# Last Seen: The State of Location Advertising and Intelligence in 2026

*Location data is sold as ground truth. In the public record, visit counts are projected or carry no stated basis, visit lift mostly rests on matched comparisons, no visit metric holds accreditation, and three companies under FTC location orders still document precise or device-level offers.*

Evgeny Popov · No Fluff Advisory · Research paper · October 2026 · Research cutoff 24 October 2026

## Abstract


*What location advertising and location intelligence actually sell in 2026: where the data comes from, how a visit becomes a number, whether anyone checks it, whether lift is causal, what the law now forbids, and whether agents can buy by place. Research current to 24 October 2026, the date the status-sensitive facts were read, with the United States as the main market.*

> **In short.**
> Location data is sold as ground truth, and the public record does not back the description. Visit numbers are projections from panels whose size is not disclosed on a comparable basis, or carry no stated basis at all. Most visit-lift products compare exposed devices with a matched or modelled group, not a holdout. No independent comparison of any vendor's projected retail visits with door counts or transactions was found, and no visit metric holds MRC accreditation; both that MRC once accredited have lost it. The FTC's five location orders mostly bar sensitive locations, not precise location as such, and three of the five ordered companies still document precise or device-level offers. Five states now ban the sale of precise geolocation outright. Agent protocols can describe a place, but in the sources screened no location signal has a public interoperability test. On the Ground Quotient, independent assurance is the weakest dimension in every cohort.

> **What a buyer can do with this.**
> Treat a visit count as a projection and a visit lift as a correlation until the vendor shows otherwise: ask which kind of number it is, what the control was, whether anyone has compared the counts with the ground, where the observations come from, and whether the vendor is under an order. [Chapter 12](#next) turns this into eight questions. This paper audits what vendors document in public by the cutoff; it does not test whose numbers are accurate.

### The question

Location is the oldest offline signal in digital advertising and the one with the most enforcement against it. The same panels feed store-visit attribution, location audiences, geofenced local buying, the US out-of-home audience currency and the foot-traffic dashboards investors read as fact. This paper asks what those products measure, where their data comes from, whether anyone has checked the numbers, whether their lift is causal, what the FTC orders and state laws changed, and whether agent-driven buying can carry location. The author's own writing calls footfall attribution "correlated and modeled by default" and DOOH outcome studies "a model on a model"; those positions were registered as rival hypotheses, not evidence.

### How it was done

Seven hypotheses and four coded questions were registered with their decision rules on 24 October 2026, before any research began, alongside a product register and scoring rules with fixed anchors. AI research agents then ran 11 workstreams, a vendor-universe sweep and 8 company profiles. A separate AI verifier, told to refute, checked each file. In all, 793 source records were registered, almost all of them opened in full, and the ledger holds 502 claims, each traced to a source; 493 research claims were checked by a verifier or a review, and the rest are author syntheses (the hypothesis readings and the privacy-cap effect) plus one legal finding added in review, New Jersey's 2026 sale ban, which a separate check read in the enacted statute. Another AI model reviewed the evidence and the design twice before scoring, then the full paper, then the live edition; every factual point it raised was checked at its primary source or against the paper's own data before anything changed, and several of the reviewer's own assertions turned out to be wrong. The corrections ledger holds 221 entries. 47 products are scored on 11 dimensions in five cohorts, and every counted cell was re-checked by a separate verifier. This is AI-assisted research with a published process. It has not been audited or peer reviewed, and no vendor saw it before publication. The author works at Samba, co-leads AgenticAdvertising.org's Signals & Measurement working group, and earlier worked at Verve Group, whose lineage includes the location-advertising company Verve Wireless; the full statement is in [chapter 13](#method).

### What it found

1. **Visit numbers are projections or unlabelled.** In cohort A, four products document a projected visit number (Foursquare, InMarket, GroundTruth and Azira), some alongside an observed count, and nine visit numbers (including Arrivalist's arrival counts) state no basis at all [16, 17, 18, 5, 449]. [Chapter 1](#defining).
2. **Supply is undocumented.** In the sources screened, no vendor discloses a US panel series on one definition across the 2019 to 2021 permission changes, and no filing attributes a quantified supply change to them, so H1 is unresolved [275, 63, 289, 427, 428, 505, 504, 88]. Ownership concentrated: T-Mobile bought Vistar Media and Blis in 2025 [36, 228], ZeroToOne.AI announced it had acquired GroundTruth in March 2026 [47, 48, 49] and Infillion announced it would acquire Foursquare in September 2026 [52, 347, 602, 603, 604]. [Chapter 2](#supply).
3. **Platform changes are dated; their effects are not.** Apple's and Google's permission changes are dated from their own documentation, and in the sources screened no filing, platform statement or independent measurement quantifies what that did to US SDK supply [505, 504, 88]. [Chapter 3](#platforms).
4. **Nobody has checked the counts against the ground at scale.** No independent multi-site comparison of projected retail visits with door counts or transactions was found; the independent checks cover parks, trails, census counts and polling places, mostly before 2021, and are adverse for SafeGraph's 2018 panel. H2 is unresolved [98, 99, 102]. [Chapter 4](#validation).
5. **Lift is mostly correlation.** Six of the eight cohort A products that document a lift method use a matched or synthetic control, and a ninth documents exposed-only visits with no lift claim; H3 is contradicted [5, 19, 119, 17, 20, 21, 22, 25, 6, 446]. The comparator decides the number: one vendor's case shows 29% lift against its holdout and 310% against unexposed devices [12]. [Chapter 5](#measurement).
6. **What audiences are built from is not disclosed.** Too few audience products name their primary source to say whether bidstream or SDK data dominates; H4 is unresolved [265, 420, 421, 422, 458, 457]. T-Mobile's app-based audience segments have been on Vistar's DOOH buying platform since May 2025, and on 5 October 2026 Vistar announced enhanced segments with location insights, available only there [334]. [Chapter 6](#audiences).
7. **The orders did not end the trade.** Only InMarket's FTC order bans selling precise location outright [130, 131, 132]; three of the five ordered companies still document precise or device-level offers, so H5 is contradicted [128, 137, 142, 145, 134, 130]. Gravy Analytics, a Unacast company, confirmed a breach of its cloud storage in January 2025 [605]. [Chapter 7](#regulation).
8. **Nobody accredits the number.** MRC's only location standard dates from 2017 and says nothing about lift [445, 600, 601]; on 24 October 2026 no visit or location metric was on any of its lists, so H6 is contradicted [171, 172, 173, 174, 175, 176, 177, 178, 179, 180]. [Chapter 8](#standards).
9. **Out-of-home runs on mobility data.** Half the US DOOH products coded name no mobility source for their audience numbers [182, 23, 192, 195, 608, 194], and the US audience currency is being rebuilt, with a pilot in the second half of 2026 [606, 607]. [Chapter 9](#dooh).
10. **Agents can describe a place; production buying by place is not established.** In the protocols and registry screened on 24 October 2026, the furthest any location signal reaches in agent-driven buying is one implemented connector; H7 is unresolved [198, 200, 201, 205]. [Chapter 10](#agents).
11. **On the Ground Quotient, independent assurance is the weakest dimension everywhere.** The field GQ is 41 in visit measurement, 44 in audiences, 41 in foot-traffic analytics and 38 in raw movement data. 1 product reaches the Most documented band, by less than a quarter of a point. [Chapter 11](#gq).

### What it means

On this paper's reading, location data is a useful signal sold with more certainty than the public record supports. A buyer who wants to treat a visit count as a count, or a lift as causal, should ask for the kind of number, the control, a ground-truth comparison, the sources and the vendor's regulatory record. [Chapter 12](#next) sets out those questions.



## 1. What a location number is


*Every location number is one of four kinds: observed, projected, declared or inferred. Most visit numbers in the documentation opened for this paper are projections or carry no stated basis. "Precise" has no single boundary. The US statutes read draw it at 1,750 or 1,850 feet, industry rules at two decimal places of latitude and longitude, and the bid-request standard draws none. A visit counted after exposure is visitation, and lift needs a control. Most of the controls vendors document are matched or modelled.*

> **In short.**
> Under this paper's coding rule, the visit totals reported by Foursquare Attribution, InMarket's LCI, GroundTruth Projected Visits and Azira are projected. For nine of the visit numbers coded across the fourteen cohort A seed products, Arrivalist's arrival counts among them, the basis is not disclosed [16, 17, 18, 5, 449] [5, 19, 119, 17, 20, 21, 22, 25, 6]. Placer.ai's population visit estimates are extrapolations from a device panel; it also exposes raw panel counts [3, 283, 386]. The US statutes read for this paper define precise geolocation by a radius of 1,750 feet, or 1,850 feet in California [152, 154, 151, 155, 162, 161, 143]. The Network Advertising Initiative (NAI) treats two or fewer decimal places as imprecise, Google caps bid-request coordinates at 0.01 degrees, and OpenRTB 2.6 has no rule on truncating coordinates [461] [444] [219, 460]. Apple quantises approximate location to a region with a typical radius of about five kilometres. Google documents Android's approximate location as accurate to within about 3 square kilometres [68, 69, 70] [86, 87, 431]. Nine cohort A products document a general visit-measurement method and eight of them claim lift. Six of the eight use a matched or synthetic control and two a randomised or holdout control; the ninth reports exposed-only visits with no lift claim [5, 19, 119, 17, 20, 21, 22, 25, 6]. In one undated PlaceIQ case, the same campaign showed 29% lift against a holdout and 310% against the open population [12].

### The definitions

- Location advertising: buying, targeting or measuring media on where a device or person was. Location intelligence: analytics about places and movement sold outside media, to retail, real estate, finance and the public sector. The same vendors and panels often serve both. The buying tasks differ, so the cohorts differ.
- Cohorts, by the output the buyer purchases. A is visit measurement and attribution. B is location audiences and targeting. C1 is foot-traffic and visit analytics, and C2 is raw movement data. D is geofencing and visit-detection SDKs and APIs, scored but not banded. Place-only datasets and maps or places APIs are listed, not scored.
- Observed visit: a panel device detected inside a defined place under stated dwell and accuracy rules.
- Projected visit (also "modelled", "estimated", "extrapolated", "normalised"): an observed count scaled to a population by a model. A projected number is never called observed here, whatever the vendor calls it.
- Declared visit: a visit taken from transactions, loyalty records, Wi-Fi log-ins or door counters.
- Inferred location: a location taken from a bid request, an IP address or a user-declared field.
- Panel: the devices a vendor observes through SDK or first-party supply in a stated period and geography, reported as US monthly devices with location observations unless the vendor defines it otherwise.
- Precise location: a fix that can place a device within a small radius, recorded with the vendor's threshold and the statutory one. Coarse location: ZIP, city, designated market area (DMA) or an operating-system "approximate" fix. Truncated coordinates are coarse.
- Place: a point of interest drawn as a polygon (a building or parcel footprint) or a radius geofence, recorded with its dwell rule and accuracy filter. A visit to a chain aggregates places.
- Visitation: visits counted or projected after exposure. Incremental lift: the difference against a control that was not exposed, coded randomised or holdout, matched or synthetic, or none.
- Every number carries two fields: its measurement basis (observed, projected, declared, inferred) and its use stage (planning, in-flight, post-campaign, or analytics outside media).

![Four of the 13 cohort A products that sell a visit number say which kind it is, and all four sell a projected number.](figures/fig-number-kinds.svg)

*Figure fig-number-kinds. **Four of the 13 cohort A products that sell a visit number say which kind it is, and all four sell a projected number.** The brief’s four kinds of location number, and the counts when the numbers opened were coded under question Q1: a sample of 16 numbers vendors sell or publish (stream C) and the 20 numbers sold by the cohort A products (stream E, after verification). In cohort A, the basis is not stated for 9 of 20 numbers; the four products that state a visit basis (Foursquare, InMarket, GroundTruth, Azira) each sell a projected visit count, three of them next to an observed one. Units: count of numbers coded. Scope: numbers opened in vendor documentation, US, at the cutoff. Sources [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [5], [21], [22], [23], [24], [25], [26], [27], [28], [11]. Caveat: Codes describe what the documentation states, not what a delivered report contains. A number coded "not disclosed" may be projected; the vendor does not say. Data: data/figure-data.json#number_kinds*

A company can have rows in several cohorts, and evidence is not inherited between them. Foursquare's 2026 product menu lists Attribution, Audience and Proximity for marketers, a Geospatial Intelligence Platform, and the Places API and Movement SDK for developers [347]. Placer.ai sells an Analytics Platform, Enrichment and a Data Marketplace, and its advertising page offers to measure lift without naming a separate measurement product [293, 294]. The product register holds 86 rows, 29 of them listed and 10 historical ([chapter 13](#method)).

### Four kinds of number

Most of the cohort A record is projection or silence, under the Q1 coding rule registered on 24 October 2026. A number is coded observed only when the source documents detection of panel devices for the places and period reported. It is coded projected when any scaling, extrapolation, normalisation or estimation to a population is documented, or when the vendor's own term implies one. Transactions, loyalty records, Wi-Fi and counters make a number declared, and bid requests, IP addresses and user-declared fields make it inferred. A number whose basis is not stated is coded not disclosed, and that gap is scored under method disclosure.

Foursquare Attribution, InMarket LCI's projected store visits, GroundTruth Projected Visits and Azira's extrapolated total visits are documented as projected. GroundTruth Observed Visits, Azira's observed visits and InMarket's exposed device visits are documented as observed [16, 17, 18, 5, 449]. Arrivalist MediaInsights' spend lift is declared, from card transactions, but the basis of its arrival counts is not stated [16, 17, 18, 5, 449]. For Blis, Reveal, Adsquare, Cuebiq, Vistar, Kochava, Lifesight and Precisely the basis is not disclosed [16, 17, 18, 5, 449]. Unacast's pages name no attribution product in the sources screened [125].

The projections are documented to different depths. Foursquare's 2022 method post describes census weighting of the panel to the national population and projection of visit volumes on match rates by platform and partner [16]. InMarket's console defines Overall Projected Store Visits as exposed device visits multiplied by an "ND Device / DMA Size Ratio", in undated text from the NinthDecimal era [17]. Azira counts each observed device as 1/P(Detection) devices, extrapolates only above about 1,000 observed visits and caps the multiplier at 20x [5, 447]. GroundTruth applies the visitation of a per-campaign panel of impressed devices to all impressed devices. The projection appears once a campaign reaches about 40 Observed Visits and 350,000 impressions, and GroundTruth reports no uncertainty for it [18].

Foot-traffic products show the scaling outright. Placer.ai's documentation scales 74.8 thousand panel device visits to one Michigan grocery store to an estimated 2.4 million [3, 283, 386]. That implies a scale factor of about 32 [289, 3]. Its API keeps raw panelVisits apart from extrapolated visits [283, 3]. Arrivalist's 2021 Lee County analysis scaled 492,897 panel trips to 9,861,477 projected US adult trips, a factor of 20.0 [4]. Unacast defines its machine-learning visits as an estimate of "unique people with a stay at a location in a day", produced by a model with more than 150 features [325, 106]. This paper therefore codes the basis of any dashboard "visits" figure before quoting it.

Observed counts have their own bias. GroundTruth calls Observed Visits a "raw" observed count [6, 367]. Its help centre says observed visits "may be lower than actual visits" because not all users are trackable [6, 18].

Declared numbers mostly enter as sales. Foursquare documents over 2 billion annual card transactions mapped to places across 450+ US chains [236]. InMarket's Retail Sales Lift add-on uses card transactions from a subset of the same exposed and control users [423, 11, 23], and Arrivalist names Visa, Mastercard and Discover as its card sources [25, 449, 450, 451]. Exposure can be inferred as well as location. Vistar's DOOH cases infer exposure from ad-log timestamps, device location, venue type and screen location [8, 9, 463]. Adsquare infers it from the proximity of mobile advertising IDs to screen coordinates while an ad is live, with no radius stated [464]. GroundTruth's 2021 help centre says its CTV-to-store visits rely on TapAd converting IP addresses to mobile ad IDs [452, 453]. Foursquare and Cuebiq document linear-TV-to-visit measurement without naming the viewing-data supplier [22, 119, 274]. None of the cohort A documentation opened names Samba, the TV data company where the author works, as a source of TV exposure data [22, 119, 274].

### Where the observations come from

MRC's 2017 Location-Based Advertising Measurement Guidelines list GPS, Wi-Fi, IP, cell trilateration, registration, beacon, panel and bid-request location as sources. They rate beacons within metres, GPS at 10-100 m, Wi-Fi at 30-100 m and cell at ZIP or neighbourhood level, and they call mobile IP generally neither accurate nor precise [445]. Those ratings date from March 2017 [445, 600, 601], before Apple's iOS 13 and Google's Android 10 permission changes of September 2019 [65, 66, 67] [82, 83].

SDK and panel supply is disclosed in units that do not compare, and mostly without dates [117, 17, 3, 15, 118]. Placer's help centre describes tens of millions of US devices from hundreds of partner apps, about 7% of the US population. Its September 2024 deck gives 30 million active devices from 500 SDK apps on one slide and 25 million-plus on another, with neither figure defined by period [3, 102, 289]. InMarket names its own apps, a partner SDK app network and unnamed third-party sources, with no proportions [267, 134]. Foursquare names GPS pings from its own consumer apps and vetted third parties [236], and it retired its City Guide app on 15 December 2024 [64, 51, 52, 236]. In the sources screened, no US location vendor discloses a panel series on one definition that spans a pre-2021 and a post-2023 figure [275, 63, 289, 427, 428]. [Chapter 2](#supply) tests H1 on that record.

Bidstream location arrives in the OpenRTB 2.6 Geo object. It carries latitude and longitude, a type (1 GPS or location services, 2 IP address, 3 user-provided), an accuracy in metres, the seconds since the fix and the IP location service used. The specification says coordinates should be passed only when they match the accuracy of the stated type, and that a postal-code centroid should not be passed [219, 460]. Two 2015 vendor studies measured how often the field was usable. Thinknear's vendor-run Q1 2015 index found location on 62% of ad requests and put 37% of located impressions within 100 m [441, 442, 443]. Placed's 2015 analysis found that less than 1% of the business visits its panel observed generated an ad-request location within 100 m [440]. This paper does not read either as a description of 2026 supply. Azira documents a blend of SDK integrations and location from the bid requests of over 100,000 apps, with no proportions [5]. InMarket's GeoLink scores the location carried in ad requests before bidding [276, 40]. The FTC's January 2025 Mobilewalla order is the only regulator record found in the audiences stream that documents bidstream as a cohort B source [140]. Google's exchange sends only IP-derived location, capped at 0.01 degrees [444]. Under Google's Authorized Buyers guidelines, any precise location a buyer receives by other means may be used only to bid on that impression [433].

The one IP-geolocation accreditation found in the MRC communications screened has lapsed. MRC moved Digital Envoy's NetAcuity service from hiatus to non-accredited in January 2020, and no IP-geolocation service was on MRC's lists on 24 October 2026 [165, 171]. Platform changes touch IP location only at the edges. iCloud Private Relay maps to a roughly city-level area by default but covers only Safari, DNS and insecure HTTP traffic, leaving HTTPS in-app traffic untouched [71, 72, 429, 430]. Chrome's IP Protection was confined to Incognito third-party contexts and is being withdrawn [434, 435, 436, 91]. Some vendors refuse IP outright. GroundTruth says it rejects centroids and carrier-IP locations [357, 259], and Simpli.fi says it does not use IP because of truncation [456, 457].

Telco and first-party sources are named more often than they are described. T-Mobile's app-based segments have been on Vistar's DSP since May 2025, and on 5 October 2026 Vistar announced enhanced, location-enriched segments available only through it [334]. Per AdExchanger, it combines mobile location data with app installation and usage that T-Mobile infers from network requests [338, 335]. Blis's 2024 strategic report cites partnerships with telco providers across North America, the UK and Europe [226]. Among the suppliers screened, only Quadrant documents a first-party server-to-server route with explicit counts, which are global. Outlogic says "No Bidstream Data" but gives no count [425, 426, 128]. "Opt-in" and "consented" are claims about the consent route. Adsquare's signals are "sourced from vetted partners with explicit, auditable consent", with no route named [420, 455]. No cohort B product documents a primary source, so H4 is unresolved on the documentation read [265, 420, 421, 422, 458, 457]. The dated clarification of its rule does not change that reading ([chapter 6](#audiences)).

### Precise and coarse

The 2017 MRC guidelines separate accuracy, meaning closeness to the true location, from precision, meaning the number of decimal places, and require vendors to disclose both [445]. They equate two decimal places with about one kilometre, four with about ten metres and five with about one metre, and warn that decimal count must not be equated with accuracy [445]. The table sets out the boundaries in the rules read for this paper, as current when they were read in October 2026.

| Rule | Where precise ends | Evidence |
|---|---|---|
| Maryland, Oregon, Washington (precise location information), Connecticut, Virginia | Radius of 1,750 feet | [152, 154, 151, 155, 162, 161, 143] [152] [154, 153] [151] [155, 159] |
| California, Civil Code 1798.140(w) | Radius of 1,850 feet | [161, 163] |
| The five FTC location orders | Coarse data excluded, for example ZIP code or census block with a radius of at least 1,850 feet | [152, 154, 151, 155, 162, 161, 143] [126] |
| NAI guidance (February 2020) | Two or fewer decimals, an area at least a 500 m-radius circle, or a non-unique descriptor is imprecise | [461] |
| Google Authorized Buyers | IP-derived only, 0.01 degrees | [444] |
| Prebid | Two decimals, only when a rule denies transmitPreciseGeo; the default is allow | [437, 438, 439] |
| OpenRTB 2.6 | No truncation or rounding rule | [219, 460] |
| Apple iOS 14 approximate location | Region with a typical radius of about 5 km | [68, 69, 70] |
| Android 12 approximate location; Google Play Data safety | About 3 square kilometres; precise is less than 3 square kilometres | [86, 87, 431] [432, 89, 90] |

Two decimal places is the recurring line, but the bodies attach different distances to it. NAI's 500 m radius is about 6% smaller than the 1,750-foot (533 m) statutory radius [461, 445, 462, 444, 437]. Washington's geofencing ban uses a third distance: a boundary of 2,000 feet or less around in-person health care providers [151]. The operating-system definitions differ in kind as well as size, because Apple states a radius and Google an area [68, 69, 70] [86, 87, 431]. Apple says apps cannot opt out of a user's approximate choice, and region monitoring needs full accuracy [68, 69, 70]. Android 12 lets users set approximate accuracy for any app, whatever it requests [86, 87, 431]. The paper codes operating-system approximate fixes and truncated coordinates as coarse. Because OpenRTB sets no truncation rule, the precision of a bid request's coordinates is the sender's choice, and their decimal count does not show their accuracy [219, 460] [445].

### Places, polygons, radii and dwell

The 2017 guidelines define a geofence as either a point-radius or a boundary (polygon). They define a physical visit as the intersection of device location and place, using speed or longitudinal analysis to exclude drive-bys, and note that geofences generally cannot represent multi-storey buildings [445]. Vendors fill in the rest themselves, and the rules differ.

| Product | Place | Visit or stop rule | Evidence |
|---|---|---|---|
| Placer.ai | Device signals matched to a geofenced POI | From 1 minute of dwell since 31 March 2024; at most once per person per POI per day | [284, 283, 285, 287, 382] |
| Foursquare Visits feed | Venue, with place shapes in dense locations | A stop of at least two minutes; the Movement SDK FAQ says there is "no standard minimum dwell time" | [241, 249, 354] |
| Unacast machine-learning products | Polygon | Multiple pings within 50 m for longer than 60 seconds | [106] |
| Radar SDK | Circles of 10 to 10,000 m, polygons or isochrones | Stopped after 140 seconds within 70 m (Responsive and Continuous presets on iOS, all presets on Android); dwell thresholds configurable | [298] [300] |
| GroundTruth | Blueprints polygons, 4.8 million across the US and Canada; 12-metre radial fence where none exists | Named tests, no numeric threshold | [252, 253, 258, 358, 361] [357, 259] |
| Adsquare | 2D POI polygons, "no fuzzy radius logic" | Not stated | [448, 21] |

GroundTruth asserts, without citing a study, that radial geofences measure visits "with up to 75% inaccuracy" [254, 361]. No cohort B documentation opened states a fix-accuracy threshold with a method or a detection error statistic. The only numeric precision claim found is Azira's unexplained "within 3 meters of precision" [263, 122, 420, 457, 459]. Disclosed attribution windows run from 4 days (Vistar with Foursquare, for DOOH) to a 30-day look-back (Simpli.fi), and window choice alone moves a visit count materially [8, 5, 6, 17, 10, 118]. AdCP, a protocol for agent-driven buying, uses the same vocabulary. AdCP v3.2.6 (23 October 2026) represents point-based proximity as a radius, a travel-time isochrone or a buyer-supplied GeoJSON polygon, and identifier-based places, as schema only [197, 221]. The author co-leads AgenticAdvertising.org's Signals & Measurement working group, where his cell-grid RFC, AdCP issue #5969, was open when read [205, 206, 207]. [Chapter 10](#agents) treats that as a declared interest.

### Visitation is not lift

A visit rate among exposed devices says nothing about what the ad caused until it is compared with a control. MRC's 2017 location guidelines contain no provision on visit lift, incrementality or control groups [445, 454]. GroundTruth's Observed Visits is the plainest case of visitation. It is an exposed-only count with no control group, crediting a visit within 14 days of exposure by last touch [6, 18]. GroundTruth's Cost Per Visit model bills on those verified visits [6, 367].

The comparator decides the size of the number. PlaceIQ's undated A/B case for a national big-box retailer reports a Place Visit Rate of 2.75% for exposed devices against 2.13% for a standing 10% holdout, a 29% lift. Against the open population the rate is 0.67%, a 310% "open lift". The same campaign shows a 10.7-fold larger lift when the comparator changes [12].

Of fourteen cohort A seed products, nine document a general visit-measurement method and eight of them claim lift. Six of the eight use a matched or synthetic control (Foursquare, InMarket, Reveal, Adsquare, Cuebiq, Arrivalist), two a randomised or holdout control (Azira, Blis), and GroundTruth reports exposed-only visits with no lift claim [5, 19, 119, 17, 20, 21, 22, 25, 6]. Blis's Smart Holdout Groups balance target and control ZIP codes by machine learning, and Blis does not describe the assignment as randomised [19, 424]. Vendor names for these designs overstate them. Cuebiq describes "causal machine learning to simulate a randomized control trial" with behaviourally matched controls [446]. InMarket calls its synthetic control causal [121]. Foursquare's product page says "incremental lift" for an output its blog and metrics page call "behavioral lift" [119, 236]. Under this paper's rules a matched or synthetic comparison is correlation, whatever the vendor calls it, and the vendor's term is reported beside the code.

On those counts H3 reads contradicted: seven of the nine documented methods are not holdouts, and excluding GroundTruth, six of eight are matched or synthetic [5, 19, 119, 17, 20, 21, 22, 25, 6]. The dated clarification of 24 October 2026 fixed the decision tree and did not by itself move the verdict. The author's glossary line that footfall attribution is "correlated and modeled by default" is H3's registered rival hypothesis, not evidence, and Azira and Blis are exceptions to it [5, 19, 119, 17, 20, 21, 22, 25, 6]. [Chapter 5](#measurement) codes each product, [chapter 4](#validation) asks whether the projected visits agree with ground truth, and [chapter 12](#next) turns these definitions into questions a buyer can ask.



## 2. Who supplies the data


*Several US location vendors changed hands in 2025 and 2026, two of them to a carrier. T-Mobile bought Vistar Media and Blis in early 2025, ZeroToOne.AI bought GroundTruth in March 2026, and Infillion announced in September 2026 that it was buying Foursquare. Near Intelligence went through Chapter 11. Whether the supply underneath shrank after Apple's and Google's permission changes cannot be read from the record: no vendor in the sources screened discloses a US panel series on one definition from before 2021 to after 2023, so H1 is unresolved.*

> **In short.**
> T-Mobile US completed its purchases of Vistar Media on 3 February 2025 and Blis on 3 March 2025, for US$621 million and US$180 million in cash [36, 228]. ZeroToOne.AI announced on 25 March 2026 that it had acquired GroundTruth [47, 489, 49]. Infillion, which already owns Gimbal, announced on 18 September 2026 that it would acquire Foursquare; by 24 October 2026 participants described the deal as done, but no closing record or date was found [52, 347, 602, 603, 604]. Gravy Analytics trades as Unacast, Precisely owns PlaceIQ and Rezolve AI owns Bluedot [137, 333, 135, 44, 490] [33, 34] [474, 475]. Near Intelligence filed for Chapter 11 in December 2023 and sold substantially all its assets to a Blue Torch vehicle; that Azira is that business is this review's inference [54] [55, 467, 466] [458, 468, 427]. Vendors' panel and scale figures use incompatible units, and none forms a US series on a constant definition from before 2021 to after 2023, so H1 is unresolved on the sources screened [275, 63, 289, 427, 428]. Some vendors document sources other than their own SDKs, but no figure shows a market-wide contraction or substitution [329, 331, 137] [5] [275, 63, 289, 427, 428].

### The ownership map at the cutoff

The table records who owned each business on 24 October 2026, when the pages and registries were read, and the event that put it there. Each event carries one class under shared rule 11, with dates from the source that settles it; an event whose completion no opened source states is recorded as announced (question Q3) [50, 51] [44].

| Business | Owner or status at the cutoff | Event class | Dates | Evidence |
|---|---|---|---|---|
| Vistar Media | T-Mobile US | Acquisition, completed | Agreed 20 Dec 2024; closed 3 Feb 2025; US$621m cash | [36] [36] |
| Blis | T-Mobile US ("part of T-Ads" on Blis's homepage) | Acquisition, completed | Agreed 18 Feb 2025; closed 3 Mar 2025; US$180m cash | [36] [225, 486] [487, 228] |
| Foursquare | Infillion, by participants' statements | Acquisition, announced; closing unverified | Announced 18 Sep 2026; no closing date found | [52, 347, 602, 603, 604] [52, 51, 347, 488] |
| Gimbal | Infillion | Not dated in this review | n/a | [51, 50] [473, 52] |
| GroundTruth (formerly xAd) | ZeroToOne.AI | Acquisition, announced as completed | Announced 25 Mar 2026; counsel confirmed 24 Apr 2026 | [47, 48, 49] [47, 489, 49] [56] |
| Gravy Analytics and Venntel | Trade as Unacast | Merger, announced; structure and completion not stated | Announced 29 Nov 2023 | [137, 333, 135, 44, 490] |
| Unacast's Aggregated Location Insights business | Pine59 | Management buyout, completed per the buyer | Announced 1 Sep 2026 | [326, 413, 325] |
| PlaceIQ | Precisely; product sold as "Precisely Audiences" | Acquisition, completed | Agreed 19 Jan 2022; completed Feb 2022 | [33, 34] [482, 483, 484, 421] |
| Bluedot | Rezolve AI | Acquisition, completed (related party) | Agreed 4 Feb 2025; closed 20 Feb 2025 | [474, 475] |
| Near Intelligence | Liquidated; assets to BTC Near Holdco LLC; Azira as successor is inferred | Insolvency | Chapter 11 8 Dec 2023; deregistration filed 28 Mar 2024 | [54] [55, 467, 466] [458, 468, 427] |
| Outlogic (formerly X-Mode) | Digital Envoy | Acquisition, then rebrand | Reported 5 Aug 2021 | [42] [469] |
| Hivestack | Perion | Acquisition, completed | Closed 12 Dec 2023; US$100m cash at closing | [35] |
| Place Exchange | Broadsign | Acquisition, announced as completed | Announced 25 Nov 2025 | [45] |

### Carriers and holding companies

T-Mobile owns two of the businesses in the map, and its FY2025 10-K puts the fair value of consideration at US$617 million for Vistar and US$174 million for Blis, after excluding payments that settled preexisting relationships [36, 228]. The filing calls Blis "a provider of advertising solutions"; the location characterisation comes from Blis's own pages, and T-Mobile states that Blis's results to 31 December 2025 were not material [36]. Lamar Advertising received US$115.1 million in cash at the Vistar closing for its 20% stake, with up to a further US$15.1 million in escrow, per its 8-K exhibit of 20 February 2025 [37]. Blis's About page, modified 8 October 2026, still calls it an "Independent DSP" and does not mention T-Mobile [487, 228].

The carrier brings its own data, described mostly as app data. Since 10 September 2025 the Blis DSP has carried T-Mobile's US app ownership and engagement data, which Blis says it combines with movement patterns and transactions [227, 228]. On 5 October 2026 Vistar announced enhanced T-Mobile segments for DOOH, available only through Vistar, whose signal list now adds "location insights" [334] [492, 334]. AdExchanger reported the same day that T-Mobile's data for Vistar includes mobile location data, and that for users who opted into location tracking T-Mobile can detect proximity to OOH ads; T-Mobile tied the exclusivity to tightening regulation of location data and mobile device IDs [335].

Infillion would hold two location product families if the Foursquare deal has closed: the Gimbal Location Platform, an SDK sold to app owners for their own users' data, and Foursquare's attribution, audience, places and SDK products [473, 52]. Infillion also bought Catalina (closed February 2026) and MediaMath (out of bankruptcy in 2023, for US$22 million) [51, 50]. Infillion's release says leadership will join "after the transaction closes", while Foursquare's homepage read "Foursquare has joined Infillion" on 24 October 2026 [52, 51, 347, 488]. Business Insider reported Dennis Crowley's account that equity holders received $0 [52, 51, 347, 488]. This paper records the event as announced, with completion resting on participants' statements [52, 347, 602, 603, 604]. ZeroToOne.AI disclosed no seller, price or closing record for GroundTruth, and its release names debt-financing advisers [47, 489, 49].

The Unacast and Gravy combination is harder to read. Gravy Analytics, Inc., whose subsidiary is Venntel, opens its September 2026 privacy policy as "Gravy Analytics, Inc. dba 'Unacast'", and gravyanalytics.com redirects to unacast.com [137, 333, 135, 44, 490]. On 1 September 2026 Pine59 said it was "the new home of Unacast's former Aggregated Location Insights business" after a management buyout led by two Unacast co-founders and its former CTO [326, 413, 325]. Unacast's own pricing page still listed the machine-learning foot-traffic datasets and the Location Insights platform on 24 October 2026, and no Unacast statement of the sale was found in the sources opened [326, 413, 325] [331, 332].

### Near Intelligence and Azira

Near Intelligence's collapse is coded as an insolvency [54]. It listed on Nasdaq through a SPAC combination completed on 23 March 2023 [465, 466]. Its 10-Q for the first half of 2023 reported revenue of US$33.2 million, a net loss of US$36.9 million and substantial doubt about its ability to continue as a going concern [465]. It filed Chapter 11 petitions in Delaware on 8 December 2023, with a stalking-horse credit bid of not less than US$50 million from BTC Near Holdco LLC [54]. The court approved the sale of substantially all its assets to that buyer on 22 February 2024, the sale closed on 1 March, and a Form 15-12G deregistration was filed on 28 March 2024 [55, 467, 466].

Azira operates as Azira LLC in Pasadena and claims 13 trillion location signals and 43 million points of interest. No filing, registry record or Azira statement opened names Near, so the lineage from Near's assets to Azira is this review's inference [458, 468, 427]. The Chapter 11 plan was a plan of liquidation, so on that inference Azira is the buyer's business, not a reorganised Near [55, 467, 466] [458, 468, 427].

### Consolidation since 2019, by class

![Five companies were sold or agreed to be sold in 2025 and 2026, among them Vistar, Blis, GroundTruth and Foursquare.](figures/fig-consolidation.svg)

*Figure fig-consolidation. **Five companies were sold or agreed to be sold in 2025 and 2026, among them Vistar, Blis, GroundTruth and Foursquare.** Acquisitions, mergers, the Near insolvency, rebrands, splits, SDK removals and product retirements in the location universe since 2017, from filings, releases and trade reports (stream D, corrected by its verifier and the review rounds). T-Mobile completed its purchases of Vistar Media on 3 February 2025 and of Blis on 3 March 2025, transferring US$621 million and US$180 million in cash. Units: dates. Scope: companies and products in the vendor universe, 2017 to the cutoff. Sources [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61], [62], [63], [64]. Caveat: Events are classified, not counted as market exits; no event found documents a failure of a location metric as its cause. Funding rounds, marketplace listings and regulatory orders are in other figures and tables. Data: data/figure-data.json#events*

Q3 is a coding question, and its rule reads no event as evidence that location data or a method failed unless a source establishes the cause. Foursquare bought Placed from Snap in June 2019, for US$66.9 million of purchase consideration in Snap's filing [30, 29]. Telenav sold Thinknear to inMarket on 16 August 2019 for a 14.5% member interest valued at US$15.6 million [31, 32]. InMarket bought NinthDecimal's assets in a deal reported on 10 September 2020, not in 2019 as the seed list had it [41, 40]. Advan Research bought SafeGraph's Patterns business in November 2022, a product retirement for SafeGraph [43, 63]. Qualcomm acquired Skyhook, an early Wi-Fi positioning supplier, in May 2022 [476]. Media and Games Invest agreed in January 2020, through its subsidiary gamigo, to buy substantially all the assets of Verve Wireless, and renamed itself Verve Group SE on 13 June 2024, a rebrand [495, 496].

Neither merger in the record has a completion date in the sources opened. Foursquare and Factual announced theirs on 6 April 2020 [39, 38]. Unacast and Gravy announced theirs on 29 November 2023 with no structure stated [44]. Cuebiq's clean room was split out as Spectus in February 2022, a restructuring [57, 58]. On 24 October 2026 spectus.ai redirected to cuebiq.com; whether Spectus has been folded back is an inference from the redirect, not an established event [479, 480].

Several events touch supply directly. Apple and Google barred apps from using X-Mode's SDK in December 2020, which this review found only in trade reporting [60]. Cuebiq announced on 23 February 2021 that it was sunsetting its proprietary SDK in favour of a clean-room product, keeping an open-source version [60]. Foursquare retired its City Guide app on 15 December 2024 and the web version on 28 April 2025 [64]. Its consumer apps supply the check-ins it uses as visit ground truth and its first-party panel [64, 51, 52, 236], and the retirement page does not say whether City Guide contributed to that panel [64]. Lifesight's 2026 homepage names no footfall or location product, coded as a pivot by inference [481]. Arity announced on 16 October 2026 that it would acquire INRIX's assets, expected to close in 2026 [493, 494].

### Panels in incompatible units

Every scale figure found carries its own definition. The table keeps each with its date and unit as published [117, 17, 3, 15, 118]; none forms a US series on one definition from before 2021 to after 2023 [275, 63, 289, 427, 428].

| Vendor | Figure as published | Unit and date | Evidence |
|---|---|---|---|
| SafeGraph | 28,914,920 opted-in devices, by carrier | One release, July 2020; not a monthly series | [63] |
| InMarket | "About 100 million unique devices" sending location each year | FTC complaint, January 2024; each year from 2016 | [275] |
| InMarket | "200 million+ Location-based consumers" and other figures | Current pages, no period | [268, 273, 23, 274, 40] |
| Foursquare | 100 million monthly US devices; 500 million worldwide; 250 million US devices of "aggregated coverage" | May 2019; April 2020; September 2026 | [30, 38, 241, 347, 52] |
| Placer.ai | "Over 20 million active devices" to "roughly 30 million mobile devices" | May 2022 to September 2026; one monthly basis | [385, 289, 290, 291] |
| Gravy and Unacast | More than 17 billion signals from around a billion devices daily; "more than 1 billion monthly devices across 180+ countries" | FTC release, December 2024; product page, undated; global | [328, 330, 428] |
| Advan | "45M U.S. & 65M global devices" | At access, 24 October 2026; no period | [477] |
| GroundTruth | 95 million "active monthly users" | June 2017; no current US figure | [262, 56, 356, 47] |
| Blis | "370m opted-in global planning and measurement panel" | August 2021; global | [230, 343] |
| Near | 1.6 billion "unique user IDs" | March 2023; no geography | [427] |

The figures cannot be read as trends. Foursquare's 100 million US devices in 2019 and 250 million in 2026 look like growth, but the 2026 term is undefined and may include partner or bidstream data [30, 470, 38, 50]. Placer's own figures range from about 7% to about 9% of the US population, and only a city fact sheet gives a monthly basis [3] [385, 289, 290, 291]. Placer also restates its history when it changes its model, so its series read at two dates can differ for the same past period [285, 102].

### Where the supply comes from

Several vendors document sources other than an SDK of their own, without proportions [329, 331, 137] [5]. Unacast says it chose to license data "instead of collecting location data through our own SDK", from aggregators representing SDKs across tens of thousands of apps [329, 331, 137]. Azira's measurement panel blends SDK integrations with bidstream location from the bid requests of over 100,000 apps [5]. Foursquare names its own apps and vetted third-party partners, plus card transactions mapped to places [236]. Foursquare also infers TV exposure by joining viewership data with ad run-logs [236], without naming the viewing-data supplier; no opened cohort A documentation names Samba, the TV data company where the author works [22, 119, 274]. Placer documents background location from partner apps that carry its SDK, and no source opened documents bidstream use [289, 379, 380, 385].

Regulators' records describe aggregation too [135]. The FTC alleged that Gravy collected no data directly from consumers and bought from suppliers who may draw on the mobile advertising marketplace [135]. Its Mobilewalla order bans collecting data from real-time bidding exchanges for purposes other than taking part in the auctions, which documents an allegation, not an admission [491].

Foursquare adjusts its projection weights for technical changes such as an OS update and names iOS 13 among the causes of step changes in raw visits; the documentation opened gives no size or direction for them [240]. Cuebiq's CEO cited privacy and security risk for the SDK sunset; the article, not Cuebiq, linked the move to Apple's tracking changes [60]. [Chapter 3](#platforms) dates the permission changes themselves.

### Marketplaces and resale

Location data also moves through marketplaces. Foursquare documented AWS Data Exchange from 13 November 2019 and Snowflake from 8 February 2021 [470, 471, 472]. Placer delivers through its API, bulk feeds, private Snowflake sharing and the Bloomberg Terminal; this review found no public marketplace listing for Placer in the searches logged [280, 281, 282, 224]. Outlogic licenses an "Observation Panel", attribution data and audiences through Snowflake, AWS S3, a REST API, Eyeota and Adsquare [128]. Datarade's foot-traffic category, headed "40 Foot Traffic Data Providers" on 24 October 2026, adds sellers not in the seed list, with self-reported figures [485]. At least three location sellers screened, Unacast, Veraset and Complementics, also sell links between mobile advertising IDs and hashed emails or IPs [125, 478, 426, 425]. Marketplace discovery is incomplete: this review found no Databricks listing for five named vendors in the searches logged, and the universe sweep did not screen the Snowflake Marketplace or AWS Data Exchange search pages [470, 471, 472] [485].

### H1, coded by the rule

H1 states that SDK-sourced precise location supply to US vendors contracted after the 2019 to 2021 permission changes and that vendors substituted other sources. The registered rule needs either two vendors' series falling on a constant definition, or a filing or official statement attributing a supply reduction to a named platform change, and at least one vendor documenting a shift of primary source away from SDK supply. Vendor commentary without a figure does not support on its own, and absence of a comparable series is unresolved, not contradicted. Foursquare's iOS 13 note gives neither a size nor a direction [240]. In the filings, statements and vendor pages opened, no US vendor discloses a series on one definition that spans a pre-2021 and a post-2023 figure, so H1 is unresolved on this evidence [275, 63, 289, 427, 428]. Cuebiq's partial SDK sunset is a documented shift away from own-SDK supply, with no panel figure [60]. Read literally, InMarket's flat 100 million devices a year from 2016 would count against contraction for that vendor, but it is one rounded allegation, not a series [275, 63, 289, 427, 428]. The absence holds only for the sources screened. Two vendors' disclosed US series on one definition from before 2021 to after 2023, or a filing attributing a supply reduction to a named platform change, would let the rule decide it ([chapter 13](#method)).



## 3. What Apple and Google changed


*Between 2019 and 2021 Apple and Google split background location from foreground use, added a one-time grant and let users hand apps an approximate location instead of a precise one. Google Play also put background location under review and now bars the sale of device location. The platforms' own pages date each change. What the changes did to US SDK supply is another matter. In the sources screened, no filing, official statement or independent measurement puts a number on it. The one platform figure found counts permission choices, and the vendors that name a change give no size.*

> **In short.**
> Apple shipped Allow Once in iOS 13 on 19 September 2019 and approximate location in iOS 14 on 16 September 2020 [65, 66, 67] [68, 69, 70]. App Tracking Transparency (ATT) became mandatory with iOS 14.5 on 26 April 2021 [68, 76, 77, 78, 69]. Android 10 (3 September 2019) separated background location, Android 11 (from 8 September 2020) added a one-time grant and Android 12 (pushed to AOSP on 4 October 2021) added approximate location [82, 83] [84, 85] [86, 87, 431]. Google Play's policy, read on 24 October 2026, prohibits selling device location and never permits it solely for advertising or analytics [501, 94] [88, 84, 94, 500]. The later changes are narrower: privacy manifests enforced from 1 May 2024, a WWDC24 model that ties Always authorization to a session started in the foreground, Private Relay for Safari and insecure app traffic, and Chrome's IP Protection, listed for retirement on 17 October 2025 [497, 79, 80, 498] [73, 74, 499] [71, 72, 429, 430] [434, 435, 436, 91]. The December 2020 removal of X-Mode's SDK from both app stores rests on trade reporting, not on a platform statement [60]. Google's statement that over half of Android 10 users chose "While app is in use" is the only effect figure a platform attributes to a change, and it measures grants, not supply [88]. Foursquare, Near and a trade article on Cuebiq name platform changes without a figure, which the Q4 rule codes asserted [240] [504] [60]. In the sources screened, nothing attributes a quantified change in US SDK-sourced location supply to a named Apple or Google permission change [505, 504, 88].

### How the changes are dated

The registered Q4 rule dates a change only from the platform's own documentation, and keeps announcement, release and enforcement dates apart. It codes an effect documented when a filing, official statement or independent measurement attributes it to the change, asserted when vendor commentary names the change without a figure, and coincident when a disclosed series moves at the same time with no attribution. A change whose effect no source attributes carries no code, and that absence is not evidence of no effect [505, 504, 88]. The platforms' pages are the settling record for dates. They are the platforms' own statements, and this paper does not treat them as independent evidence of what a change did [65, 66, 67].

| Platform | Change | Announced | Released or effective | Enforced | Evidence |
|---|---|---|---|---|---|
| Apple | iOS 13: Allow Once, background tracking alerts, Wi-Fi and Bluetooth protections | 3 June 2019 | 19 September 2019 | n/a | [65, 66, 67] |
| Google | Android 10: ACCESS_BACKGROUND_LOCATION, allow only while in use | Not stated | 3 September 2019 | n/a | [82, 83] |
| Google | Play background-location approval | 19 February 2020 | Projected 3 August 2020 (new apps), 2 November 2020 (existing) | Existing apps deferred to 2021 on 8 September 2020 | [88, 84, 94, 500] |
| Google | Android 11: one-time grant, Allow all the time only in Settings | Not stated | Rollout from 8 September 2020 | n/a | [84, 85] |
| Apple | iOS 14: approximate location | 22 June 2020 | 16 September 2020 | n/a | [68, 69, 70] |
| Apple | App Tracking Transparency | 22 June 2020; deferred 3 September 2020 | 26 April 2021 (iOS 14.5) | 26 April 2021 | [68, 76, 77, 78, 69] |
| Apple | iCloud Private Relay (public beta, iCloud+) | Not stated | 20 September 2021 (iOS 15) | n/a | [71, 72, 429, 430] |
| Google | Android 12: approximate location | Not stated | AOSP 4 October 2021 | n/a | [86, 87, 431] |
| Google | Play Data safety section | 6 May 2021 | Visible 26 April 2022 | Mandatory after 20 July 2022 | [432, 89, 90] |
| Apple | Privacy manifests and required-reason APIs | WWDC23; "spring 2024" on 7 December 2023; dated 29 February 2024 | Warnings from 13 March 2024 | Uploads rejected from 1 May 2024 | [497, 79, 80, 498] |
| Apple | WWDC24 authorization model (CLServiceSession) | WWDC24 | iOS 18, 16 September 2024 | n/a | [73, 74, 499] |
| Google | Privacy Sandbox and IP Protection retirement | 17 October 2025 | No dates given | n/a | [91] [434, 435, 436, 91] |

![Most platform location-permission changes date from 2019 to 2021; Apple’s last was in iOS 18, and Google’s next documented change takes effect in January 2027.](figures/fig-platform-timeline.svg)

*Figure fig-platform-timeline. **Most platform location-permission changes date from 2019 to 2021; Apple’s last was in iOS 18, and Google’s next documented change takes effect in January 2027.** Apple and Google location-permission and policy changes since 2019, from the platforms’ release notes, developer pages and policy posts (stream G, corrected by its verifier). No filing, platform statement or independent measurement in the sources screened attributes a quantified change in US SDK location supply to any of these changes. Units: dates. Scope: Apple and Google platform documentation, 2019 to the cutoff. Sources [65], [66], [67], [68], [69], [70], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81], [82], [83], [84], [85], [86], [87], [88], [89], [90], [91], [92], [93], [94]. Caveat: Timing is not effect: a coincidence in time between a platform change and a vendor’s supply is recorded as a coincidence (Q4 rule). Google Play’s ban on selling device location has no settled start date and is not plotted. Data: data/figure-data.json#platform*

### Apple

Apple announced a one-time location option and background-use information on 3 June 2019 and shipped them in iOS 13 on 19 September 2019, as Allow Once, background tracking alerts and Wi-Fi and Bluetooth protections against location inference [65, 66, 67].

iOS 14, released on 16 September 2020, let users give an app approximate location instead of precise. Apple's WWDC20 session says apps cannot opt out. A reduced-accuracy fix is quantised to a region with a typical radius of about five kilometres, a couple of kilometres in dense cities and ten or more in rural areas, recomputed about four times an hour. Region monitoring requires full accuracy [68, 69, 70]. Under this paper's definitions ([Chapter 1](#defining)), an approximate fix is coarse location [68, 69, 70]. This review found no Apple statistic on how many users choose it [68, 69, 70].

ATT is a tracking rule, not a location permission. Apple announced it on 22 June 2020 as an iOS 14 requirement, deferred it on 3 September 2020 to "early next year" and made it mandatory for all apps on 26 April 2021 with iOS 14.5. Apple's definition of tracking includes sharing user or device data with data brokers [68, 76, 77, 78, 69].

Privacy manifests followed. Apple introduced them, with SDK signatures and required-reason APIs, at WWDC23. On 29 February 2024 it dated the requirement: warning emails from 13 March 2024 and, from 1 May 2024, rejection of uploads that use a required-reason API without an approved reason. Tracking domains declared in a manifest are blocked unless the user has granted ATT permission [497, 79, 80, 498]. Whether any location API is a required-reason API is not established here, because the category list could not be read [497, 79, 80, 498]. Radar's iOS SDK manifest, last changed on 1 June 2024, shows what a location vendor declares: precise and coarse location collected and linked to the user, tracking set to false, and purposes limited to app functionality and analytics [406, 304]. That is the vendor's own declaration.

At WWDC24 Apple introduced CLServiceSession. Under it, Always authorization is effective only while the app holds a session started in the foreground, and approximate location is delivered only every 15 to 20 minutes. iOS 18 shipped on 16 September 2024 [73, 74, 499]. The 2020 figure of four recomputations an hour and the 2024 figure for delivery are both Apple's, and this review does not reconcile them [68, 69, 70] [73, 74, 499]. iOS 26 shipped on 15 September 2025 and iOS 27 on 14 September 2026. Apple's Core Location change log lists no change after June 2024, and the Apple pages screened show no change to the user-facing location permission in either release. The WWDC25 and WWDC26 session listings were not opened [73, 74, 499].

iCloud Private Relay acts on IP, not on the location permission. It shipped as a public beta with iOS 15 on 20 September 2021 for iCloud+ subscribers who enable it. It replaces the client IP with a relay IP that maps by default to a roughly city-level area, chosen from a four-character geohash of roughly 800 km2, or only to country and time zone. It covers Safari, DNS queries and insecure HTTP app traffic, and leaves HTTPS in-app traffic and its IP geolocation untouched [71, 72, 429, 430]. No Apple adoption figure was found, so the share of traffic affected is unknown [71, 72, 429, 430].

### Google

Android 10, released on 3 September 2019, made background location a separate permission, ACCESS_BACKGROUND_LOCATION, and let users allow location only while an app is in use. Apps targeting Android 9 or lower had background access added automatically. Wi-Fi and Bluetooth scanning began to require fine location for apps targeting Android 10, and MAC addresses were randomised by default [82, 83].

Android 11 began rolling out on 8 September 2020. It added a one-time grant, "Only this time", and removed the background option from the runtime dialog, so "Allow all the time" can be set only in Settings. Apps targeting Android 11 must request background location separately, and unused apps have their permissions reset [84, 85]. Android 12, pushed to AOSP on 4 October 2021, lets users grant only approximate location even when an app asks for precise, in the dialog or in Settings for any app. Google documents approximate location as accurate to within about 3 square kilometres [86, 87, 431].

Google Play moved alongside the OS. On 19 February 2020 it announced that background location would need approval, with a projected timeline, subject to change, of 3 August 2020 for new apps and 2 November 2020 for existing ones. On 8 September 2020 it said it would not enforce the policy for existing apps until 2021 [88, 84, 94, 500]. The final 2021 date was not read from a Google page [88, 84, 94, 500]. The current policy limits all device location, foreground and background, to functions central to the app's core purpose, never permits it solely for advertising or analytics, and reserves background access for critical features [88, 84, 94, 500].

The Play policies, as read on 24 October 2026, list device location as personal and sensitive user data and prohibit selling it, where a sale is a transfer to a third party for money. Developers must ensure their SDK providers do not sell it. The Permissions policy adds "Don't sell device location" [501, 94]. The pages carry no last-updated dates, so when the sale clause entered is not established [501, 94]. The FTC's January 2024 complaint records a paid SDK market: the X-Mode SDK sat in more than 300 apps, and X-Mode paid developers per device whose location it collected [502, 503].

The Data safety section was pre-announced on 6 May 2021, shown to users from 26 April 2022 and mandatory after a grace period ending 20 July 2022. Declarations must cover data collected through third-party SDKs and location inferred from IP address. They define approximate location as an area of at least 3 km2 and precise location as less than that [432, 89, 90]. The labels are self-declared, and the threshold differs from the radius definitions in state laws ([Chapter 7](#regulation)) [432, 89, 90].

On the buying side, Google's Authorized Buyers guidelines (last updated 15 May 2026) state that Google does not share precise location through the programme. Precise location a buyer receives by other means may be used only to bid on that impression and kept no longer than that needs. Program Data may not be sold, and non-aggregated Program Data may be kept for at most 18 months [433]. The guidelines bind Google's exchange only [433]. Prebid coarsens coordinates only when a rule or privacy module denies precise geo, its default is to allow, and its Android mobile SDK sets no precision limit by default [437, 438, 439]. [Chapter 6](#audiences) takes up bidstream-derived location.

Chrome's IP Protection was designed for Incognito mode only, for third-party requests to listed domains, through a two-hop proxy whose IPs represent coarse location with a public city-level geofeed. Google said it would reach Chrome Stable no sooner than July 2025 [434, 435, 436, 91]. On 17 October 2025 Google announced the retirement of the Attribution Reporting API, Protected Audience and Topics on Chrome and Android, of SDK Runtime and of IP Protection, among others, with no retirement dates. CHIPS, FedCM and Private State Tokens continue [91]. Google's documentation classed IP Protection by 14 August 2026 as scheduled for phase-out with no date, and whether it ever reached Stable is not established [434, 435, 436, 91].

### The X-Mode SDK removal

In December 2020 Apple and Google barred apps in their stores from using X-Mode's SDK. This review found the event only in trade reporting, from AdExchanger in February 2021, and opened no statement by either platform [60]. Under the Q4 rule the date therefore stands as reported, not as documented by the platforms. It is coded a platform SDK removal under Q3, not a regulatory event [60]. The FTC's January 2024 complaint against X-Mode Social and Outlogic does not mention the removal [502, 503]. [Chapter 7](#regulation) covers the order.

### What sources attribute to the changes

| Source | Change named | What it states | Q4 code | Evidence |
|---|---|---|---|---|
| Google, 19 February 2020 | Android 10 location options | Over half of users select "While app is in use"; many apps asking for background location did not need it | Documented, for permission grants; no vendor panel | [88] |
| Foursquare methodology documentation | iOS 13, OS updates | Names iOS 13 among causes of step changes in raw visits and reweights to offset them; no size | Asserted | [240] |
| Near Intelligence S-4/A, February 2023 | iOS 14 approximate location, iOS 14 and 15 opt-in measures, Android 12 | Lists them under "key trends around the lack of precise location data"; no figure for its own supply | Asserted | [504] |
| Trade article on Cuebiq, 23 February 2021 | ATT, set beside the move by the article | Cuebiq sunsets its proprietary SDK for a clean room; its CEO cites privacy and security risk | Asserted; the link to ATT is the article's | [60] |
| Working-paper abstract, April 2022 | ATT | Collection frequency from iOS devices fell for one multi-app provider | Documented for direction only (independent measurement); no size | [110] |
| SafeGraph sampling rate, academic reading | None | Dips 2020-2021 and recovers in 2022; authors cite the pandemic | Coincident at most | [95] |
| Telenav 10-Ks, FY2019 and FY2020 | None | No mention of iOS 13 or location permissions | No effect coded | [505, 504, 88] |

The one figure a platform attaches to a change is about choices. Google said on 19 February 2020 that on Android 10 "over half of users select 'While app is in use'" when offered the new options [88]. It describes user decisions, not the volume of observations reaching SDK vendors, and cannot be turned into a supply change. No equivalent Apple figure was found [88].

Vendors that name a change give no size. Foursquare adjusts its normalised-visit weights for "changes to how we calculate a visit in our SDK due to an OS update", and names iOS 13 as one cause of step changes in raw visits [240]. Near's registration statement treats the platform changes as market trends and attributes no change in its own supply to them [504]. Cuebiq's sunset of its proprietary SDK in February 2021 is a documented partial shift away from its own SDK supply, with no panel figure. The ATT link comes from the article, not from Cuebiq [60].

The independent evidence is thin and indirect. An April 2022 working-paper abstract, using data from more than 7 million iOS and Android devices, reports that collection from iOS devices fell after ATT, both because devices went unobserved and because observed devices sent fewer pings. It names no vendor and gives no size in the abstract read [110]. In this paper's reading it attributes a direction to ATT, which is a tracking rule rather than a location permission [68, 76, 77, 78, 69]. SafeGraph's residing-device sampling rate fell from 8.1% in 2019 to 7.2% in 2020 and 6.6% in 2021, then rose to 8.5% in 2022. The authors attribute the fall to the pandemic and do not discuss the permission changes, so the series is at most coincident with them [95].

### The bounded finding

In the sources screened by the cutoff, this review found no filing, official platform statement or independent measurement that attributes a quantified change in US SDK-sourced location supply to a named Apple or Google permission change. Telenav, which sold its Thinknear advertising business to inMarket under an agreement dated 8 August 2019, does not mention iOS 13 or location permissions in its FY2019 and FY2020 10-Ks. EDGAR full-text searches for 10-Ks pairing location data with ATT, iOS 13 or background location returned no location vendor [505, 504, 88] [505]. The search has limits. Foursquare, Placer.ai, InMarket, GroundTruth and Blis are private and file no 10-K or 10-Q reports, the filings these searches covered (Placer Labs has filed Form D notices of private offerings), and the stream that ran these searches did not screen vendor blogs [505, 504, 88]. Under the Q4 rule, an absence of attribution is not evidence of no effect.

The absence matters for H1. No US vendor in the sources opened discloses a panel series on one definition from before 2021 to after 2023, so H1 is unresolved on this evidence ([Chapter 2](#supply)) [275, 63, 289, 427, 428]. Published validation has the same gap in time. Every independent SafeGraph ground-truth comparison found uses data from 2018 to 2020, so it describes panels before or during the changes, not at the cutoff ([Chapter 4](#validation)) [98, 99, 100, 101].

What the documentation does establish is where the changes reach. The OS permissions govern what an SDK may collect, and approximate grants deliver coarse location by design [68, 69, 70] [86, 87, 431]. Radar's SDK, for example, requests an accuracy level from the OS, and what it receives depends on the device, the environment and the user's choice [300]. The IP measures cover Safari and insecure app traffic for subscribers who opt in, and Incognito third-party requests in Chrome, which is being withdrawn [71, 72, 429, 430] [434, 435, 436, 91]. Outside Google's exchange, bid requests may carry device coordinates under each exchange's own terms, and Prebid coarsens them only when a publisher rule or privacy module denies precise geo [433] [437, 438, 439]. A buyer can ask a vendor which permission level its panel devices granted, and whether its source mix changed after 2020, with figures ([Chapter 12](#next)).



## 4. Do the numbers match the ground?


*The test that would settle it is missing from the record screened. In the sources screened by the cutoff, no independent study compares a vendor's projected retail visits with door counts or transactions across many places and months, and no vendor publishes a held-out validation with error statistics on the pages opened. The independent checks that exist are narrow and old: a 2018 polling-place audit of SafeGraph's panel, park and trail counts from 2018 to 2021, and one trail-site check of Placer.ai. H2 is unresolved overall, with adverse historical evidence for SafeGraph's 2018 panel.*

> **In short.**
> In the sources screened by the cutoff, this review found no independent comparison of any named vendor's projected retail visits with door counts or transactions across many places and months, and no vendor publishes a held-out validation with error statistics on the pages opened [98, 99, 102]. Census comparisons show SafeGraph's device counts tracking population closely at county level and less closely at tract level; they test who is in the panel, not whether visits are counted correctly [95]. A 2018 audit against North Carolina's voter file found lower coverage at polling places with older and with more non-white voters [99]. In Yellowstone, SafeGraph visits at the two most-visited points of interest correlated with official monthly visits at 0.85 and 0.88; at one Minnesota recreation area, Placer.ai's total ran 30.5% above trail counters in a comparison that is not like for like [98] [101]. Every independent SafeGraph ground-truth comparison found uses 2018-2020 data, so published validation does not describe the panels as they were at the cutoff [98, 99, 100, 101]. Azira's vendor-asserted correlation of 0.85 and median absolute percent error of 10%, the only such comparison in the cohort A documentation opened, comes with no design, sample, period or places [274, 254, 5]. The bidstream accuracy series found are vendor-run and date from 2014 to 2019, and no peer-reviewed measurement with published statistics was found [441, 442, 443] [506, 507] [508, 509]. In randomised Facebook benchmarks, matching and regression methods mostly overstated lift [111]. H2 is unresolved overall, with adverse historical evidence for SafeGraph's 2018 panel and use [98, 99, 102].

H2 says that published visit estimates from US location-measurement and location-intelligence products agree with independent ground-truth counts (door counters, transactions, official visitor counts) within the error the vendor states. Its registered rule counts two kinds of test: a comparison with a published method by a party without a financial interest, and a vendor comparison with a held-out design and published method. Support is recorded per product and per place type, and it does not carry to a different panel vintage. Absence of accessible comparisons codes as unresolved, not contradicted ([Chapter 13](#method)).

MRC's 2017 location guidelines set the industry's own bar: they encourage first-party "ground truth" validation and set a general expectation of less than 5% error on primary reported characteristics, with disclosure above that [445]. Projected visits, as [Chapter 1](#defining) defines them, are observed panel counts scaled to a population by a model. Placer.ai's visit numbers are projections of this kind, and its own example scales 74.8 thousand panel visits to 2.4 million [3, 102, 289]. Studies the searches did not surface could exist [98, 99, 102].

![No independent comparison with retail door counts or transactions was found for any named vendor; the third-party checks that exist use census, park and polling-place counts.](figures/fig-ground-truth.svg)

*Figure fig-ground-truth. **No independent comparison with retail door counts or transactions was found for any named vendor; the third-party checks that exist use census, park and polling-place counts.** Each panel or vendor against four kinds of reference count. A third-party comparison is one with a published method by a party other than the vendor; it counts as independent only where its funding is stated, and four of the studies charted (Kupfer, Leslie, Choi, UW 2022) state none. The retail column holds only vendor claims. The third-party comparisons that exist are historical and narrow: the newest data end in January 2023, so no comparison covers a panel vintage sold at the cutoff. Units: data period of each comparison. Scope: comparisons found in the sources screened by the cutoff; US unless marked. Sources [95], [96], [97], [98], [99], [100], [101], [102], [103], [104], [105], [5], [106], [14], [107]. Caveat: Absence is bounded to the searches logged in stream B and its verifier. Agreement with census population validates coverage, not visit counts. H2 stays unresolved overall (LOC-B-F31). Data: data/figure-data.json#ground_truth*

### Who is in the panel

Census comparisons test who is in a panel, not whether its visits are counted correctly [95]. SafeGraph's own 2019 analysis (a vendor blog and notebook) imputed each device's demographics from its home census block group and compared them with national census totals (r = 0.99 for education and income), and compared device counts with county population (r = 0.97; a block-group correlation of 0.176 per Li et al. 2024). It concluded the panel was "well-sampled across demographic categories" and did not test bias in place-specific visit inferences [99].

An independent census comparison for 2018-2022 put SafeGraph's monthly sampling rate at 7.5% on average, with monthly values from 4.5% to 14.5% across the national, urban and rural series. In 2019, device counts tracked population closely by county (Pearson r 0.986 urban, 0.934 rural) and moderately by tract (r above 0.683 urban, above 0.830 rural), and the urban-tract fit weakened to r-squared 0.375 by 2022 [95]. Hispanic residents were under-represented from March 2020 to July 2021, low-income and low-education groups slightly under-represented, more so during the pandemic, with the pattern reversed at state level in 2022, and most group biases fell within about plus or minus 0.05; rural units were sampled at higher rates than urban ones at tract and block-group level from April 2019 [95].

Aggregate flows agree with other aggregates, but no ground truth exists for them. SafeGraph-derived county population flows for March-May 2020 correlated with pre-pandemic commuting flows from the American Community Survey (ACS) at 0.974 to 0.985, and with another vendor's mobility indices in five metros at 0.92 or above. The authors state that no ground truth exists for these flows [96].

Other panels show bias that changes with place and source. The only Veraset coverage study found, a June 2026 preprint on 2,478 Mexican municipalities, finds users concentrated in larger, wealthier and better-connected places, with spatially clustered bias that differs from Facebook's single-app data [105]. Two 2025 preprints find bias in all ten US cities studied but specific to each city, and find that multi-app UK datasets do not necessarily reduce bias relative to single-app sources; the US authors say this raises serious questions about general debiasing approaches [108, 109]. For Cuebiq, a 2017 Federal Highway Administration case study found biases by traveller age and income, and a 2022 University of Washington analysis of 11 metros found the direction of racial bias varying by metro [104, 103].

### Polling places: coverage bias by place

SafeGraph's 2018 panel under-covered polling places with older and more non-white voters. An independent audit tested its point-of-interest data (about 47 million US devices) against North Carolina's 2018 general-election voter file. Polling places with older and with more non-white voters had lower coverage: rank correlations of coverage with the share of voters over 65 of -0.14 (p < 0.001) and with the share of non-white voters of -0.11 (p = 0.0067) [99]. Election-day traffic tracked turnout in rank (0.383, p < 0.001), but place-level estimates were noisy: the root-mean-squared error was 1,375 voters, and polling places that registered 20 marginal SafeGraph devices had roughly 300 to 2,300 actual voters [99].

Allocating a resource by SafeGraph traffic rather than by actual voters would have under-allocated by 37% to the oldest and most non-white group of polling places and over-allocated by 33% to the youngest and whitest group (both p < 0.05) [99]. This is the adverse evidence in the H2 reading. It covers SafeGraph's 2018 panel and this use, not other products or later vintages [98, 99, 102].

### Parks, trails and one Placer.ai check

Most of the independent ground-truth comparisons found use parks and trails, which keep official counts [98, 99, 102]. A peer-reviewed study of six western national parks compared SafeGraph records with National Park Service (NPS) statistics for 2019-2020 and concluded that device data capture visitation patterns at finer granularity; the abstract read reports no correlation statistic [97]. In Yellowstone, SafeGraph Patterns visits from January 2018 to November 2020 correlated with NPS monthly visits at r = 0.88 and 0.85 for the two most-visited points of interest, with a trail counter's daily counts at 0.76 (2018) and 0.77 (2019), and with park-wide gate vehicle counts at 0.79 in 2019 but 0.47 in 2020, when the compared facility imposed indoor capacity limits [98]. The same study found only 40 of the 80 SafeGraph points of interest inside the park valid, visitor demographics imputed from the ACS differing significantly from the visitor survey on education and income, and no way to assign a home block group to short-term international visitors, 23% of visitors per NPS surveys [98]. Inference: a correlation shows that two series rise and fall together; it does not show that their levels agree.

One later panel has a park check. A 2025 preprint using Advan Weekly Visit Patterns (about 35 million US devices from about 1,000 apps) for January 2019 to January 2023 reports a Pearson correlation of 0.875 with NPS visitation for the ten most-visited parks, a comparison the authors call qualitative and based on visual inspection [100].

In the only third-party comparison of Placer.ai found, a University of Minnesota Extension case study whose funding is not stated (April 2024) at one recreation area, Placer.ai estimated 13,233 visits and StreetLight 14,544 for 29 May to 6 September 2021, 30.5% and 43.5% above the 10,138 recorded by trail counters. The authors found no statistically significant difference, but the comparison is not like for like: polygons differ, counters cover only the cross-country trails, and events are handled differently [101]. The platforms showed weekends 7 times busier than weekdays, against 2.1 times in the counter averages [101]. Placer.ai's inferred visitor demographics differed significantly from the intercept survey's: 30.8% with a bachelor's degree or higher and 21.8% with household income over US$100,000, against 62% and 61% [101].

### Vintage and restatement

Every independent SafeGraph ground-truth comparison found uses data from 2018-2020, before or during the iOS 13 and 14, App Tracking Transparency and Android 12 changes. The only later third-party comparisons are the Advan park series and the Minnesota case, neither with its funding stated, so published validation does not describe the panels as they existed at the cutoff [98, 99, 100, 101]. The panels moved in the meantime: SafeGraph's sampling rate peaked at 13.4% in May 2022 and dropped to 4.9% in July 2022, a fall the authors say "may relate" to a data-provider disruption [95]. A working-paper abstract (April 2022) reports that one multi-app provider's iOS data collection fell after App Tracking Transparency ([Chapter 2](#supply)) [110].

Vendors also change the past. Placer.ai's Data Version 2.0 (31 March 2024) updated all historical data from 1 January 2017, and its accuracy page says de-biasing adjustments happen "usually no more than once a year" and "can sometimes impact previously run numbers". A Placer.ai series read at two dates can differ for the same past period [285, 102].

### What vendors say about their own accuracy

The vendor record is assertions without method. Of the documentation opened on 24 October 2026 for the fourteen cohort A products, only Azira's undated white paper reports a comparison with an independent reference: a vendor-asserted median absolute percent error of 10% and a correlation of 0.85 between extrapolated visits and "ground truth sales data and foot traffic counters", with no study design, sample, period or places disclosed [274, 254, 5]. No product publishes such a comparison as a documented study [274, 254, 5].

| Vendor | What it publishes | Reference named | Statistic and method | Evidence class |
|---|---|---|---|---|
| Azira | White paper, undated | Sales data and foot-traffic counters | Correlation 0.85, median absolute percent error 10%; no design, sample, period or places | Vendor assertion [274, 254, 5] |
| Placer.ai | Home page and help centre | Baseball attendance, transactions, store revenues, conference attendees, vehicle counters | "+92% accuracy" on the home page; no result, sample or statistic | Vendor assertion [293, 290, 102, 289] |
| Unacast | Two posts | One unnamed clothing retailer; 2019 plant headcounts | R-squared "of 91.6% or higher", no places, period or holdout; four of five plants "within 20%" | Vendor assertion [106, 14] |
| GroundTruth | Audit posts, 2017-2019 | Receipts from 10,000 matched panelists | "94.4% visit accuracy"; no report, period, method or error statistic | Commissioned [259, 258] |
| Foursquare | Trend check | Public financial statements, transaction trends | No places, period or statistic | Vendor assertion [236] |
| InMarket | Validation statements | Comscore; unnamed "truth sets" | No report, date or scope | Vendor assertion [274, 40, 267, 374] |
| Huq (Germany) | Footfall page, 6 August 2026 | "Hundreds of authoritative external sources" | Median MAPE 14.5%; references, sites and period unnamed | Vendor assertion [116, 117] |

Several of these measure something other than visit counts. GroundTruth's audit checks whether recorded visits are confirmed, not whether visit counts match a reference, and its help pages later claim "99% accurate as verified by Numerator" [259, 258]. Unacast's plant comparison uses headcount rather than visits [14]. Advan's 2019 "99.5% accuracy of geofences" is a claim about geofence geometry [116, 117]. GroundTruth asserts, without a cited study, that radial geofences measure visits "with up to 75% inaccuracy" [254, 361]. Foursquare's "Third-party verified" item is a Forrester Total Economic Impact study commissioned by Foursquare (June 2022), not a validation [236]. Its visit model trains on check-ins, and a 2013 third-party study of 244 users, funding not stated, found check-ins covered about 10% of the places users visited, with about 75% of check-ins matching no GPS-detected visit [355, 28].

Third-party and production uses do not fill the gap. A site-selection vendor's blog reports that Placer.ai visit trends and six chains' SEC-reported traffic or transaction figures "agreed on direction 9 times out of 12" across 12 retailer-quarter comparisons [383]. The Federal Reserve Bank of Chicago's weekly retail index lists SafeGraph and Advan Research among its sources and benchmarks to Census retail sales; that is documented production use, not a published validation of visit counts [512]. Geopath reports that its reach model aligns with its own mobile input data to within ±5%, which is not a comparison with an independent count ([Chapter 9](#dooh)) [184]. Under the registered rule, a vendor comparison without a held-out design or published method leaves a product unresolved [98, 99, 102].

### Bidstream location: historical and vendor-run

The public series on bidstream location accuracy found in the sources screened are vendor-run and date from 2014 to 2019 [441, 442, 443] [506, 507] [508, 509]. Thinknear's Location Score Index reported in 2014 that 34% of location-bearing impressions were within 100 metres of the user, and its Q1 2015 edition, on more than one billion impressions, put 37% within 100 m and 10% beyond 100 km [441, 442, 443]. A May 2016 edition put 32% within 100 m, the lowest industry score since tracking began [514, 515]. Placed's 2015 analysis of exchange bid requests against its own panel found that 27% carried any location, that 91% of observed business visits generated no ad request and that about 73% of bid requests occurred in transit [440]. Location Sciences, which sold a verification product, said in September 2019 that only 14% of 550 million impressions it verified carried high-quality GPS signals [506, 507]. Cuebiq's 2017 paper cites a 2015 MMA report for "less than 33%" of exchange location data being accurate within 100 metres; neither the stream nor its verifier could read the report [118].

Verification vendors have their own interests. Location Sciences' 2019 verification of Blis was a paid engagement; in March 2021 Location Sciences stated that Blis threatened legal action after Verify reported "data manipulation", and Adweek reported that Blis denied the allegations [343, 231, 232, 233]. This review found no peer-reviewed measurement of bidstream location accuracy with published statistics, on web searches only; the ACM, IEEE, USENIX and PETS databases were not searched directly. The 2017 ADINT study showed that ads bought for about US$1,000 could localise a device to about 8 metres once it stayed about 5 minutes [508, 509]. Google's Authorized Buyers documentation (updated 17 September 2026, read on 24 October 2026) says its bid-request location comes only from IP geolocation, with latitude and longitude precision limited to 0.01 degrees [444]. The author's essays argue that a mobile impression's location carries real error; that is his position, not evidence. His geographic cell-grid proposal, AdCP issue #5969, which he opened and which was open when read on 24 October 2026, sits in the AgenticAdvertising.org working group he co-leads and has a stake in the answer [205, 206, 207].

### The causal benchmarks

Agreement with ground truth is one question; whether a visit lift is causal is another, coded under H3 in [Chapter 5](#measurement). In the randomised benchmarks, observational methods miss the experimental lift, usually by overstating it [111] [112] [114]. Fifteen US Facebook campaigns run as randomised trials in 2015 found matching and regression methods on rich individual data mostly overestimating lift, and in half of the studies purchase lift was off by a factor of three across all methods. Two coauthors are affiliated with Facebook and the study runs on Facebook's own data; the two Kellogg academics declare no financial interest [111]. Across 663 Facebook experiments, median RCT lifts of 29%, 18% and 5% by funnel stage compared with 173%, 176% and 64% from stratified propensity-score matching, and the authors conclude they could not reliably estimate a campaign's causal effect [112].

Randomisation does not make the problem easy. Across 25 field experiments with US retailers and brokerages, the median confidence interval on return on investment was over 100 percentage points wide, and informative experiments can need more than 10 million person-weeks [113]. A randomised experiment with 1.6 million customers of a major retailer on Yahoo! measured a 5% purchase increase from online ads, 93% of it in stores, while a cross-sectional observational estimate on the same data understated the effect, with a negative bias three times the experimental effect [114]. Ghost ads and geo experiments, both from Google-affiliated authors, are randomised designs with published methods that location attribution can borrow [115, 513]. Randomised field experiments of location-targeted promotions also exist, with store-level outcomes [510, 511]. Inference: none of the benchmarks screened tests visit lift measured by a location panel; the nearest ones measure in-store purchases matched by customer record and promotions redeemed on location [114] [510, 511].

### H2, coded

H2 stays unresolved overall, and no single verdict is given across products and data vintages [98, 99, 102]. SafeGraph's 2018 panel carries adverse historical evidence for one place type and use [98, 99, 102]. Placer.ai has one third-party trail-site check, funding not stated, that is not like for like, and Advan one visual park comparison [101] [100]. For the other products in the sources screened, the record holds coverage studies, vendor assertions or nothing [98, 99, 102]. No product meets the supported branch: no independent comparison, and no held-out vendor comparison with a published method, tests a named product against an error the vendor states [98, 99, 102] [274, 254, 5].

### What study would settle it

One way to close the current retail gap would be an independent study with this design; the registered rule would also accept a vendor-run comparison with a held-out design and a published method. Take a named product's projected weekly visits for a few hundred retail places of more than one type, and the door counts or transactions for the same places and weeks, supplied by the retailers. Fix the model version before the data arrive and hold out places and later weeks. Report error in levels as well as correlation in patterns, by place type, region and income band, with the vendor's stated error alongside. Repeat after the next model restatement.

The study table lists the 26 study records coded from independent, academic, commissioned and vendor work, one row per study and outcome, with design, sample, sponsor, vintage and uncertainty.

**Study evidence table**

| Study | Year | Design | Sample | Setting | Format / channel | Outcome measure | Headline result | Uncertainty | Sponsor | Generalisability | Src |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Coston et al. 2021, polling-place bias audit | 2021 | Panel visits compared with official in-person turnout per polling place; election-day traffic adjusted for non-election traffic; placebo distribution | 558 polling places | North Carolina, 2018 general election (data October-November 2018) | SafeGraph panel, about 47 million US devices as described | Place-level coverage and its correlation with voter age and race | Rank correlation of marginal traffic with turnout 0.383; RMSE 1,375 voters; coverage falls with share of voters over 65 (-0.14) and non-white (-0.11); policy simulation under-allocates 37% to the oldest, most non-white group | p<0.001 and p=0.0067 for the coverage coefficients (labelled preliminary); placebo p=0.024 | Academic (NSF GRFP, Stanford HAI and RISE); SafeGraph supplied the data, answered questions and gave feedback | One state, one election day, a special place type, 2018 panel vintage; adverse evidence for that panel and use only | [99] |
| Li et al. 2024, SafeGraph against census population 2018-2022 | 2024 | Monthly residing-device counts compared with census population at county, tract and block-group level; demographic bias from imputed home composition | US counties, tracts and block groups, 60 monthly releases | US, 2018-2022 | SafeGraph residing-device counts | Sampling rate and agreement with census population | Average sampling rate 7.5% (monthly range 4.5-14.5%; overall trend peak 13.4% May 2022, low 4.9% July 2022); 2019 county r 0.986 urban, 0.934 rural; tract r above 0.683 urban, 0.830 rural; urban-tract r-squared 0.651 (2018) to 0.375 (2022) | Correlations reported without intervals in the passages read | Academic authors; funding not recorded in the source record | Validates coverage against population, not visit counts; SafeGraph lineage only | [95] |
| Kang et al. 2020, multiscale mobility flows | 2020 | Origin-destination flows derived from the panel compared with pre-pandemic ACS commuting flows and Descartes Labs indices | US county flows, weekly and daily, March-May 2020 | US, March-May 2020 | SafeGraph panel (described as about 10% of the population; weekly active devices from over 22 million to 13 million) | Agreement of inferred flows with survey commuting flows | Pearson r 0.974-0.985 for inferred population flows, 0.915-0.961 for raw visitor flows; at least 0.92 with Descartes Labs in five metros | Not stated; the authors say no ground truth exists for these flows | Academic authors; funding not recorded in the source record | Flows, not place visits; pandemic period | [96] |
| Kupfer et al. 2021, six western national parks | 2021 | Social Distancing Metric visitation patterns compared with National Park Service visitation statistics | 6 national parks | Western US, 2019-2020 | SafeGraph Social Distancing Metric | Park visitation patterns | Qualitative agreement: device records 'can effectively capture park visitation patterns'; no coefficient in the abstract | Not reported in the abstract | Academic authors; abstract only read (funding not stated: independence not established, round 3 R3-11) | Parks, not retail; abstract only | [97] |
| Liang et al. 2022, Yellowstone visitor counts | 2022 | POI visit series compared with NPS monthly visits, a calibrated trail counter and gate vehicle counts; POI validity check against Google Maps | 2 POIs for correlations; 80 POIs checked | Yellowstone National Park, 2018-2020 | SafeGraph Patterns and Core Places | Correlation of visit series with official and counter series; POI validity | Monthly r 0.88 and 0.85 with NPS visits; daily r 0.76 and 0.77 with a trail counter 700 m away; daily r 0.79 (2019) and 0.47 (2020) with gate counts; 40 of 80 POIs valid | p<0.001 for the monthly correlations | Academic authors | Strongly seasonal park series inflate correlations; levels not compared; not retail | [98] |
| Leslie et al. 2025, Advan and NPS visitation | 2025 | Weekly panel visits aggregated by sector compared with NPS visitation by visual inspection | 10 most-visited national parks | US, January 2019 to January 2023 | Advan Weekly Patterns (raw visit counts; about 35 million devices, about 10% sample, as described) | Agreement of park visit series | Pearson r 0.875 | None; the authors call the comparison qualitative | Academic authors; data bought through a Dewey subscription (funding not stated: independence not established, round 3 R3-11) | Parks only; newest ground-truth data point found (January 2023) | [100] |
| Choi 2024 (University of Minnesota Extension), Detroit Mountain | 2024 | Platform visit totals compared with automated trail counters and an intercept survey at one recreation area | 1 site; survey n=116 | Detroit Mountain Recreation Area, Minnesota, 29 May-6 September 2021 | Placer.ai and StreetLight (vehicle trips times three passengers) | Visit totals, weekend ratio, visitor demographics | Placer.ai 13,233 (+30.5%) and StreetLight 14,544 (+43.5%) against 10,138 counter visits; weekend/weekday 7x on the platforms v 2.1x on counters; platform demographics differ from the survey | Authors report no statistically significant difference in totals | University extension case study; no funder or data arrangement stated (funding not stated: independence not established, round 3 R3-11) | One site; polygons differ, counters cover only the cross-country trails, events handled differently; not like for like | [101] |
| FHWA 2017 (RSG), locational app data for origin-destination analysis | 2017 | App location data compared with a smartphone GPS household travel survey | Not stated in the record | US, 2017 | Cuebiq app location data | Demographic and trip-length bias | Biases in traveller age and income; trips longer than 9 miles over-represented | Not stated in the record | Federal Highway Administration (government report) | Travel survey, not visits; 2017 vintage | [104] |
| University of Washington bigdata4mobility 2022, Cuebiq sampling bias | 2022 | Device counts compared with block-group population by race, assuming no bias within block groups | 11 metropolitan statistical areas | US MSAs; data vintage not stated | Cuebiq dataset (named in the title) | Device-to-population bias by group | Los Angeles: other race groups -9.60% against white -1.04%; pattern reversed in Boston and Seattle | Not stated | University research project (funding not stated: independence not established, round 3 R3-11) | Coverage, not visits; period not stated | [103] |
| Cabrera et al. 2026, representativeness of GPS mobility data in Mexico | 2026 | Two sources' coverage compared with the 2020 census | 2,478 municipalities | Mexico; Veraset data 15-22 March 2020; 2020 census benchmark | Veraset multi-app data and Facebook data | Coverage bias | Multi-app data concentrate users in larger, wealthier, more digitally connected places; Facebook coverage higher and more even | Model-based; not recorded in this review | Academic preprint; UK ESRC funding (ES/Y010787/1), no conflict declared, public code (full text read 24 October 2026) | Non-US; coverage, not visits | [105] |
| Ten-city US data-production bias preprint 2025 | 2025 | Tract-level models of data production against census demographics | 10 US cities | US; period not stated on the abstract page | Mobility data, vendor unnamed | Data produced per resident by tract characteristics | Bias in all ten cities, specific to each; wealth, ethnicity and education predict data production | Not read (abstract) | Funding not stated | Vendor unnamed; cannot be attributed to a product | [108] |
| UK four-dataset population-coverage preprint 2025 | 2025 | Population coverage indicators for four app datasets against the 2021 census | 4 datasets | UK, 2021 census | Single-app and multi-app mobile phone datasets | Coverage bias | Multi-app datasets do not necessarily reduce bias against single-app sources; phone data cover more of the population than national surveys | Not read (abstract) | Funding not stated | Non-US; coverage, not visits | [109] |
| App Tracking working paper (Bocconi abstract), supply after ATT | 2022 | iOS against Android ping frequency around Apple's ATT change, with matching and placebo tests | More than 7 million iOS and Android devices | Not stated in the abstract | One multi-app location data provider (unnamed) | Frequency of data collection from iOS devices | Collection from iOS devices decreases after the policy change, through devices becoming unobserved and fewer pings; more pronounced for visitors to privacy-sensitive places | Magnitudes not given in the abstract | Funding not stated; data presumably from the provider | One unnamed provider; abstract only | [110] |
| Gordon, Zettelmeyer, Bhargava and Chapsky 2019 | 2019 | Observational lift methods benchmarked against randomised controlled trials | 15 RCTs; 500 million user-experiment observations; 1.6 billion impressions | US Facebook advertising campaigns, 2015 | Online ads; purchase and other online outcomes | Error of observational lift against experimental lift | Matching and regression methods mostly overestimated lift; in half of the studies purchase lift was off by a factor of three across all methods | Per-study confidence intervals in the paper (not charted here) | Academic-Facebook collaboration on Facebook data; two coauthors affiliated with Facebook; the Kellogg academics declare no financial interest | Online outcomes, not panel-detected store visits | [111] |
| Gordon, Moakler and Zettelmeyer 2023, 'Close Enough?' | 2023 | Double machine learning and stratified propensity-score matching against RCTs | 663 RCTs; more than 5,000 user features | Facebook advertising experiments | Online ads; funnel outcomes | Median lift by method and funnel stage | Median RCT lift 29% / 18% / 5% (upper, middle, lower funnel) against DML 83% / 58% / 24% and SPSM 173% / 176% / 64% | Abstract only; per-experiment error statistics not read | Facebook experiment data; authors' affiliations as given in the paper | Online outcomes, not panel-detected store visits | [112] |
| Lewis and Rao 2015, the unfavourable economics of measuring returns | 2015 | Power analysis of randomised field experiments with sales outcomes | 25 field experiments; US$2.8 million of spend | Yahoo!, major US retailers and brokerages, before 2015 | Online display ads; individual sales | Precision of return-on-investment estimates | Median confidence interval on ROI over 100 percentage points; individual sales coefficient of variation about 10; more than 10 million person-weeks needed for an informative test | As stated (interval widths) | Authors former Yahoo! employees, as acknowledged | Sales, not visits; visit outcomes are more frequent, which changes the power arithmetic | [113] |
| Lewis and Reiley 2014, online ads and offline sales | 2014 | Individual-level randomised experiment with matched retailer purchase records; observational estimators run on the same data | 1.6 million retailer customers | Yahoo!, one major US retailer, before 2014 | Online display ads; in-store and online purchases | Purchase lift; bias of observational estimators | 5% purchase lift, 93% of it in stores and 78% from people who never clicked; a cross-sectional estimator showed a negative bias three times the experimental effect | As reported in the paper (abstract read) | Yahoo! research | Purchases matched by customer record, not visits detected by location; the bias runs opposite to the overestimates above | [114] |
| Johnson, Lewis and Nubbemeyer 2016, ghost ads | 2016 | Ghost-ad control (counterfactual exposure logging) against PSA and intent-to-treat designs | One retailer's display retargeting campaign | Online display, before 2016 | Display retargeting; website visits and purchases | Lift and cost of measuring it | Website visits +17%, purchases +11%; equal precision at least an order of magnitude cheaper; PSA experiments invalid under performance-optimising algorithms | As reported in the working paper | Authors affiliated with Rochester, Netflix and Google | Method demonstration; not applied to store visits measured by a location panel | [115] |
| Placer.ai visitation accuracy page |  | Vendor statement of testing against customers' first-party data and public sources; no method, sample or holdout | Not stated | US; undated | Placer.ai projected visits | Agreement with MLB attendance, transactions, store revenues, conference attendance and vehicle counters | No statistic published | None | Vendor | Not assessable | [102] |
| Unacast ground-truth post, five manufacturing plants | 2021 | Vendor comparison of average workday foot traffic with 2019 employee counts from online sources | 5 plants | US; 2019 reference; post published 25 February 2021, updated 5 August 2024 | Unacast foot-traffic estimates | Agreement with headcount | 'All other data matched within 20%', the Everett plant the only outlier; values only in an image | None | Vendor | Headcount is a proxy, not visits; five sites | [14] |
| Unacast foot-traffic accuracy statement |  | Vendor statement of fit against one unnamed nationwide clothing retailer's foot traffic | Not stated | US; undated | Unacast foot-traffic estimates | Fit to a retailer's foot traffic | R-squared 'of 91.6% or higher' | None; no places, period, sample or holdout stated | Vendor | Not assessable | [106, 14] |
| Azira white paper, extrapolated visits against ground truth | 2025 | Vendor assertion of comparison with 'ground truth sales data and foot traffic counters'; no study design, sample, period or places disclosed | 'Millions of campaigns' (as stated) | Undated white paper (c. 2025) | Azira extrapolated total visits | Error of extrapolated against actualised visits | Median absolute percent error 10%; correlation 0.85 | None | Vendor | Not assessable; vendor assertion | [5] |
| Huq footfall accuracy statement | 2026 | Vendor statement of regression testing against 'hundreds of authoritative external sources' (not named) | Not stated | Germany; post of 6 August 2026 | Huq footfall estimates | Error against ground truth | Median MAPE 14.5% | None; sites and period not stated | Vendor | Non-US; references unnamed | [116] |
| Advan geofence accuracy guarantee | 2019 | Vendor quality assurance of geofence geometry by manual review | 1.2 million locations of 1,541 brands | US; 11 June 2019 | Advan geofences | Geofence accuracy | 99.5% accuracy guaranteed | None | Vendor | Geometry, not visit counts | [117] |
| Azira visit confidence scoring | 2025 | Vendor method description: polygons drawn from satellite imagery and POI data, with a confidence score | Not stated | US; 2025 documentation | Azira visit detection | Visit confidence | No error statistic | None | Vendor | Method description only | [5] |
| Cuebiq 2017 benchmark paper, citing MMA 2015 | 2017 | Second-hand citation of an industry report gated behind a member login | Not stated | US; 2015 report cited in a 2017 vendor paper | Exchange-collected (bidstream) location | Accuracy within 100 metres | 'Less than 33%' accurate within 100 m | None; sample and method untraced | Vendor citing an industry body's vendor-sourced figure | Not assessable | [118] |



## 5. How visit lift is measured


*Most US visit-lift products compare exposed devices with a modelled or matched control, not with a group held out from the campaign. Of the nine cohort A products that document a default method, seven use no holdout, so H3 is contradicted. The comparator decides the number: one campaign reported 29% lift against its holdout and 310% against the open population. Every product that documents the basis of its visit totals projects them, though three also report observed counts.*

> **In short.**
> Of fourteen cohort A seed products, nine document a general visit-measurement method that governs their ordinary reporting, on documentation read on 24 October 2026 [5, 19, 119, 17, 20, 21, 22, 25, 6]. Six use a matched or synthetic control (Foursquare, InMarket, Reveal, Adsquare, Cuebiq, Arrivalist), two a randomised or holdout control (Azira's PSA holdout and Blis's geographic holdout), and GroundTruth reports exposed-only visits with no lift claim [5, 19, 119, 17, 20, 21, 22, 25, 6]. Seven of the nine documented methods are non-holdout; of the eight that claim lift, six are matched or synthetic, so H3 is contradicted under either count [5, 19, 119, 17, 20, 21, 22, 25, 6]. One undated PlaceIQ case reports a 29% lift against a standing holdout and a 310% "open lift" against the open population for the same campaign [12]. Azira is the only vendor whose opened public pages disclose numeric minimum-sample and significance rules [17, 5, 8, 518]. Disclosed attribution windows run from 4 days to a 30-day look-back [8, 5, 6, 17, 10, 118]. Where the basis is documented, visit totals are projected (Foursquare, InMarket, GroundTruth Projected Visits, Azira), with observed counts alongside at GroundTruth, Azira and InMarket; for nine visit numbers, Arrivalist's arrival counts among them, the basis is not disclosed [16, 17, 18, 5, 449]. No opened cohort A documentation names Samba, the author's employer, as a source of TV exposure data [22, 119, 274]. Among the products opened, only GroundTruth states a payment model, Cost Per Visit [266].

H3 asks whether the visit-lift numbers that US location-attribution products report rest, by default, on randomised or held-out control groups. The rival hypothesis, from the author's glossary, is that footfall attribution is correlated and modelled by default; it is a hypothesis, not evidence. The registered test codes each cohort A product's documented default as randomised or holdout (ghost ads, PSA holdouts and geo-experiments included), matched or synthetic, or exposed-only, and H3 is contradicted when more than half of the products with a documented default use a matched or exposed-only design ([chapter 13](#method)). A clarification dated 24 October 2026, written after extraction, fixed the decision tree: below four codable products, or at exactly half, the reading is unresolved, a product-wide method description counts as the default even without the word "default", geographic holdouts count as holdouts, and a product that makes no lift claim is reported in a separate strict count. The clarification did not move the verdict; better sources did. This chapter uses the paper's terms from [chapter 1](#defining): visitation counts visits after exposure, lift is a difference against a control, and a matched comparison is not causal evidence.

### What each product counts

Visit definitions are thin in the public record. GroundTruth's Observed Visits counts a user's visit to the brand or "drive-to" location within 14 days of exposure to a GroundTruth ad, credited by last touch, and shows only visits verified after a 7-day period; its help centre says observed visits "may be lower than actual visits" [6]. Its places are Blueprints, polygons traced from satellite and street-view imagery in in-store, on-lot and retail-block layers; one GroundTruth article of 25 February 2026 counts "4.8M Blueprints" and another of 3 March 2026 "over 5 million" [6, 367] [254, 361]. GroundTruth asserts, without a cited study, that radial geofences measure visits "with up to 75% inaccuracy" [254, 361]. Azira draws polygons by hand from satellite imagery and builds its panel from SDK integrations and bidstream location signals from the bid requests of over 100,000 apps, with no panel size and no SDK-to-bidstream split stated [5]. Under this paper's definitions, location taken from a bid request is inferred location [5]. Adsquare matches lat/long to 2D POI polygons with "no fuzzy radius logic" [448, 21]. Foursquare Attribution documents no visit-definition thresholds [119, 28], and Blis publishes no visit definition and no source for its footfall outcome [124, 229, 19, 519].

Where the basis is documented, the visit totals are scaled. Foursquare weights its panel to census block groups and applies "projection weighting" that scales matched panelists to each publisher's delivered impressions [236, 120]. InMarket's console defines Overall Projected Store Visits as exposed device visits rated up by a device-to-DMA ratio, and Incremental Projected Store Visits as that total multiplied by the LCI Lift; only "Exposed Device Visits" are panel observations [17]. An incremental visit in an LCI report is therefore a projection multiplied by a matched-control lift [17] [17, 11]. GroundTruth's Projected Visits applies the visitation of a per-campaign panel of impressed devices to all impressed devices, appears once a campaign usually reaches 40 Observed Visits and 350,000 impressions, and reports no uncertainty [18]. Azira projects total visits by a Poisson detection model in which each observed device stands for 1/P(Detection) devices, only above about 1,000 observed visits and with the multiplier capped at 20x [5, 447]. Across cohort A, the basis of the visit numbers is not disclosed for Blis, Reveal, Adsquare, Cuebiq, Vistar, Kochava, Lifesight and Precisely [16, 17, 18, 5, 449]. [Chapter 4](#validation) covers how such numbers are validated.

### H3: the default control is modelled

Nine of the fourteen seed products document a general method; five document none [5, 19, 119, 17, 20, 21, 22, 25, 6]. Vistar is a partner wrapper, Kochava lists MediaLift synthetic control as one selectable option, Precisely and Unacast have no named attribution product, and Lifesight's location module is not on its current site [5, 19, 119, 17, 20, 21, 22, 25, 6]. Under the clarified rule, D is 9 and N is 14.

| Product | Vendor's description of the comparator | H3 code | Window | Documented minimums and uncertainty | Visit basis |
|---|---|---|---|---|---|
| Foursquare Attribution | "Modeled control group" on 10+ attributes; 2022 post: regression adjustment to a synthetic control, then difference-in-differences [119, 16] | Matched or synthetic | 14-day example window plus 7 days of reconciliation [8, 5, 6, 17, 10, 118] | Feasibility score, no numeric minimum or significance rule [247, 246] | Projected [16] |
| InMarket LCI (also Campaign Incrementality) | 1:1 audience-matched control on eight traits (undated console); synthetic control from historical visitation (July 2026 post) [17] [121] | Matched or synthetic | Two weeks after versus two weeks before exposure [17, 11] | 10 million or 25 million impressions; standard error; ">95%" confidence (legacy console, read once, now behind a sign-in) [17] | Projected [17] |
| Reveal Net Lift Study (OOH) | Synthetic control by propensity-score matching from a "persistent panel of 60M US DAUs" [20, 518] | Matched or synthetic | Not stated | Confidence intervals and p-values; no level stated as a rule [20, 518] | Not disclosed [16, 17, 18, 5, 449] |
| Adsquare Attribution | Simulated "Synthetic Control Condition" in place of control groups; US not named as a market [448, 21] | Matched or synthetic | Not stated | None stated [448, 21] | Not disclosed [16, 17, 18, 5, 449] |
| Cuebiq measurement | Behaviourally matched control [22]; "causal machine learning to simulate a randomized control trial" [446] | Matched or synthetic | Not stated | None stated [22] | Not disclosed [16, 17, 18, 5, 449] |
| Arrivalist MediaInsights | Matched control of cardholders not exposed [25, 449, 450, 451] | Matched or synthetic | Not stated | None; headline statistics render as zero placeholders [25, 449, 450, 451] | Spend lift declared; arrivals not stated [16, 17, 18, 5, 449] |
| Azira Measurement | Randomised PSA holdout on 10-15% of impressions; synthetic control as the alternative [5] | Randomised or holdout | 7 days (QSR), 14-30 days (auto) [8, 5, 6, 17, 10, 118] | About 1,000 observed visits and 100,000 exposed devices; 99% intervals; two-tailed t-test [5] | Observed and projected [16, 17, 18, 5, 449] |
| Blis Smart Holdout Groups | Machine-balanced pairs of zip codes, one control and one target, with unserved buffer zones [123, 19, 124] | Holdout (assignment not documented as random) | Not stated | None stated [124, 229, 19, 519] | Not disclosed [16, 17, 18, 5, 449] |
| GroundTruth Observed and Projected Visits | No control [6, 122, 257] | Exposed-only, no lift claim | 14 days, last touch, plus 7-day verification [6] | Projected Visits shown from about 40 Observed Visits and 350,000 impressions; no uncertainty [18] | Observed and projected [16, 17, 18, 5, 449] |

Seven of the nine documented methods are non-holdout, so M + E is 7 against a threshold of more than 4.5, and H3 is contradicted [5, 19, 119, 17, 20, 21, 22, 25, 6]. In the strict lift-claim count, which drops GroundTruth because it makes no lift claim, six of eight are matched or synthetic, and H3 is contradicted there too [5, 19, 119, 17, 20, 21, 22, 25, 6]. Adsquare's US availability is unconfirmed; dropping it leaves six non-holdout methods of eight [5, 19, 119, 17, 20, 21, 22, 25, 6]. Blis's pair designation is not documented as random [123, 19, 124], and coding it as a matched geographic design would make the count eight of nine without changing the reading. The rival line does not hold for every product: Azira and Blis document holdouts [5, 19, 119, 17, 20, 21, 22, 25, 6]. Vendor documentation can lag practice, and a randomised design available on request does not count as the default under H3 [5, 19, 119, 17, 20, 21, 22, 25, 6]. No randomised or holdout option appears in Foursquare's opened documentation [119, 16], and every lift design InMarket and its partners describe is matched or modelled [11, 272, 270, 23].

![Seven of the nine documented cohort A visit-measurement methods use no randomised or holdout control; of the eight that claim lift, six use a matched or synthetic control.](figures/fig-control-design.svg)

*Figure fig-control-design. **Seven of the nine documented cohort A visit-measurement methods use no randomised or holdout control; of the eight that claim lift, six use a matched or synthetic control.** The 14 cohort A products in the register, by the control design their documentation gives as the general method for visit measurement, eight of them with a lift claim (final wording of LOC-E-F28, every code re-derived on 24 October 2026). Six use a matched or synthetic control and one reports exposed-only visits; of the two holdout designs, Azira’s randomised PSA control is documented as its default in an undated white paper, and Blis’s geographic holdout is machine-balanced, with no documented random assignment. Five products document no default. Units: products. Scope: cohort A products in the product register, US, at the cutoff. Sources [119], [119], [120], [17], [17], [121], [6], [6], [122], [19], [19], [123], [124], [20], [5], [125], [21], [26], [22], [23], [24], [25], [27]. Caveat: Codes describe documented defaults, not the design of any delivered study; a vendor may run a holdout on request. The count is bounded to the 14-row register. Data: data/figure-data.json#control*

Vendors describe these designs in stronger terms than the designs support. Foursquare's product page calls its output "incremental lift" while its blog and metrics page call it "behavioral lift" [119, 236], a term no opened page defines [119, 248]. InMarket calls its synthetic-control Visit Lift causal [121]. Adsquare says its model "removes reliance on traditional control groups" [21]. Under this paper's codes, language that describes a matched or synthetic design as simulating or replacing a randomised trial does not move it out of the matched class [446].

### The two holdouts

Azira's methodology page (footer 2025) is the only cohort A document opened in stream C that names a randomised control as the default: 10-15% of impressions serve public-service creative as a PSA holdout [5]. Its own Domestos case from Rize, Turkey reports an exposed footfall rate of 1.672% against a control rate of 1.168%, presented as a 43% increase in store traffic, without stating the control design, window or dates [516]. A buyer therefore cannot tell from that published result which of Azira's two designs produced it [516]. The Google-affiliated authors of ghost ads designed the method to measure lift without PSA controls, which they say become invalid under performance-optimised delivery [115, 513].

Blis announced the US rollout of Smart Holdout Groups on 11 July 2024 and the global release on 9 October 2024, and says footfall lift is "Only available in the US" [123, 124, 19]. Areas are scored on seven data points including demographics, household characteristics and income, similar areas are paired with one designated control and one target, and buffer zones built from Blis movement data are left unserved and unmeasured [123, 19, 124]. On its product page, launch releases and technology page, Blis publishes no visit definition, outcome source, minimum sample, significance rule, confidence interval or projection method for it, and states that its measurement "can be independently verified" [124, 229, 19, 519]. The geographic experiment is a published design: Google's 2011 method assigns non-overlapping regions at random to treatment or control [115, 513]. Blis's documents do not say its designation is random [123, 19, 124]. The 55% "store visit uplift (control vs exposed)" in Blis's US restaurant case was published on 12 September 2019, about five years before Smart Holdout Groups were announced [13].

### The comparator decides the number

On one PlaceIQ campaign, changing the comparator multiplies the lift by 10.7. In that undated case for a national big-box retailer, exposed devices visited at a Place Visit Rate of 2.75% against 2.13% for a standing 10% holdout never served a PlaceIQ ad, a 29% lift; against the open population, at 0.67%, the same campaign shows a 310% "open lift" [12]. The document says how the holdout is maintained but not how it is selected, and it does not show whether PlaceIQ, now owned by Precisely, runs the design today [12]. Precisely's audience product page documents no named PlaceIQ attribution product [27].

Published lift figures rarely carry the comparator's rate. Foursquare's linear-TV QSR case reports 1.32% behavioral lift with 3.4M+ incremental and 58.9M total attributed store visits, and its product page reports a "23% lift in in-store visits among exposed users versus the control group", none with exposed and control rates, sample, period or control construction [351, 28]. InMarket reports 121 of 126 TikTok campaigns with "statistically significant lift" (2024; a footnote gives a base of 121) and a 26% incremental visit lift against a 23% benchmark for an unnamed dining chain (2026), with no design, base or uncertainty beyond its own descriptions [11, 121, 267, 375]. Vistar's Foursquare-measured retail case (August 2021 to January 2022) reports a 6.51% lift in visitation "compared with non-exposed consumers" without describing the control [8]. Simpli.fi, which is not a seed product, defines its 31.76% "Geo-Conversion Lift" as the difference between visitors who received an ad and visitors who did not, with no matching described [10, 520]. Cuebiq's benchmarks show how vintage and definition move the norm: its 2017 paper put retail uplift at 10-30% [118], while its 2022 report, on 2020-21 campaigns, put the medium retail band at 1.01-7.61% [517].

In the academic record, observational estimates miss experimental lift by wide margins, and not always in the same direction. On 15 US Facebook campaigns run as randomised trials, matching and regression methods mostly overestimated lift relative to the experiments [111]. Across 663 Facebook experiments, median randomised lifts of 29%, 18% and 5% by funnel stage compared with 83%, 58% and 24% from double machine learning and 173%, 176% and 64% from stratified propensity-score matching [112]. Those are online outcomes on platform data, not store visits [111]. In a randomised Yahoo! experiment with 1.6 million retail customers, a cross-sectional observational estimator showed a negative bias three times the size of the experimental effect, so it understated the effect [114].

### Samples, significance and windows

Azira is the only vendor whose opened public pages disclose numeric minimum-sample and significance rules: about 1,000 observed visits and 100,000 exposed devices, 99% confidence intervals and a two-tailed t-test [17, 5, 8, 518]. InMarket's LCI console states minimums of 10 million impressions for mobile app and 25 million for mobile web or desktop per measured dimension, a standard error and a ">95%+" confidence rule [17], but stream C placed that glossary behind a sign-in and could not verify it from a public route [17, 5, 8, 518]. Reveal discloses confidence intervals and p-values without a numeric minimum, Vistar states only that it requires "enough scale", and Foursquare, Cuebiq, Adsquare, Blis, Simpli.fi and GroundTruth state none on the pages opened [17, 5, 8, 518]. Foursquare releases reports once results are "statistically viable", a criterion it does not define [247, 246]. Scale matters: across 25 randomised retail and brokerage experiments, the median confidence interval on return on investment was over 100 percentage points wide, and informative experiments can need more than 10 million person-weeks [113].

Windows vary as much as controls. Disclosed windows run from 4 days (Vistar with Foursquare, DOOH) through 7 days for QSR and 14-30 days for auto (Azira), 14 days plus a 7-day verification lag (GroundTruth), two weeks before and after exposure (InMarket), to a 30-day look-back (Simpli.fi) [8, 5, 6, 17, 10, 118]. In one 2017 Cuebiq campaign, 37% of exposed visitors arrived within 2.6 days and 14% between 10.5 and 13 days, so the window alone moves a visit count materially [8, 5, 6, 17, 10, 118].

### A formula that does not compute

Azira's white paper defines lift as (exposed rate - control rate) / control rate, then computes incremental visits as "Exposed Group Lift * Exposed Group Devices" [5]. Read literally, that multiplies a relative lift by a device count. Under the paper's own definition, incremental visits are (exposed rate - control rate) x exposed devices [5]. InMarket's console has a related problem: it gives two denominators for LCI Lift on the same page, "Visits" in the glossary and "(Exposed Visits + Control Visits)" in the calculations [17]. A buyer who recomputes a reported incremental figure needs the formula actually applied.

### TV to visit

No opened cohort A documentation names Samba as a source of TV exposure data [22, 119, 274]. The author is employed by Samba, so this finding records only what the measurement vendors' documentation says [22, 119, 274]. Foursquare infers TV exposure by joining viewership data with ad run-logs and names no viewing-data supplier [236]. Cuebiq takes linear TV exposure from broadcast partners' impression logs at household level, without naming them [22]. GroundTruth's help centre (2021, not since updated) says CTV-to-store visits rely on TapAd converting IP addresses to mobile ad IDs, an extra match before the location step [452, 453]. Blis's privacy centre lists Samba TV among its ad-effectiveness partners with Cuebiq, Foursquare, InMarket, Kochava and Lifesight, and no source says which, if any, supplies Smart Holdout Groups' footfall outcomes [342]. Samba TV's consumer site refers business visitors to samba.com, which this review did not open [524, 525, 526]. CTV or TV is a named channel for Foursquare, InMarket, Azira, Blis, Cuebiq and Arrivalist; a listed channel is a claimed capability, not evidence of production measurement in it [28, 19, 5, 25, 27].

### Who pays

GroundTruth is the only product whose opened pages state a payment model: Cost Per Visit, introduced in spring 2017, under which "marketers only pay for ads that result in a store visit"; among the other thirteen products the opened pages state no payer or fee [266]. A November 2024 case reports an auto dealers' association paying "only for verified visits" for "over 1,200 visits" in a month, with no cost, impressions or control [266, 258, 367]. Inference: under Cost Per Visit the seller of the media also counts the exposed-only visits it bills on, and this review found no independent audit of that billing [266, 258, 367]. Vistar's activation guide says its partner studies "require a feasibility test and have varying minimums and costs", and that it offers device ID passback for free [337, 522]. Foursquare documents no pricing [119, 28]. InMarket's visit-lift measurement reaches buyers through partner platforms (TikTok from 2024, Reddit for all US advertisers from 12 November 2024, Vistar for US DOOH, Basis from 27 May 2026) as well as its own console and exports [11, 269, 23, 270, 271, 272]. Its Maestrø beta, announced on 18 June 2026, includes a fully managed "Guaranteed iROAS" offer [273, 523]; inference: a vendor that guarantees a return and measures it sits on both sides of the guarantee [273, 523].

No lift, incrementality or attribution metric of a location product is recorded in the 35 MRC communications from January 2020 to October 2026 screened by name or on the current lists; both MRC location accreditations on record since 2020 covered data sets [166, 168, 521, 171]. MRC's 2017 location guidelines contain no provision on visit lift, incrementality or control groups [445, 454]. Foursquare's Visits data-set accreditation was granted in 2020, suspended in October 2022 and revoked in December 2022, and its own 2020 release said Attribution had not been audited or accredited [166, 167, 245]. [Chapter 8](#standards) has the assurance record.

### What would change the reading

H3 would move if the matched-default vendors documented a randomised or holdout control as their ordinary design, or if Azira and Blis published results that state which design produced them. A buyer can ask for the same things now: the control's construction and visit rate beside the exposed rate, the window, the minimum sample and significance rule, whether the visit totals are observed or projected, and the formula that turns a lift into incremental visits. [Chapter 12](#next) turns these into questions.



## 6. How location audiences are built


*A location audience is a set of devices that a vendor says were at a place, or are at one when the ad is served. Vendors document the shapes they draw and the time windows they use. The documentation opened gives no current source mix, no fix accuracy with a method and no test of whether a segment reaches the people it names. On that record H4, the registered test of bidstream against SDK sourcing, cannot be decided.*

> **In short.**
> Foursquare Audience qualifies devices on past visits, with high, medium and low frequency buckets and a look-back of up to 18 months; Foursquare Proximity is a real-time radius with no look-back [528, 459] [536, 537, 237]. No cohort B documentation opened by 24 October 2026 states a fix-accuracy threshold with a method or a detection error statistic [263, 122, 420, 457, 459]. H4 is unresolved: in the documentation opened, no product documents a primary source; Foursquare and InMarket disclose several sources without naming a primary one, and Azira and GroundTruth disclose blends without proportions [265, 420, 421, 422, 458, 457]. Bidstream appears as a documented source in the FTC's Mobilewalla order, on Azira's data page, in GroundTruth's help centre and on Vistar's partner page for GroundTruth [140] [260, 261, 191] [191, 539, 416, 540, 23]. Sensitive-place handling sits in vendor privacy documents as self-attested policy, alongside FTC orders that impose programmes on InMarket, Mobilewalla, and Kochava with Collective Data Solutions [265, 263, 132, 140, 144, 459]. In the documentation opened, this review found no independent on-target or accuracy validation of a location audience with a published design [263, 459, 422, 268, 529, 456]. T-Mobile's enhanced, location-enriched audience segments, announced on 5 October 2026, are available for out-of-home (OOH) targeting only through Vistar Media, which T-Mobile owns; its app-based segments have been there since May 2025 [334] [36].

### How a device qualifies

The segment rules opened state a place geometry (a polygon or a radius) and a time window, and some state a frequency count; this chapter reads each rule on those terms ([chapter 1](#defining)). Foursquare Audience, live in 2026, builds segments from past visitation with location-specific frequency buckets and a recommended look-back of 3 to 6 months, and is sold off the shelf on The Trade Desk, LiveRamp and DV360 [528, 459]. Its minimum segment size is 5,000 devices in the FAQ and 10,000 on the requirements page, and it is priced at US$1.50 CPM "in most scenarios", against US$0.75 for Proximity [528, 238].

| Product | What qualifies a device | Time window and frequency | Evidence |
|---|---|---|---|
| Foursquare Audience | Past visits to a place | 3 to 6 months recommended, up to 18; high, medium and low frequency buckets | [528, 459] [528, 238] |
| Foursquare Proximity | Inside a radius geofence, validated against the impression's lat/long | Real time, no look-back | [536, 537, 237] |
| GroundTruth Ads Manager | Target Locations (city to business address or lat/long); Target by Audience (historic visitation); Target Neighborhoods (trade-area lookalikes); Target On-Premise (In Store, On Lot, Retail Block, Radial) | On-Premise is real time and mobile only; Target by Audience uses historic visitation | [263, 122] [263] |
| Adsquare ID-Precise Audiences | Mobile advertising ID (MAID) segments from "verified visits, offline purchases"; home and work at postcode level | Look-backs of 14, 30, 90 and up to 180 days | [420, 455] |
| Adsquare Geo-Contextual Audiences | No device ID; a postcode, H3 cell or polygon | Scored by the hour | [420, 455] |
| Simpli.fi Addressable Geo-Fencing | Any device seen within parcel plat lines, up to one million addresses per campaign | Up to 30 days, "variable recency settings" | [456, 457] |
| InMarket GeoLink | Block-level geo-audiences, point-of-interest geofences, past store visitors | Geofences in real time; retargeting on past visits | [276, 40] |

Precision is asserted by geometry, not by a threshold. The one numeric figure found is Azira Audiences' "within 3 meters of precision", with no method. GroundTruth On-Premise is "strictly inside the physical building walls", Simpli.fi uses parcel plat lines, Adsquare offers "precise polygons or custom radii" alongside postcode and H3 cells, and GroundTruth also sells coarse ZIP, state and DMA options [263, 122, 420, 457, 459]. GroundTruth names the tests it applies to each location signal and visit, among them stale, outlier, drive-by speed and open-store hours, and rejects centroids and carrier-IP locations, but publishes no numeric threshold for dwell, horizontal accuracy or speed [357, 259]. Foursquare Proximity is delivered where "your ad platform is able to validate that a user is within the radius boundaries of the geofence by matching the lat/long of the impression" [536, 537, 237]. This paper infers, from that wording alone, that delivery rests on the coordinate carried in the bid request.

GroundTruth and InMarket also sell catalogues built by models. Since ZeroToOne.AI's acquisition, GroundTruth sells "Predictive AI Audiences" from the acquirer's model, built from visitation, interests, transactions, digital engagement and demographics. They are marketed at "over 90%" predictive accuracy with no published definition, and the catalogue went from an "initial" 40 segments in June 2026 to "1,500+" in October 2026 [263, 264, 122]. InMarket sells 10,000+ syndicated segments and GeoType personas built by machine learning on 1,000+ attributes [268, 271]. A Vistar-hosted GroundTruth page states one example segment with both a frequency and a window: "Visitors who visited general clinics at least 1 time in the past 120 days nationwide" [191, 339].

### Where the location comes from: H4

H4 is unresolved: in the documentation opened, no cohort B product documents a primary source of location [265, 420, 421, 422, 458, 457]. H4 says that location-based audience and targeting products sold in the US at the cutoff rest mainly on bidstream or IP-derived location rather than on SDK- or panel-observed location. As registered, it is unresolved if fewer than four cohort B products document a primary source or if the bidstream-or-IP and SDK-or-panel counts are equal, and blends without proportions are excluded from both counts. A clarification dated 24 October 2026, written after extraction in response to external review rounds 2 and 2b, counts only products documenting one of those two families as primary (b for bidstream or IP, s for SDK or panel), makes H4 unresolved if b + s is below four or b equals s, and lists products with other primary sources, blends without proportions or no source separately. It is a substantive change to the population, and H4 is unresolved under both readings ([chapter 13](#method)). A verifier's two SDK-primary codes do not stand after review [265, 420, 421, 422, 458, 457].

| Product | Sources its own documents name | H4 code | Evidence |
|---|---|---|---|
| Foursquare Audience | Its own apps, with opt-in precise or background location; SDK and API partners that "may license data to us"; data suppliers, brokers and ad networks | Several sources, primary not established | [265, 420, 421, 422, 458, 457] [244, 243] |
| InMarket Audiences | Its own apps (CheckPoints, ListEase, Key Ring, Subscription Stopper, Out of Milk), a partner SDK app network, unnamed "trusted & certified third-party sources" | Several sources, primary not established | [265, 420, 421, 422, 458, 457] [267, 134] |
| Azira Audience | "premium SDK, bid-stream, and verified consumer transactions" | Blend, proportions not disclosed | [458, 529] [265, 420, 421, 422, 458, 457] |
| GroundTruth Ads Manager | Privacy policy: SDKs and scripts, publishers, service providers, consumer data resellers. Help centre: WeatherBug, direct app partnerships, SDK integrations and the bid-stream | Blend; one proportion, from 2018 | [265] [260, 261, 191] |
| Blis | Stream F could not read the product pages; the audience page read for the profile names no collection route; the company notice names GPS and IP-derived location from data partners, SSPs, exchanges and clients | Not disclosed for the product | [228] [229] [235] |
| Adsquare, Mobilewalla, Simpli.fi, Reveal, Precisely Audiences | "Vetted partners", third-party ad-tech providers, GPS or "anonymized mobile signals", with no collection route | Route not stated | [420, 455] [422, 141] [456, 457] [530, 531] [421] |
| Broadsign Ads (registered; US availability unconfirmed, so outside the count) | Native Touch data "pulled from active application SDKs and APIs" | SDK via a named partner; outside the count | [538] |

GroundTruth's only proportion, "90% of our location signals" from WeatherBug, app partnerships and SDK integrations, sits in a help article last edited in April 2018 [260, 261, 191]. GroundTruth's privacy policy, effective 2 April 2026, does not use the words bid request or bidstream [265]. The regulator records describe both routes before the orders. The InMarket complaint described collection through InMarket's own apps and through third-party apps using its SDK [132], and the FTC alleged that Mobilewalla collected bidstream data, an allegation rather than a finding [491].

### Bidstream, where it is documented

The FTC's Mobilewalla order (Docket C-4811), issued 13 January 2025 and announced as final on 14 January on a 4-1 vote, bars Mobilewalla from collecting or retaining information it accesses in online advertising auctions for any purpose other than participating in them. It is the only regulator record in stream F that documents bidstream as a cohort B location source [140]. Mobilewalla's business notice, updated 9 January 2026, says all its data comes from third-party ad-tech information providers and names the movement-data fields (MAID, latitude and longitude, IP, timestamp) without saying whether they come from SDKs or bid requests [422, 141]. The same notice says Mobilewalla sold personal information to clients and ad-tech partners in the last year, and its site sells Audience Segments and "Location-Based Marketing" with no stated precision class [142, 141].

Azira's privacy policy, updated 6 August 2025, names bid requests it receives when acting as a DSP, alongside third-party providers, data brokers and app-collected movement data [527]. InMarket's GeoLink applies a machine-learned "Location Score" to the location carried in ad requests before bidding, so InMarket's targeting uses bidstream location as a filtered input alongside SDK data [276, 40]. Blis's August 2021 press kit cited "18 billion" ad requests a day and a "Smart Pin" filter that discards about 80% of raw location data [230, 343]. Simpli.fi says it uses "GPS data paired with plat lines" and rejects IP because of truncation, without saying where the GPS fixes come from [456, 457].

OpenRTB 2.6 lets a request state whether its location came from GPS, IP or the user, with accuracy in metres and the age of the fix [219, 460]. Google's Authorized Buyers documentation, last updated 17 September 2026, says bid-request location comes only from IP geolocation, with lat and lon limited to 0.01 degrees [444]. Thinknear's vendor-run index put 32% to 37% of location-bearing impressions or location data within 100 metres of the user in its 2014 to 2016 editions, and a reseller deck advertising the Q4 2017 index shows 37% on an unlabelled slide ([chapter 4](#validation)) [441, 442, 443] [514, 515].

### Sensitive places: vendor policies and FTC orders

None of the cohort B product pages opened describes sensitive-place suppression, but vendor privacy documents do. Foursquare commits to the Network Advertising Initiative (NAI) enhanced standard and lists 13 sensitive-place categories. Mobilewalla says it does not knowingly collect visits to places of worship, shelters, treatment centres or reproductive health clinics, and Azira describes a programme that removes sensitive locations from its datasets. These are self-attested statements with no verified enforcement [265, 263, 132, 140, 144, 459]. GroundTruth's help-centre guidelines, updated 28 September 2026, define Consumer-Protected places (1-mile minimum; no device-level collection), Healthcare-Protected places (0.4-mile minimum for healthcare ads in WA, NY, CT, NV and CA; visits not tracked) and Prohibited places (no location targeting; visits not measured), each with a published category list [256]. GroundTruth's privacy notice that it "may sell your sensitive personal data" is a state-law disclosure and does not show that it sells sensitive-place visits [265, 263, 132, 140, 144, 459].

InMarket's FTC order, final on 1 May 2024, prohibits it from selling or licensing precise location data, as the order defines it, and, separately, from using, selling, licensing, transferring or sharing any product that categorises or targets consumers on sensitive location data [132]. Its sensitive-location ban covers the locations InMarket's own programme identifies, in nine categories; the order excludes coarse data at a radius of at least 1,850 feet and runs 20 years from issuance [130, 132]. Mobilewalla's order covers ten categories with only a conversion exception and no direct-relationship consent exception [139, 140]. The FTC's 4 May 2026 release on the Kochava settlement says Collective Data Solutions "has taken over Kochava's data broker business" and will be barred from selling sensitive location data without affirmative express consent [147, 144]. On 24 October 2026, Collective Data Solutions' own site offered data feeds including "precision location (protected by Privacy Block)", audience targeting and competitor conquesting [145, 146, 143, 147]. InMarket documents its own sensitive-place handling only as company-level statements, among them suppression of sexual and reproductive healthcare providers and NAI membership [376, 535, 133, 134]. At 30 June 2026 InMarket's privacy policy says it does not sell precise geolocation "in the plain language sense", while its state-law notice says it "sold (or shared for interest-based advertising)" coarse and precise geolocation to third-party partners and uses precise location to build audience segments [134]. This paper does not assess compliance with any order [128, 137, 142, 145, 134, 130]; [chapter 7](#regulation) reads the orders and H5.

State laws change what vendors sell, by the vendors' own account. GroundTruth says that for locations in Maryland, Oregon, Virginia, New Jersey and Connecticut, from dates it gives between 1 October 2025 and 1 October 2026, visit tracking, On-Premise and Neighborhood targeting, radius targeting below 1 mile and audience building from new precise location signals are unavailable [256]. Foursquare requires a separate New York, Washington and Connecticut set with a minimum 3,000-foot radius for national Trade Desk Proximity segments, "to follow state privacy laws" [536, 537, 237]. The statutes read for this paper define precise geolocation by a radius of 1,750 or 1,850 feet [152, 154, 151, 155, 162, 161, 143], so both vendors' minimums are wider than those radii.

### Activation routes and channels

GroundTruth Ads Manager documents six channels: mobile, CTV and OTT, desktop and tablet, audio, digital out-of-home (DOOH) and direct mail. Azira's Advance DSP also documents six, so the two tie for the widest documented channel set among the products opened [122, 459, 530, 529, 335]. Azira, Vistar and Simpli.fi activate through their own platforms, while Foursquare, Adsquare, Reveal and Mobilewalla name external DSPs or marketplaces [122, 459, 530, 529, 335]. Foursquare's eligible-platforms page lists 26 activation platforms for Audience, including Meta, TikTok, Roku, Spotify and Vistar Media, and Vistar says over 2,300 Foursquare audiences are available in its DOOH DSP [237, 238, 416]. InMarket claims activation on "any" DSP and names one external DSP, Basis [268, 271]. Blis activates through its own DSP and, from 17 June 2025, PMG's Alli Marketplace [229, 532, 533]. Outside its own platform, the only named route for GroundTruth audiences is Vistar's programmatic DOOH integration [191, 258, 122, 363]. On The Trade Desk, Vistar's taxonomy is available through auction packages and private marketplaces, and third-party audiences must contain mobile device IDs [337].

The accuracy statements that travel with these segments are vendor assertions without method. They include GroundTruth's "over 90% predictive accuracy", Foursquare's "industry's most accurate location-based datasets", InMarket's 3.7x against Wordstream display benchmarks, and Azira's 80-90% addressability. InMarket also says its data is "Comscore verified and Location Sciences certified" but publishes no certificate or scope [263, 459, 422, 268, 529, 456]. Location Sciences audited several InMarket audiences in the first half of 2020 and rated them "Optimal"; the certifier's announcement does not say who paid for those audits, and no later audit or renewal was found [378, 373, 267]. In March 2021 the same certifier stated that Blis had threatened legal action after its Verify product reported "data manipulation"; Adweek reported that Blis denied the allegations [343, 231, 232, 233]. [Chapter 11](#gq) scores cohort B on this record.

### Geofencing for local advertisers

Geofencing and geo-conquesting reach US local advertisers through at least three routes with different mechanics [263, 122, 456, 530]. GroundTruth Ads Manager sells real-time On-Premise and "Pure Geographic" delivery to anyone inside a boundary, plus retargeting of a competitor's past visitors, with no minimum spend [263, 122, 456, 530] [122, 534]. Simpli.fi's Addressable Geo-Fencing turns up to one million street addresses into parcel-level fences drawn from property-tax and land-survey records, then follows "any device seen within those plat lines" with mobile, video and CTV ads for up to 30 days through a cross-device graph [456, 457]. Reveal's Geofencing Solutions captures MAIDs at "your locations, competitor locations, and audience hotspots" for The Trade Desk, LiveRamp, Basis and social platforms, and names no source of device location [530, 531]. In Simpli.fi's case studies a visit counts when a served device later enters a Conversion Zone around the store, an exposed-only design, and "Geo-Conversion Lift" compares visitors who received an ad with visitors who did not, with no matching described [10, 520]. Under this paper's rules that is a visitation comparison, not incremental lift ([chapter 5](#measurement)).

### Telco data enters DOOH targeting

Vistar's own guide documents DOOH audience targeting from "aggregated and anonymized location data" plus third-party partners: Epsilon, Eyeota, InMarket and Affinity Answers [338, 335]. T-Mobile's app-ownership and engagement segments have been on Vistar's self-service DSP since 6 May 2025 [334]. On 5 October 2026, Vistar announced enhanced T-Mobile segments that add "location insights", available only through the Vistar DSP; T-Mobile turns them into area-level audience scores that Vistar matches to screens [334]. Vistar's older T-Mobile partner page describes the segments as "built from app ownership data" from devices on T-Mobile's network [492, 334]. AdExchanger, relaying T-Mobile's account, reported that its data for Vistar also includes mobile location data used to estimate inventory impressions, and for users who opted into location tracking T-Mobile can detect proximity to OOH ads and retarget on other channels [335]. T-Mobile's Jess Zhu attributed the exclusivity to tightening regulation of location data and mobile device IDs [338, 335]. The same carrier owns Blis, whose DSP has carried T-Mobile US app data since 10 September 2025 [227, 228] [36, 228].

Vistar's wording and its legal notice point in different directions. The audience guide says DOOH targeting works "without ever relying on personal information", while the Services Privacy Notice says Vistar receives device location histories tied to device identifiers and sells and shares personal information [338, 339]. On an undated page, Broadsign Ads names its mobility supplier and route: Native Touch data from "active application SDKs and APIs", collected once phones enter defined boundaries and turned into six targeting groups and over 190 personas per screen, with no panel size, consent route or region [538]. Broadsign's measurement page, by contrast, shows its partners as logos and documents no provider [194]. GroundTruth's DOOH product, launched through Place Exchange on 9 October 2025, captures mobile device IDs within viewing range of a screen when an ad plays and reuses them for retargeting [257, 254]. [Chapter 9](#dooh) sets out the DOOH sources in full.

A buyer can ask each seller for the segment rule (geometry, dwell, window and frequency), the source mix behind it and any on-target test ([chapter 12](#next)).



## 7. What the law now forbids


*Five FTC orders now bind the US location data companies they name, and only one, InMarket's, bans selling precise location data as such. The other four bar selling or using sensitive location data, and the X-Mode and Gravy orders do not ban selling precise location as such. Five states have since banned the sale of precise geolocation itself: Maryland, Oregon, New Jersey (from 30 June 2026), Virginia and, from 1 October 2026, Connecticut. At the cutoff three of the five ordered companies still document current offers of precise location to third parties, so H5, the regulatory-withdrawal hypothesis, is contradicted. This review reads product sets, not compliance, and nothing in it finds that any company breached its order.*

> **In short.**
> The FTC's orders against X-Mode/Outlogic and InMarket (April 2024) and against Gravy Analytics with Venntel and Mobilewalla (January 2025) run 20 years; the Kochava order, entered by a federal court on 25 June 2026 after almost four years of litigation, runs ten [126, 127] [130, 131, 132] [135, 136] [139, 140] [542, 543, 544, 148, 143, 144] [143]. InMarket's order bans selling or licensing precise location data, as it defines it; the X-Mode and Gravy orders do not ban the sale of precise location, and the Mobilewalla and Kochava terms reach sensitive location data [130, 131, 132] [126, 127] [135, 136] [139, 140] [147, 144]. Gravy told Norway's regulator that an intruder, who contacted it on 4 January 2025, obtained "some files"; the "millions of smartphones" figure is the intruder's claim [605] [545]. Maryland (1 October 2025), Oregon (1 January 2026), New Jersey (30 June 2026), Virginia (1 July 2026) and Connecticut (1 October 2026) ban the sale of precise geolocation defined by a 1,750-foot radius; Washington bans geofences around in-person health care, and California defines precise geolocation by 1,850 feet [152] [154, 153] [156] [162, 157] [155, 159] [151] [161, 163]. In the searches logged, the one state attorney-general lawsuit against an alleged location data broker found for 2023 to 2026 is Texas's against Allstate and Arity [555]. Under the registered rule H5 is contradicted: Outlogic, Collective Data Solutions for Kochava and Gravy trading as Unacast document precise-location offers, Mobilewalla does not state the precision of what it licenses, and InMarket documents no such product [128, 137, 142, 145, 134, 130].

### Five orders, one sale ban

Only the InMarket order bans selling precise location data; the other four reach sensitive location data [130, 131, 132] [126, 127] [135, 136] [139, 140] [147, 144]. The table reads each from the FTC's own record or, for Kochava, the court order.

| Order | Dates | What it bans | Exception | Clean-up | Term |
|---|---|---|---|---|---|
| X-Mode Social and Outlogic (C-4802) | Proposed January 2024; issued 11 April 2024; final 3-0-2, announced 12 April | Selling, licensing, transferring, sharing, disclosing or using sensitive location data (seven categories of Sensitive Locations); not precise location as such | Direct-relationship consent | Historic Location Data deleted in 60 days unless consented, or deidentified or rendered non-sensitive within 90; Data Products deleted in 90 | 20 years [126, 127] [126] |
| InMarket (C-4803) | Issued 29 April 2024; final 3-0-2, announced 1 May 2024 | Selling or licensing precise Location Data for any valuable consideration; products that categorise or target consumers on sensitive location data (nine categories) | Carve-out for compliance use | Own-app Historic Location Data deleted in 90 days unless the consumer consents to retention; third-party data deleted, deidentified or rendered non-sensitive in 120; segments built on it deleted in 120 | 20 years [130, 132] [130, 131, 132] [130] |
| Gravy Analytics and Venntel (C-4810) | Complaint and proposed order 3 December 2024; issued 13 January 2025; final 5-0, announced 14 January | Selling, disclosing or using sensitive location data (eight categories, adding military installations); not precise location sold to commercial customers | Direct-relationship consent; security, federal national-security and federal law-enforcement emergency uses fall outside Location Data | Historic Location Data deleted in 60 days unless, within 90, consent records are obtained or the data deidentified or rendered non-sensitive | 20 years [135, 136] [135, 428, 136] |
| Mobilewalla (C-4811) | Issued 13 January 2025; final 4-1, announced 14 January, Commissioner Holyoak dissenting | Using or selling sensitive location data (ten categories); collecting or retaining ad-auction data for any purpose other than taking part in the auction | Conversion to non-sensitive or non-location data only; no consent exception | Historic Location Data deleted in 90 days; Data Products in 120 | 20 years [139, 140] [140] |
| Kochava and Collective Data Solutions (D. Idaho, 2:22-cv-00377) | Announced 4 May 2026 (Commission vote 2-0); entered 25 June 2026 | Selling, sharing or disclosing sensitive location data without affirmative express consent, in the FTC's description | Consent limited to data used for a service the consumer directly requested | Historical Location Data deidentified or rendered non-sensitive in 90 days unless consent records are obtained | 10 years [542, 543, 544, 148, 143, 144] [147, 144] [143] |

![Only InMarket’s order bans selling location data outright, and three of the five ordered companies still document selling precise geolocation.](figures/fig-ftc-orders.svg)

*Figure fig-ftc-orders. **Only InMarket’s order bans selling location data outright, and three of the five ordered companies still document selling precise geolocation.** The five FTC location orders of 2024–2026 and the GM/OnStar order of January 2026, from the orders and the FTC’s releases, with each ordered company’s documented product set at the cutoff coded against H5 (final wording of LOC-H-F23). Outlogic, Collective Data Solutions for Kochava, and Gravy Analytics trading as Unacast document the sale or licensing of precise geolocation to third parties; Mobilewalla documents licensing device-linked latitude and longitude without stating its precision; InMarket documents no such product. Under the preregistered rule, H5 is contradicted. Units: order terms; categories; feet; years. Scope: FTC orders on location data, January 2024 to the cutoff, US. Sources [126], [127], [128], [129], [130], [131], [132], [133], [134], [135], [136], [125], [137], [138], [139], [140], [141], [142], [143], [144], [145], [146], [147], [148], [149], [150]. Caveat: Sensitive-location category counts are the numbered items in each order’s definition and are not comparable across orders as a measure of strictness. Status-sensitive: product pages read in October 2026. Data: data/figure-data.json#ftc*

None of the orders defines "precise" by a radius. All five instead exclude coarse data, giving zip code or census block location with a radius of at least 1,850 feet as the example [152, 154, 151, 155, 162, 161, 143]. The Kochava order also leaves out of Precise Location Data any data used for security purposes, federal national security, or a law-enforcement response to an imminent risk of death or serious bodily harm [143]. The Gravy order puts its national-security carve-out in the definition of Location Data instead [135, 136].

The orders reach supply as well as sales. X-Mode/Outlogic and Gravy must run a Supplier Assessment Program, and from 90 days after the effective date neither may collect or use Location Data without a record of the consumer's consent [126] [135, 428, 136]. InMarket must run a Sensitive Location Data Program and an SDK Supplier Assessment Program that confirms consent for SDK-supplied data [130]. Mobilewalla's case was the first in which the FTC alleged that collecting consumer data from real-time bidding exchanges was unfair; that documents an allegation that bidstream was among Mobilewalla's sources, not an admission or a finding [491]. Gravy and Venntel neither admit nor deny the allegations [135, 428, 136].

### Kochava fought for almost four years

Kochava's order came after almost four years of federal litigation [542, 543, 544, 148, 143, 144]. The FTC sued in the District of Idaho on 29 August 2022, and the court granted Kochava's first motion to dismiss, with leave to amend, on 4 May 2023 [542, 543, 544, 148, 143, 144]. The complaint described a data sample covering more than 61 million unique mobile devices in one week [541, 144]. The First Amended Complaint followed on 5 June 2023, and the court denied the motion to dismiss it on 3 February 2024 [542, 543, 544, 148, 143, 144]. Kochava then moved its data-broker business into a subsidiary, Collective Data Solutions (CDS), and a Second Amended Complaint naming CDS was filed on 15 July 2024. The court denied the motion to dismiss that one on 3 February 2025, and the stipulated order was announced on 4 May 2026 and entered on 25 June 2026 [542, 543, 544, 148, 143, 144]. The former kochava.com/collective address now serves an app product page with no Collective content [147, 144].

### Connected cars sit outside H5

On 14 January 2026 the FTC announced a final order against General Motors and OnStar, proposed on 16 January 2025, that bans GM for five years from disclosing consumers' geolocation and driver-behaviour data to consumer reporting agencies [149]. It is a connected-vehicle case and lies outside the five companies H5 covers [149].

### What Gravy confirmed about its breach

The Gravy incident is a security incident, not an enforcement action [605]. In an undated notice to Norway's Datatilsynet, Gravy Analytics, a Unacast subsidiary, said the intruder told it on 4 January 2025 of unauthorised access to its AWS cloud storage, apparently through a misappropriated access key [605]. It said the intruder obtained some files whose contents and personal-data status were under investigation, that it was still assessing whether the incident was a reportable personal data breach, and that its data-processing platform was operational again by 9 January 2025 [605]. Unacast's Texas registration, submitted on 9 January 2026, adds that a small subset of copied data, "covering primarily a few days around New Year's 2025", was briefly posted on a dark web forum [333, 278].

The widely cited scale, "several terabytes" and "historical location data of millions of smartphones", is the hacker's claim on a closed-access Russian-language forum, first reported by 404 Media [545]. In the sources read, the only other figure is security researcher Baptiste Robert's statement that his copy of the leaked dataset held more than 30 million location data points, by a method he did not publish [545]. 404 Media found much of the location data was inferred from IP addresses [545].

### Five states now ban the sale

Five states ban the sale of precise geolocation, New Jersey among them since 30 June 2026 [156]; Washington bans geofences around health care; California defines the term and runs a deletion system [152] [154, 153] [162, 157] [155, 159] [151] [161, 163].

| State | Instrument | What it bars | Radius | Effective |
|---|---|---|---|---|
| Maryland | Online Data Privacy Act (2024 ch. 455, SB 541) | Sale of sensitive data, which includes precise geolocation, with no consent exception; geofences within 1,750 feet of mental health or reproductive or sexual health facilities | 1,750 ft | 1 October 2025 [152] |
| Oregon | Oregon Laws 2025 ch. 251 (HB 2008), amending ORS 646A.578 | Sale of personal data that identifies a consumer's or linked device's present or past location, or of data on a consumer known to be under 16 | 1,750 ft | 1 January 2026 [154, 153] |
| New Jersey | P.L.2026, c.25 (A5328 1R), amending P.L.2023, c.266 | Sale of sensitive data, including precise geolocation, by any controller regardless of consumer count; sale or licensing of sensitive data by data brokers and data collectors, with ten listed exemptions | 1,750 ft | 30 June 2026, immediately; the public registry follows after 270 days [156] |
| Virginia | Code 59.1-578(A)(6), added by 2026 Acts ch. 820 (SB 338) | Selling or offering for sale precise geolocation data, with no consent exception | 1,750 ft | 1 July 2026 [162, 157] [152, 154, 151, 155, 162, 161, 143] |
| Connecticut | Public Act 26-64 (SB 4) | Sale of any consumer's precise geolocation data by a controller or third party, replacing the consent-to-sell rule in force from 1 July 2026 | 1,750 ft | 1 October 2026 [155, 159] |
| Washington | My Health My Data Act (2023 ch. 191, ESHB 1155) | Geofences around in-person health care entities to identify or track consumers, collect consumer health data or send messages | 2,000 ft geofence; 1,750 ft precise location | Geofence ban 23 July 2023; health-data duties 31 March 2024 [151] |
| California | Civil Code 1798.140(w) and (ae); Delete Act | Precise geolocation is sensitive personal information; data brokers register each January and process DROP deletion requests | 1,850 ft | DROP from 1 August 2026 [161, 163] |

![5 states ban the sale of precise geolocation, all from dates between October 2025 and October 2026, each at 1,750 feet.](figures/fig-state-laws.svg)

*Figure fig-state-laws. **5 states ban the sale of precise geolocation, all from dates between October 2025 and October 2026, each at 1,750 feet.** US state rules on precise geolocation opened in stream H, with the corrections of its verifier and the review rounds applied: Virginia’s Code 59.1-578(A)(6) (effective 1 July 2026) and Connecticut’s Public Act 26-64 (effective 1 October 2026) ban the sale of precise geolocation, joining Maryland (1 October 2025), Oregon (1 January 2026) and New Jersey (P.L.2026, c.25, 30 June 2026). Washington’s My Health My Data Act bans geofences of 2,000 feet or less around health care entities and has done so since 23 July 2023. Units: effective dates; feet. Scope: state statutes opened by the cutoff, US. Sources [151], [152], [153], [154], [155], [156], [157], [158], [159], [160], [161]. Caveat: Only the states listed were opened; other states with precise-geolocation definitions were not swept. Virginia’s effective date rests on the legislature’s bill history; each act’s coverage thresholds and exemptions apply. Data: data/figure-data.json#states*

Maryland's ban is subject to the act's coverage thresholds and exemptions, and a clause that limits one section to processing from 1 April 2026 does not postpone the sale ban [152]. Virginia still requires consent for other processing of precise geolocation [162, 157]. Washington treats precise location that could indicate an attempt to obtain health services as consumer health data [151].

GroundTruth says that for locations in Maryland, Oregon, Virginia, New Jersey and Connecticut, visit tracking, radius targeting below one mile and audience building from new precise signals are unavailable, and attributes this to state precise-geolocation laws [256]. For national segments on The Trade Desk, Foursquare requires a separate New York, Washington and Connecticut set with a minimum 3,000-foot radius "to follow state privacy laws" [536, 537, 237]. Placer names Washington's act and Nevada's SB 370 as reasons it withholds reports on in-person health care locations [286]. All three are vendor assertions [256] [536, 537, 237] [286].

### What state attorneys general have done

Texas has brought the only state attorney-general lawsuit against an alleged location data broker found for 2023 to 2026, in the researcher's 14 searches and the verifier's 8 [555]. On 13 January 2025 the Attorney General announced a suit in Montgomery County against Allstate, two Allstate insurers and three Arity entities, under the Texas Data Privacy and Security Act, the Texas Data Broker Law and the Insurance Code [546, 547]. The state alleges that Arity paid app developers to embed its SDK, naming Life360, and collected location and driving-behaviour data from over 45 million consumers without notice or consent; it calls the suit the first enforcement by a state attorney general under a comprehensive privacy law [546, 547]. The petition pleads a Data Broker Law count against Arity, and the case remains in litigation [555]. Arity sells an advertising product, Arity Audiences, and on 16 October 2026 announced an agreement to buy substantially all of INRIX's assets [493, 494]. In those searches no state attorney-general lawsuit against an advertising-technology company over location data was found [555].

The other state actions found concern a carmaker, Google and one advertiser [555]. Texas sued General Motors and OnStar on 13 August 2024 over driving data that included GPS coordinates, and Arkansas, Nebraska and Iowa followed between February 2025 and February 2026; all four suits are allegations [548, 549, 550] [551, 552, 553]. California's Attorney General and four district attorneys settled with GM for US$12.75 million over sales of location and driving data to Verisk and LexisNexis Risk Solutions, with final judgment entered on 12 May 2026 [554]. State attorneys general have settled with Google over deceptive location tracking at least six times between October 2022 and October 2025, from Indiana's US$20 million to Texas's US$1.375 billion, which also resolves incognito-browsing and biometric claims [556, 557, 558, 559]. In 2017 Massachusetts bound Copley Advertising not to geofence at or near Massachusetts health care facilities to infer health status, after Copley had geofenced reproductive-health and methadone clinics in other states [560]. California's March 2025 sweep of the location data industry is an investigation, not a lawsuit [555].

### What the registries show

The data broker registries record what companies declare [251, 163]. In California's file of 603 registrants that operated as data brokers in 2025, 115 self-report collecting precise geolocation, among them Collective Data Solutions, Outlogic, InMarket, Mobilewalla, Unacast and Venntel; Kochava Inc. is not listed [251, 163]. Unacast and Venntel also answer yes to past-year sale or sharing with the federal government and other state governments [251]. On 24 October 2026 the Texas register's 456 records included Outlogic, Collective Data Solutions, Unacast and Venntel but not Mobilewalla, InMarket or Kochava [561, 562, 563]. Vermont's revised statute requires disclosure of precise-geolocation collection from 1 January 2027, and Connecticut requires data brokers to register from the same date [561, 562, 563] [155, 159]. A registration is a filing about collection in a past year, not evidence of what a company sold at the cutoff or an assurance of compliance [251, 163] [561, 562, 563].

### H5 is contradicted

H5 predicted that the five ordered companies would no longer sell precise-location or sensitive-location data products in the US at the cutoff. A clarification dated 24 October 2026, added after extraction following external review, sets precedence: one verified current US offer to sell or license precise location data to third parties contradicts H5, a current public offer is the test rather than proof of a sale, and a statutory notice of sales in the past twelve months is not on its own a current offer, and continued licensing is not a finding of an order breach [128, 137, 142, 145, 134, 130]. The verdict does not depend on the clarification, because the original contradicted branch is met [128, 137, 142, 145, 134, 130].

| Company at the cutoff | What its own pages document | Reading |
|---|---|---|
| Outlogic (X-Mode) | It may "receive and sell the precise latitude and longitude coordinates of your mobile device"; data to license through data exchanges, platform partners or direct, via Snowflake, AWS S3 and a REST API; audiences through Eyeota and Adsquare; filtering of sensitive health-based locations in specific states | Precise offer [128, 129] |
| Collective Data Solutions (Kochava) | "Data Feeds: License premium data including ... precision location (protected by Privacy Block)", with Privacy Block marketed as blocking health clinics, religious institutions and schools | Precise offer [145, 146, 143, 147] [128, 137, 142, 145, 134, 130] |
| Gravy Analytics dba Unacast (with Venntel) | Location Data, Foot Traffic Data and MAID-to-hashed-email linkages; device-level location data for the US and abroad; Venntel sells location products to national-security buyers | Precise offer [137, 125, 138] [328, 332] |
| Mobilewalla | Movement data with MAID, latitude and longitude, IP and timestamp; personal information sold to clients and ad-tech partners in the past year; no precision class for what is sold | Unclear [142, 141] [128, 137, 142, 145, 134, 130] |
| InMarket | Does "not sell (in the plain language sense of that word) or license to customers" precise geolocation, while its state-law notice lists precise geolocation "sold (or shared for interest-based advertising)" to Third Party Partners | No such product [133, 134] |

Unacast's offer survived a split. On 1 September 2026 Pine59 announced it had taken over Unacast's former Aggregated Location Insights business in a management buyout, yet Unacast's own pages still market device-level location data on 24 October 2026 [326, 413, 325] [328, 332]. InMarket's reading rests on its home page and privacy policy, the only InMarket pages read for H5 [133, 134].

The contradiction is about product sets, not breaches [128, 137, 142, 145, 134, 130]. Because the X-Mode and Gravy orders do not ban selling precise location outside sensitive places, Outlogic's and Unacast's offers are not on their face outside those orders, and compliance with any order is not assessed [126, 127] [135, 136] [128, 137, 142, 145, 134, 130]. Each of the companies that sells documents some sensitive-location suppression, from Outlogic's state filtering to CDS's Privacy Block and Unacast's PrivacyCheck, and each is the company's own statement [128, 129] [145, 146, 143, 147] [327, 332, 135]. Outlogic and InMarket also appear on the signatory page, last updated March 2025, of the NAI's Precise Location Enhanced Standards, a self-regulatory code [564, 565, 243]. [Chapter 6](#audiences) reads the audience products these companies sell, and [chapter 11](#gq) scores the regulatory record under the privacy-posture dimension.

### The instruments read

The instrument lists the state laws and registries read for this chapter. Each row describes the primary text as read; none is legal advice.

**Rules and governance instruments**

| Jurisdiction | Instrument | What it requires | Effective | Scope | Law or practice | Src |
|---|---|---|---|---|---|---|
| Maryland | Online Data Privacy Act of 2024 (Ch. 455, SB 541) | Controller may not sell sensitive data, which includes precise geolocation data; no geofence within 1,750 ft of mental health or reproductive or sexual health facilities for consumer health data | 2025-10-01 | controllers subject to the act | sale ban + geofencing ban | [152] |
| Oregon | Oregon Laws 2025 ch. 251 (HB 2008), amending ORS 646A.578 | Controller may not sell personal data accurately identifying a consumer's or linked device's present or past location within 1,750 ft, or personal data of a consumer under 16 | 2026-01-01 | controllers subject to ORS 646A.570-589 | sale ban | [153, 154] |
| Washington | My Health My Data Act (2023 c 191, ESHB 1155), sec. 10 | Unlawful to implement a geofence (2,000 ft or less from perimeter) around an in-person health care entity to identify or track consumers, collect consumer health data or send them messages | 2023-07-23 | any person | geofencing ban | [151] |
| Washington | My Health My Data Act (2023 c 191), secs. 3-9 | Consumer health data (including precise location information, 1,750 ft, that could indicate an attempt to obtain health services) collected or shared only with consent or as necessary for a requested service | 2024-03-31 (small businesses 2024-06-30) | regulated entities and small businesses | consent rule | [151] |
| Connecticut | Public Act 25-113 (SB 1295), sec. 9 amending CGS 42-520 | Controller shall not sell the sensitive data of a consumer (includes precise geolocation data) without the consumer's consent (superseded for sale by Public Act 26-64 from 1 October 2026) | 2026-07-01 | controllers subject to CTDPA | consent rule | [155] |
| Connecticut | Public Act 25-113 (SB 1295), sec. 15 amending CGS 42-529a | No collection of a known minor's precise geolocation unless strictly necessary, with a signal shown for the duration | 2026-07-01 | controllers offering online services to minors | collection limit | [155] |
| Virginia | Consumer Data Protection Act, Code 59.1-575 (definitions) | Precise geolocation data (1,750 ft) is sensitive data; the sale ban is in 59.1-578(A)(6), listed separately | 2023-01-01 (not read here) | controllers subject to VCDPA | consent rule (by definition only) | [162] |
| California | CCPA/CPRA, Civil Code 1798.140(w) and (ae) | Precise geolocation (radius 1,850 ft) is sensitive personal information subject to the right to limit use; no sale ban | current at access | businesses subject to CCPA | definition + limit-use right | [161] |
| California | Delete Act data broker registry and DROP | Data brokers register January 1-31 each year; must process DROP deletion requests from 2026-08-01 | 2026-08-01 (DROP processing) | registered data brokers | registry + deletion mechanism | [163] |
| New Jersey | P.L.2026, c.25 (A5328 1R), amending P.L.2023, c.266 | No controller may sell sensitive data, including precise geolocation, regardless of consumer count; no data broker or data collector may sell or license sensitive data (ten listed exemptions) | 2026-06-30 (immediately; public registry provision after 270 days) | Controllers, data brokers and data collectors; precise geolocation within 1,750 feet | sale ban | [156] |
| Virginia | Code 59.1-578(A)(6), added by 2026 Acts ch. 820 (SB 338) | A controller shall not sell or offer for sale precise geolocation data concerning a consumer | 2026-07-01 | controllers subject to VCDPA | sale ban | [157] |
| Connecticut | Public Act 26-64 (SB 4), secs. 14 and 15 amending CGS 42-520 and 42-521 | No controller and no third party shall sell any consumer's precise geolocation data (communications content and utility smart-meter data excepted) | 2026-10-01 | controllers and third parties subject to CTDPA | sale ban | [159] |



## 8. Who checks the numbers


*On 24 October 2026 none of the eleven accredited and under-review lists of the Media Rating Council (MRC) carried a US visit or visit-lift metric, so H6 is contradicted on that reading. Two location data sets held MRC accreditation from 2020, Foursquare's and GroundTruth's, and both lost it. MRC's only location standard dates from March 2017 and says nothing about lift. The rest of the assurance on offer is self-regulation, specification text and vendor statements, and Geopath, which publishes the US out-of-home audience currency, is on no MRC list opened.*

> **In short.**
> MRC's only location-specific standard is the Location-Based Advertising Measurement Guidelines, final Version 1.0 of March 2017, with no later update on MRC's index on 24 October 2026 [445, 600, 601]. The text contains no provision on visit lift, incrementality or control groups [445, 454]. MRC's September 2018 presentation reported one accredited location service among six applicants, unnamed [166, 167, 521, 569, 164]. Foursquare's Visits data set was accredited in July 2020 and revoked in December 2022 [166, 167, 521, 569, 164]. GroundTruth's data set was accredited in July 2021 and later revoked after it withdrew, in a month the records leave unconfirmed [168, 169, 170, 356]. Both covered data sets, not a lift or attribution metric [166, 168, 521, 171]. On 24 October 2026 no visit, visitation, foot-traffic or location-determination metric was on any of MRC's six accredited or five under-review lists, so under the registered rule H6 is contradicted on that reading [171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 567, 568]. The Network Advertising Initiative's (NAI) precise-location enhanced standards had four signatories on a page last updated March 2025: Cuebiq, Foursquare, InMarket and Outlogic [564, 565, 243]. InMarket and Outlogic are under FTC orders [130, 132] [126, 127]. OpenRTB 2.6 sets no rule on truncating coordinates, so exchanges and publisher configuration set precision [219, 460] [444] [437, 438, 439]. Geopath appears on no MRC list opened, and MRC's out-of-home accredited list is empty [173, 178, 171].

### What the 2017 guidelines require

MRC issued the Location-Based Advertising Measurement Guidelines in draft for 30 days of public comment in November 2016 and finalised them on 29 March 2017, sponsored with IAB and the Mobile Marketing Association (MMA). MRC's standards index on 24 October 2026 lists them as issued March 2017 with no later update [445, 600, 601]. They cover how device and user location is assigned for physical visitation and ad exposure on mobile and static devices, including desktop and connected TV, and treat place measurement as the reference for visitation and targeting. They are guidance and a benchmark for voluntary third-party audit, not a certification [445].

The text separates accuracy, closeness to the true location, from precision, the number of decimal places, and requires both to be disclosed. A geofence is either a point radius or a boundary polygon. A physical visit is the intersection of device location and place, with speed or longitudinal analysis to exclude drive-bys, and the guidelines note that geofencing generally cannot represent multi-storey buildings [445]. Two decimal places of latitude and longitude place a device within about one kilometre, four within about ten metres and five within about one metre. The text warns that decimal count must not be equated with accuracy, encourages first-party ground-truth validation and sets a general expectation of less than 5% error on primary reported characteristics, with disclosure above that [445].

On sources, the guidelines list GPS, Wi-Fi, IP, cell trilateration, registration data, beacons, panels and ad or bid-request location, and characterise mobile IP as generally neither accurate nor precise. Bid-request location should pass validation edit rules, and zip, DMA or state centroids should not stand for precise location unless reported at market level. Vendors using ad or bid-request data should disclose the active devices they observe daily, weekly or monthly and the number of measured visits [445]. Projections need a fully disclosed universe, and mixed assignment methods should be disclosed by share, along with whether assignment is probabilistic or deterministic [445]. The guidelines also require clear opt-in for location collection and a privacy policy stating why location is collected and how it is shared [445].

The guidelines recommend annual third-party certification of every digital location-based measurement product used in buying and selling, without requiring it. Conformance to them ("built to MRC guidelines") and MRC accreditation are therefore distinct states, and only accreditation appears on MRC's lists [445, 566].

### What they leave out

The 2017 text has no occurrence of "lift", "control group" or "incremental". MRC treats foot traffic as an outcome, and incrementality including randomised controlled trials, in its separate Outcomes and Data Quality Standards of September 2022, which require foot-traffic outcomes to follow the 2017 location guidelines [445, 454]. The location standard governs how a visit is counted, and whether advertising caused it falls under the general outcomes standard [445, 454]. In the 35 MRC communications from January 2020 to October 2026 screened by name for nine location vendors, and on the current lists, no lift, incrementality or attribution metric of a location product is recorded [166, 168, 521, 171]. Of the nine cohort A products that document a general visit-measurement method, seven use no holdout; of the eight that claim lift, six use a matched or synthetic control ([chapter 5](#measurement)) [5, 19, 119, 17, 20, 21, 22, 25, 6].

### The accreditation record

MRC's own 2017 text noted that some US location products had then been "accredited by the MRC in some fashion", without naming them [445, 566]. Its September 2018 presentation reported one accredited location service among six applicants, also unnamed [166, 167, 521, 569, 164]. The records screened name three location services, and the table adds Epsilon's outcomes entry, which is not a visit metric [165, 171] [166, 167, 521, 569, 164] [168, 169, 170, 356] [570].

| Service | Scope in MRC's record | Granted | Ended | Evidence |
|---|---|---|---|---|
| Unnamed location service | One of six applicants accredited | By September 2018 | Not stated | [166, 167, 521, 569, 164] |
| Digital Element NetAcuity (Digital Envoy) | IP geolocation | Not in the records screened | Moved from hiatus to non-accredited, January 2020 | [165, 171] |
| Foursquare (formerly Placed) Visits data set | Location information reported in Visits data set | July 2020 | Suspended October 2022; revoked December 2022 | [166, 167, 521, 569, 164] [166, 167, 245] |
| GroundTruth Location, Place, and Visit data set | Data set | July 2021 | Hiatus at its request, September 2023; revoked after withdrawal, reported 26 July 2024, month unconfirmed | [168, 169, 170, 356] [169, 170, 368, 369] |
| Epsilon Correlated Outcomes (online and offline) | Desktop, mobile web, mobile in-app; not a visit metric | Continued 1 October 2025 | Current | [570] |

![No visit or location metric is on any of MRC’s eleven lists; both visit data sets MRC accredited, in 2020 and 2021, have since lost accreditation.](figures/fig-mrc-history.svg)

*Figure fig-mrc-history. **No visit or location metric is on any of MRC’s eleven lists; both visit data sets MRC accredited, in 2020 and 2021, have since lost accreditation.** Every MRC action on a location, place, visit or IP-geolocation metric found in MRC’s communications from January 2020, plus MRC’s September 2018 presentation, and the state of its lists on 24 October 2026. Foursquare’s Visits data set was accredited in July 2020, suspended in October 2022 and revoked in December 2022; GroundTruth’s Location, Place, and Visit data set was accredited in July 2021, went on a six-month hiatus at its request in September 2023 and was reported revoked in July 2024. Foursquare said its Attribution product was not audited. Units: dated MRC actions. Scope: MRC accredited-services and services-under-review lists and communications, US. Sources [164], [165], [166], [167], [168], [169], [170], [171], [172], [173], [174], [175], [176], [177], [178], [179], [180], [181]. Caveat: Communications before January 2020 were not screened apart from the September 2018 presentation (LOC-NFA-S01), which names no service. Status-sensitive: lists read on 24 October 2026. Data: data/figure-data.json#mrc*

MRC granted Foursquare's accreditation in July 2020, suspended it in October 2022 and revoked it in December 2022, per its communications of 13 October 2020 and 24 January 2023. Because of the 2018 slide, the 2020 grant is not established as the first location accreditation [166, 167, 521, 569, 164]. Foursquare's 2020 release said Attribution "has not been audited or accredited by the MRC" [166, 167, 245].

GroundTruth's accreditation came in July 2021, as MRC recorded on 20 October 2021, and covered a data set, not a visit-lift or attribution metric [168]. In September 2023 MRC granted a six-month hiatus at GroundTruth's request, during which the service was not accredited. MRC later recorded the accreditation as revoked "due to withdrawal from the MRC accreditation process", in a row labelled February inside its April to June 2024 communication of 26 July 2024 [169, 170, 368, 369]. The effective month is unconfirmed [168, 169, 170, 356]. GroundTruth's own pages still present the accreditation as current. Its August 2021 release carries no hiatus or revocation notice, and a comparison page last modified on 9 November 2023 says in the present tense that GroundTruth is "Accredited by the Media Rating Council (MRC)" for its place, location and visitation datasets [356, 358, 169].

Digital Envoy's NetAcuity, the only other MRC-accredited location-determination service found in the communications screened, became non-accredited in January 2020 after Digital Envoy chose not to proceed with a new audit. No IP-geolocation service was on MRC's lists on 24 October 2026 [165, 171]. Epsilon's Correlated Outcomes is the only accredited outcomes metric on MRC's digital list. Its letter cites impression, invalid-traffic and data-integration standards, not the location guidelines, and does not define offline outcomes as store visits, so it does not satisfy H6 [570].

### H6, coded by the rule

Under the registered rule, H6 is contradicted on MRC's lists as read on 24 October 2026 [171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 567, 568]. H6 states that at least one US footfall or visit-measurement product holds current MRC accreditation of its visit or visit-lift metric at the cutoff. The registered rule codes it contradicted when MRC's lists at the cutoff carry no accredited visit, visitation or visit-lift metric for any product. Lapsed or withdrawn accreditations are reported with their dates and do not change the verdict, and audits in progress confirmed by MRC are reported separately.

On 24 October 2026, none of MRC's six accredited-services lists and none of its five under-review lists carried a visit, visitation, foot-traffic or location-determination metric. The closest entries under review are Kevel's "Conversions and incrementality" and Kroger's attributed and incremental sales. MRC's April to June and July to September 2026 communications record no location or visit action [171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 567, 568]. The rule dates the verdict to the cutoff, 24 October 2026; a later change will be logged as a dated correction ([chapter 13](#method)).

Read vendor by vendor, the lists give the same answer. On 24 October 2026 Blis, Foursquare, GroundTruth, InMarket, Placer, Unacast and Vistar appeared on neither MRC's accredited nor its under-review digital list [171] [171, 173] [171, 177] [171, 177, 267] [171, 177, 386, 292] [171, 177, 137, 414] [171, 177, 173, 178]. InMarket's assurance claims (Comscore verification, Location Sciences certification and a Forrester TEI study) are not accreditations [171, 177, 267]. Location Sciences' certification of InMarket audiences dates from 2020, and no later audit, renewal or report was found, although InMarket's 2026 data page still says "Location Sciences certified" [378, 373, 267]. Placer's documented certifications are information-security ones [171, 177, 386, 292].

### Self-regulation: the NAI

The NAI's Precise Location Enhanced Standards were first issued on 22 June 2022 with Cuebiq, Foursquare and PlaceIQ as signatories, and reissued as "Updated and effective October 2024". The signatory page in the PDF, last updated March 2025, shows four logos: Cuebiq, Foursquare, InMarket and Outlogic. Within US activities the standards bar using or transferring activity correlated to a known sensitive point of interest (11 definitive and 2 non-definitive categories), and uses for law enforcement, national security or bounty hunting, subject to exceptions [564, 565, 243]. PlaceIQ, an initial signatory, is not among the four [564, 565, 243].

NAI's 2020 Code defines precise location information as data from technology able to determine a device's actual physical location with reasonable specificity, and requires opt-in consent for its use in tailored advertising. NAI's February 2020 guidance describes three ways to make location imprecise: two or fewer decimal places, an area at least as large as a circle of 500 m radius, or a non-unique general descriptor of the place [461].

Two of the four signatories are under FTC orders [130, 132] [126, 127]. InMarket's order, finalised in May 2024, bars it from selling or licensing precise location data and from products that categorise or target consumers on sensitive location data [130, 132]. The X-Mode order binding Outlogic, issued in April 2024, bars uses of sensitive location data and does not ban the sale of precise location data as such [126, 127]. Outlogic's own site says the company may "receive and sell the precise latitude and longitude coordinates" of a device [128, 129]. Foursquare's privacy policy says that, as a participant in the standards, it does not use or share activity correlated with 13 listed sensitive-place categories, a vendor statement [244, 243]. Unacast cites NAI membership "in good standing" with "annual reviews", a self-regulatory status with no public product-scoped result [171, 177, 137, 414]. [Chapter 7](#regulation) covers the orders.

### Who sets coordinate precision

The specification does not. OpenRTB 2.6's Geo object carries latitude and longitude as floats, a type for the source (1 GPS or location services, 2 IP address, 3 user-provided), accuracy in metres, seconds since the fix and the IP location service. Coordinates should be passed only if they match the accuracy of the stated type, and a postal-code centroid should not be passed. The 2.6 text has no rule on truncating or rounding coordinates [219, 460].

Precision is set downstream. Google's Authorized Buyers documentation, last updated 17 September 2026, states that its bid-request location comes only from IP geolocation and that Google limits latitude and longitude to 0.01 degrees, a description of its own exchange [444]. Prebid's documented coarsening happens only when a publisher rule or a configured privacy module denies the transmitPreciseGeo activity. Prebid Server then truncates to two decimals and masks part of the IP address. The default is allow, and the Prebid Mobile Android SDK sets no precision limit by default [437, 438, 439]. Across these documents two decimal places is the recurring boundary between precise and coarse, but the bodies attach different distances to it. NAI's 500 m radius is about 6% smaller than the 1,750-foot radius in several state laws [461, 445, 462, 444, 437]. MRC's 2017 text warns that decimal count must not be equated with accuracy [445]. In this paper's reading, a two-decimal rule therefore bounds precision, not accuracy.

OpenRTB remains the baseline in production: Google documents bid requests whose geo type is always IP, with coarse accuracy [219, 220]. Hierarchical geographic cells are not part of AdCP at the cutoff versions. Issue #5969, which proposes them, was opened by the author and was open with milestone 3.4.0 when read on 24 October 2026 [205, 206, 207]. It is a declared interest and a rival framing, not evidence, and [chapter 10](#agents) records the protocol states and the author's role in the working group.

### IAB guidance

IAB's location-data guidance found by the cutoff runs from the gated Mobile Location Data Buyer's Guide (July 2014) and the February 2016 Mobile Location Data Guide for Publishers to the April 2017 Location-based Marketing Playbook for Retailers. The playbook is buyer-facing. It asks whether device data come "through panels or pass through ad exchanges" and how store-visit-lift control groups are selected. No IAB or IAB Tech Lab location guide later than April 2017 was found in the sources screened [462, 571]. Verve is among the playbook's listed contributors, and Verve Group is the author's former employer [1]. Media and Games Invest agreed in January 2020 to buy substantially all of Verve Wireless's assets and renamed itself Verve Group SE in June 2024 [495, 496].

IAB published a Digital Out-Of-Home (DOOH) Measurement Guide on 8 July 2025, behind a free-account gate. IAB's blog of that day says it covers attribution through "synthetic control groups and matched market testing". The blog does not mention MRC, Geopath or OpenRTB [572, 573, 574, 575]. Under this paper's rules a matched comparison is not causal evidence.

### Geopath: a currency without MRC accreditation

On 24 October 2026 MRC's out-of-home accredited list was empty, its only out-of-home service under review was JCDecaux's Proof of Plays metric, and Geopath appeared on no MRC list opened. In the sources screened, the US out-of-home audience currency is not MRC-accredited [173, 178, 171]. Geopath publishes the audience estimates US out-of-home trades on, as annual planning datasets. The 2026 dataset went into preview on 30 September 2026 and, per its 19 October post, becomes the default transactional planning dataset on 12 November 2026, after the cutoff [579, 580]. It is a not-for-profit governed by a tripartite board of advertisers, agencies and media companies. Its own page says it was established in 1934 as the Traffic Audit Bureau for Media Measurement, which conflicts with the 1933 in the author's essay [581].

Geopath did help write MRC's standard for the medium: the Out-of-Home Measurement Standards, final December 2025, were developed with DPAA, Geopath and OAAA, and they name mobile location among deterministic inputs and require services to ascertain the accuracy of third-party data [576, 577]. Geopath's methodology overview says it is not designed to provide a comprehensive statistical validation and reports no confidence intervals [578]. Its reach and frequency model is self-reported to align with observed data within ±5%, with no provider, sample or validation method named [184]. In March 2026 Geopath and the OAAA announced an Ipsos-supported pilot in the second half of 2026, a transition beginning in 2027 and full adoption in 2028 [606, 607]. [Chapter 9](#dooh) reads its data sources.

### What assurance covers

For location in the US, the MRC record holds a 2017 standard with no lift provision, two revoked data-set accreditations and no current visit metric [445, 600, 601] [445, 454] [166, 167, 521, 569, 164] [168, 169, 170, 356] [166, 168, 521, 171] [171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 567, 568]. Everything else a buyer is shown is a different kind of assurance: NAI self-regulation, exchange documentation, vendor certifications and pages that still describe a revoked accreditation [564, 565, 243] [444] [171, 177, 267] [356, 358, 169]. A buyer can ask for the MRC letter and check its date, metrics and environments against the current list. [Chapter 11](#gq) scores independent assurance on that basis, and [chapter 12](#next) turns it into questions.



## 9. Out-of-home runs on mobility data


*US out-of-home trades on an audience currency built from smartphone location panels, and the programmatic sellers on top of it add more mobility data for targeting and attribution. Geopath names its sources, but its panel figures date from 2023 documentation and its own pages say they offer no comprehensive statistical validation. Three of six US DOOH products and currencies coded name their location sources, and only Geopath states a panel size. No DOOH footfall product or case screened documents a randomised or holdout control. Geopath and the OAAA have picked Ipsos to support a pilot of a successor system, and no document found by the cutoff says what data it will use.*

> **In short.**
> Geopath's 31 May 2023 methodology names MotionWorks device panels, Teralytics/Streetlytics and HERE for roadways, and Claritas among its demographic sources, with no proportions [182, 578]. Its pages describe more than 100 million unique US devices a month culled to seven to ten million qualified devices a week, a documentation-dated figure with no observation vintage [183] [578, 582, 185, 583]. A 19 October 2026 post names Bentley Systems as roadside data provider, and the 2026 Planning Dataset becomes the default on 12 November 2026, after the cutoff [182, 578] [579, 580]. Geopath's likelihood-to-see impressions are opportunity-to-see counts adjusted by a visibility model [584, 182], its ±5% accuracy claim compares a model with the mobile data it was trained on [184], and on 24 October 2026 it appeared on no MRC list opened [173, 178, 171]. An Ipsos-supported pilot was announced to launch in the second half of 2026, with transition from 2027 and full adoption in 2028; no document on the pilot's data sources was found by the cutoff [606, 607] [585, 579]. Of six US DOOH products and currencies coded under Q2, three name their location sources and one states a panel size [182, 23, 192, 195, 608, 194]. Of six DOOH footfall-attribution products or cases screened, none documents a randomised or holdout control [196, 195, 194, 23, 187, 257]. OpenRTB's qty object carries a DOOH impression multiplier and the measurement vendor's domain, but no field for its mobility source [586, 587]. T-Mobile has owned Vistar since 3 February 2025, and since 5 October 2026 its enhanced segments, now with "location insights", have been sold for DOOH only through the Vistar DSP [36] [334].

### The currency and its inputs

Geopath's methodology documentation, published on 31 May 2023, names MotionWorks device panels, Teralytics/Streetlytics and HERE Maps for roadways, HERE Maps for vehicle travel, and government agencies, MotionWorks and Claritas for demographics, without proportions [182, 578]. Under the Q2 coding rule Geopath is coded source disclosed [182, 23, 192, 195, 608, 194]. Geopath says it was established in 1934 as the Traffic Audit Bureau for Media Measurement and is a not-for-profit governed by a tripartite board of advertisers, agencies and media companies [581]. It is a currency body, not an accreditor or verifier [337, 8].

The documentation is a set of Confluence pages [578, 582, 185, 583]. A "Mobile Device Location & Movement Patterns" page says source data from multiple location SDK aggregators cover more than 100 million unique US devices a month and that Intermx culls them to a sample of seven to ten million qualified devices a week [578, 582, 185, 583]. Each record carries a device ID, a UTC timestamp, coordinates and horizontal precision in metres, and imprecise sightings are removed under no stated numeric threshold [578, 582, 185, 583]. The Mobile Device Panel page of the same date describes a weekly panel of GPS-sighted devices active about eight hours a day with a consistent Census Block Group home, culled by Standard Data [183]. A Movement Patterns page records AirSage origin-destination matrices for the 2019-2020 methodology and a 2021 expansion contracted to Standard Data [578, 582, 185, 583]. These device counts are a documentation-dated disclosure with no stated observation vintage, not a current audited panel count [183]. Whether MotionWorks still supplies the panel in 2026 is not stated in any document found [182, 578].

The dataset in use at the cutoff has no published source list later than 2023 [182, 578]. Geopath's 2026 Planning Dataset went into preview on 30 September 2026, and its 19 October 2026 post names Bentley Systems as roadside data provider and credits "updated data sources" without listing them [579, 580]. That dataset becomes the default transactional planning dataset on 12 November 2026, a plan at the cutoff [579, 580] [182, 578]. This review found no statement of sources later than the 2023 list for the dataset in use at the cutoff [182, 578].

![Three of six US DOOH products name their mobility source, and only Geopath states a panel size with a period.](figures/fig-dooh-sources.svg)

*Figure fig-dooh-sources. **Three of six US DOOH products name their mobility source, and only Geopath states a panel size with a period.** Question Q2 coding of US DOOH audience currency, SSP and measurement products: the mobility source named in the product’s or operator’s own documentation, the kind of source, and whether a panel size is stated with a period (final wording of LOC-I-F23). A partnership announcement establishes a relationship, not a dependence. Place Exchange, Broadsign and Perion describe device data without naming where it comes from. Units: coding. Scope: US DOOH currency, SSP and network products, documentation read by the cutoff. Sources [182], [183], [184], [185], [186], [23], [187], [188], [189], [190], [191], [192], [193], [194], [195], [196]. Caveat: Geopath’s figures are documentation-dated (2023), not a current audited panel count; its 2026 dataset becomes the default on 12 November 2026, after the cutoff. Data: data/figure-data.json#dooh*

### What a Geopath impression is

A Geopath impression is a projected, visibility-adjusted estimate [584, 182]. Its core outputs are likelihood-to-see impressions, audience composition, and reach and frequency [584, 182]. Opportunity-to-see counts are adjusted by its proprietary visual acuity index (VAI) model for the probability of noticing the unit [584, 182]. Composition comes from treating a device's longest-dwell location as its home and applying census-block demographics, computed as an annual average by hour of day [584, 182]. The VAI model's inputs and parameters are not on the pages read [584, 182]. Under the Q1 coding the impressions are projected: panel observations scaled to a population and adjusted for visibility [584, 182]. Route's eye-tracking adjustment is the Great Britain counterpart of the VAI [591, 592].

Geopath's reach-and-frequency approach, introduced with its 2023 data release, is "centered around observed mobile location data from smartphone applications and connected cars" and is self-reported to align with that data "with ±5% accuracy" [184]. The "observed data" is the same mobile-derived panel measure the model is trained on, so the ±5% is agreement between a model and its own input, not with an independent count [184]. The page names no provider, sample size or validation method [184]. Geopath's methodology overview says it is not designed to provide a comprehensive statistical validation and reports no confidence intervals [578]. In the sources screened by the cutoff, this review found no independent validation of Geopath reach and frequency [184].

On 24 October 2026, MRC's Out-of-Home accredited list was empty, the only OOH service under review was JCDecaux's Proof of Plays metric, and Geopath appeared on no MRC list opened [173, 178, 171]. MRC's Out-of-Home Measurement Standards (final December 2025, developed with DPAA, Geopath and OAAA) treat location data as foundational where persons are not measured directly [576, 577]. They require services to ascertain the accuracy of third-party data, with disclosure when a source cannot be independently verified, and name "mobile location" among the deterministic inputs for adjusting likelihood-to-see to audience [576, 577]. [Chapter 8](#standards) sets out the standards record.

### The Ipsos pilot

On 31 March 2026, Geopath and the OAAA announced that their modernisation "began in early 2025" and that Ipsos had been selected to support an industry pilot launching in the second half of 2026 [606, 607]. The pilot is to be followed by a transition to a next-generation framework beginning in 2027 and "full adoption in 2028"; the release names no data sources [606, 607]. Between that release and the cutoff, this review found no Geopath or Ipsos document describing the pilot's data sources, panel or method in the sources screened [585, 579]. Ipsos's 12 May 2026 session page mentions the pilot without method, and Geopath's 30 September 2026 dataset post does not mention it [585, 579].

Ipsos already recruits and processes the sample for Route, Great Britain's out-of-home currency for adults 15+ [591, 592]. Route combines passive GPS tracking of a recruited, consented sample of c.7,000 participants a year with volumetric counts and Lumen eye-tracking, and uses no third-party smartphone SDK panel [591, 592]. Route is an international attribute and earns no US score, and no document found says whether the US pilot will follow its design [591, 592] [585, 579]. No COMMB methodology page with data sources could be read by the cutoff, so Canada's currency is coded not disclosed [593].

### Which sellers name their sources

Programmatic is a minority of DOOH spend on WOO's figures. Its study, aggregated by PwC and announced on 5 June 2026, puts global programmatic DOOH at about US$1.4 billion of revenue in 2025, 7% of total DOOH expenditure, with the Americas (not US-only) at US$670 million and 14.2% penetration [588]. The data were submitted by 11 SSPs, and PwC "has not independently verified the information which it received from respondents"; the announcement says nothing about audience or mobility data [588].

| Product | Mobility source named | Panel size with period | Q2 code |
|---|---|---|---|
| Geopath (currency) | MotionWorks panels, SDK aggregators; roadway and demographic suppliers | 7–10M qualified devices weekly, in 2023 documentation | Source disclosed [182, 578] [578, 582, 185, 583] [182, 23, 192, 195, 608, 194] |
| Vistar attribution | InMarket SDK data; Foursquare and Placed | Not stated | Source disclosed [23] [187, 189] |
| Vistar audience targeting | Partner pages: GroundTruth, Adsquare, Foursquare, PlaceIQ, T-Mobile | Not stated | Source disclosed; dependence not documented [188, 189, 190] [182, 23, 192, 195, 608, 194] |
| Place Exchange PerView | "Daily updated mobile data", provider unnamed | Not stated | Not disclosed [192, 193] |
| Broadsign measurement | Partner logos only | Not stated | Not disclosed [194] |
| Perion (Hivestack) | "Anonymised device-level location data" | Not stated | Not disclosed [195, 589, 590] |

Of these six US products and currencies, three name their location sources in their own or the operator's documentation, and only Geopath states a panel size with a period [182, 23, 192, 195, 608, 194]. A not-disclosed code means not readable in public documentation at the cutoff, not independence from mobility data, and gated buyer pages may disclose more [182, 23, 192, 195, 608, 194]. Two products outside the six name a source. Broadsign Ads' audience targeting, registered but outside this US count because its US availability is unconfirmed, names Native Touch data "pulled from active application SDKs and APIs" [538]. GroundTruth's DOOH attribution with Place Exchange uses GroundTruth's own "location and purchase data" [257]. Broadsign announced its acquisition of Place Exchange on 25 November 2025, and Perion completed its acquisition of Hivestack on 12 December 2023 [45] [35].

Vistar documents its partners in a public directory [189]. The directory, read on 24 October 2026, tags 18 partners for US audience segments and six USA-tagged foot-traffic partners: Accretive, Arrivalist, Cuebiq, Foursquare, InMarket and Spotzi [189]. Its partner pages describe GroundTruth's data as "first-party data, third-party partnerships, and bid stream data", Adsquare's as including "mobile location data" and InMarket's as "200M+ location-based users", with no proportions or precision class [191, 539, 416, 540, 23]. Under the Q2 rule a partnership announcement establishes the relationship, not the dependence [188, 189, 190]. InMarket's "nearly 200M+ permission-based users" has no period, geography or definition of users, and the same page gives two transaction figures, ~$122B+ and $2.5T+ [23].

Exposure is inferred from where a device was, and the documents opened state no full rule for it. Place Exchange describes a "patented process of capturing exposed mobile IDs for consumers in the viewshed of every screen at the time of ad play" and names no provider for them [192, 193]. GroundTruth counts a DOOH view when a device enters a screen's mapped viewshed "for the appropriate amount of time", with no time stated [257, 254]. Adsquare, in an interview, describes inferring exposure from the proximity of mobile advertising IDs to screen coordinates while an ad is live, with no radius stated [464]. Vistar's 2022 case says exposure for studies uses "ad log timestamps, device location, venue type and screen location", and Vistar publishes no distance, dwell or line-of-sight rule [337, 8].

### The multiplier in the bid

OpenRTB 2.6 carries the audience estimate into the auction in its qty object [586, 587]. Its multiplier is the quantity of billable events deemed to occur ("a DOOH opportunity may be considered to be 14.2 impressions"), with a source type and, for vendor-provided values, the measurement vendor's domain [586, 587]. AdCOM lists the source types as unknown, measurement vendor, publisher and exchange [586, 587]. Neither document records the vendor's mobility source, panel or method, and neither enumerates multiplier vendors; one web search for a vendor list found none [586, 587]. In The Trade Desk, Vistar's audience impressions are "the result of impression multiplier calculations provided by SSPs & media owners", which its guide says "are commonly sourced from verification partners like Route and Geopath", while plain impressions count ad plays [337, 8]. Geopath and Route are currency bodies, not verifiers [337, 8]. Inference: a bid can carry a panel-derived audience count without saying which panel produced it [586, 587].

### Footfall attribution without a holdout

Of six DOOH footfall-attribution products or cases screened (four US, one UK, one global), none documents a randomised or holdout control: none is randomised or holdout, one asserts a matched control, two use unexposed comparisons and three do not disclose a control [196, 195, 194, 23, 187, 257]. With three of six undisclosed, fewer than four document a design, so H3's decision tree applied to this DOOH subset alone gives unresolved [196, 195, 194, 23, 187, 257]. H3 itself is tested on cohort A, which [Chapter 5](#measurement) reads as contradicted: seven of nine documented methods are non-holdout [5, 19, 119, 17, 20, 21, 22, 25, 6]. H3's dated clarification of 24 October 2026 fixed that decision tree and by itself did not move the verdict.

Hivestack's study for Zoom Media UK and Gillette (London, November 2022, 115 screens) compared devices seen in a screen geofence while the ad played with devices in the same geofence when it did not [196]. It reported a "2.12x lift" and 8,574 "Modeled incremental walk-ins", scaled by the ratio of an unnamed data partner's phone access to the GB population [196]. That is a non-randomised time-split comparison, not a holdout [196]. Perion's current page offers footfall lift "comparing exposed audiences with a matched control group" without a provider or matching method, and an undated Perion blog names Foursquare, Cuebiq, PlaceIQ and Adsquare as pre-integrated providers [195, 589, 590]. Broadsign's Decathlon case reports "a 46% total increase in store footfall" and "an incremental 8 percentage point lift" against "non-exposed control groups"; only the second is described as incremental, and the control's formation is not disclosed [194].

Vistar documents no attribution method of its own at the cutoff [337, 522]. Its activation guide says it "works with trusted industry leaders" for studies that "require a feasibility test and have varying minimums and costs", and vistarmedia.com/measurement returned HTTP 404 [337, 522]. It has sold foot-traffic studies through rotating third parties since at least January 2019 [594, 187]. In two OAAA-hosted cases measured by Foursquare, Target's reported a 6.51% visitation lift over a four-day look-back "compared with non-exposed consumers", and Jack in the Box's an 8.8% lift, neither describing the control [8, 9, 463]. Whether the 533,000 Target visits are observed or projected is not stated, and Vistar's visit basis is coded not disclosed [8] [16, 17, 18, 5, 449]. GroundTruth's Place Exchange release calls "more than half" of a national retailer's more than 42,000 exposed-consumer visits incremental with no control design described [257]. On 24 October 2026 no Vistar metric was found on MRC's accredited or under-review lists [171, 177, 173, 178].

The channel's guidance recommends matched designs and mentions holdouts as an aid. IAB's July 2025 DOOH Measurement Guide, gated on IAB's site but publicly retrievable as a PDF and read in the final live sweep, recommends synthetic-control and matched-market or matched-store tests for incrementality and says combining synthetic models with geographic holdouts can reduce contamination of the control group [572, 573, 574, 575] [572, 573, 575]. That is guidance, not evidence that any of the six products above uses a holdout by default. Vistar's material does contain a randomised design, in a creative test [419]. Its 23 September 2026 study with Omnicom Media and Annalect randomised 7,200 consumers in the Netherlands to a control creative or one of three store-locator placements and measured stated visit and purchase intent, which is declared, not a visit [419]. The author earlier worked at Hearts & Science, an Omnicom Media agency.

### A model on a model?

The author's essay "Measurement, on the Street's Terms" argues that a DOOH footfall or lift study is "a model on a model" [2]. This chapter treats that as a rival framing to test. Where the layers are documented, it mostly holds. The audience behind a multiplier is a projected, visibility-adjusted estimate [584, 182]. Exposure is inferred from device proximity to a screen [337, 8] [464]. Several documented visit bases are projected, as in Hivestack's modelled walk-ins and the projected store visits of Foursquare Attribution and InMarket LCI [196] [16, 17, 18, 5, 449].

Two limits apply. Some inputs are panel observations, not models: InMarket's exposed device visits and GroundTruth's Observed Visits are coded observed [16, 17, 18, 5, 449]. And for Vistar the basis is not disclosed, while Place Exchange, Broadsign and Perion do not name their sources, so for those products the framing can be neither confirmed nor rejected from public documents [16, 17, 18, 5, 449] [182, 23, 192, 195, 608, 194].

### The carrier in the stack

T-Mobile US agreed on 20 December 2024 to acquire Vistar Media and completed the deal on 3 February 2025, transferring US$621 million in cash [36]. At the 13 January 2025 announcement the parties described more than 1.1 million digital screens from nearly 370 OOH media owners [336]. Lamar received US$115.1 million at closing for its 20% stake [37]. T-Mobile also completed its acquisition of Blis on 3 March 2025 [36, 228].

T-Mobile's app-ownership and engagement segments have been on Vistar's self-service DSP since 6 May 2025, and on 5 October 2026 Vistar announced enhanced segments that add "location insights", available exclusively through the Vistar DSP [334]. Whether those insights are network-derived or app-permission location is not stated [492, 334]. AdExchanger, relaying T-Mobile's account, reported that the data include mobile location used to estimate inventory impressions, and that for users who opted into location tracking T-Mobile can detect proximity to OOH ads and retarget on other channels [335]. T-Mobile tied the exclusivity to tightening regulation of location data and device IDs, and called DOOH measurement "on our road map to look into for next year" [335]. Those are an interested party's statements, relayed by trade press [335].

Vistar's own documents disagree about what it handles. Its audience guide says DOOH targeting works "without ever relying on personal information", while its Services Privacy Notice (1 July 2026) says it receives device location histories tied to device identifiers and sells and shares personal information [338, 339] [339]. On California's 2026 Data Broker Registry, Vistar declared "Yes" to collecting precise geolocation, after declaring "No" in the 2025 file [163, 595, 596]. The FCC's US$80,080,000 forfeiture against T-Mobile concerns its carrier location programme before Vistar was acquired, not Vistar; the carriers' certiorari petition was distributed for the 6 November 2026 conference, after the cutoff [340, 341].

AdCP v3.2.6, the agentic buying protocol, accepts Geopath-prefixed venue IDs and gives geopath and route as example measurement_source values, and its foot_traffic metric carries no field declaring the control design [208, 211, 209, 597] [211, 212]. The author's RFC #5969 for geographic cells (21 July 2026) is open in the AdCP working group he co-leads and has no protocol state [205, 206, 207]; [chapter 10](#agents) treats it as a declared interest.



## 10. Can an agent buy by place?


*An agent can describe a place in AdCP in far more detail than any agent has been shown to buy one. Version 3.2.6 carries administrative geography, radius and polygon proximity, store catchments, named places, DOOH venue identifiers, per-venue delivery and a foot-traffic metric. In the sources screened, the furthest any location signal gets in open-protocol agent-driven buying is one implemented connector, built on an earlier beta, that maps countries, regions and DMAs into Google Ad Manager. No interoperability test result is public, and the vendor releases that claim geo-targeted agentic buys name no field. Under the registered rule, H7 is unresolved.*

> **In short.**
> At AdCP v3.2.6, the stable release of 23 October 2026, the targeting overlay represents countries, ISO 3166-2 regions, metros, postal areas, point-based proximity, store catchments and identifier-based named places, as schema representation and nothing more [197, 221]. Delivery reporting represents DOOH plays, a per-venue breakdown, a measurement_source and a foot_traffic metric qualified by methodology, with no field for the control design [211, 212]. Geographic cells are in neither v3.2.6 nor v3.3.0-beta.0, and the author's proposal for them is open [205, 206, 207]. The only implemented connector found, Prebid Sales Agent v2.0.0, is built against AdCP 3.1.0-beta.3; it maps administrative geography into Google Ad Manager and refuses postal areas [198]. On 24 October 2026 none of the 34 agents in AAO's public registry showed a passing geo or DOOH conformance test [200, 597]. The vendor releases that claim geo or venue targeting in agentic AdCP buys name no field and no protocol version [201, 210]. IAB Tech Lab's AAMP components carry geography only as coarse codes, and ARTF 1.0 has no standard intent to change or add a geographic field [213, 214, 215] [216]. The location vendors' agent interfaces found sit outside agent-driven buying [217, 218, 222, 200]. H7 is unresolved [198, 200, 201, 205, 598, 599].

### The test

H7 asks whether location signals (geo targeting fields, geographic cells, venue identifiers or visit metrics) reach documented production use in agent-driven buying by the cutoff. Shared rule 12 records five states for each protocol and version: schema representation, an implemented connector, a demonstrated interoperability test, documented production use and a demonstrated transaction. H7 is supported when at least one location signal reaches documented production use or a demonstrated transaction in an agent-driven buying workflow, with the protocol, version and signal recorded. It is contradicted when a source documents that a location signal was dropped, could not be carried, or failed in such a workflow or an interoperability test, and no location signal reaches documented production use. Anything short of both is unresolved, and the highest state reached is reported. Membership, working-group roles, proposals and announcements earn no state ([Chapter 13](#method)).

The protocols screened are AdCP (the Ad Context Protocol), IAB Tech Lab's AAMP (Agentic Advertising Management Protocols) components and ARTF (the Agentic Real Time Framework) 1.0, with OpenRTB 2.6 as the non-agentic baseline [198, 200, 201, 205, 598, 599] [219, 220]. The screen also covered location vendors' own agent interfaces, including servers built on MCP (the Model Context Protocol), through which AI assistants call outside tools [217, 218, 222, 200]. Versions and statuses below were read on 24 October 2026, the cutoff [197, 221] [205, 206, 207].

> **Interests.**
> The author opened the geographic cell proposal that this chapter codes and co-leads the working group in which it was raised [205, 206, 207]. Two of the 34 agents in the public registry of AgenticAdvertising.org (AAO), read below, are his own, and neither carries a geo or DOOH result [200, 597]. He also made the dated ruling, after extraction, on how H7's contradicted rule treats a connector that refuses a field by design; it is reported with the verdict below. None of these earns a state.

### What AdCP can represent

AdCP's latest stable release at the screen was v3.2.6, published on 23 October 2026 [197, 221]. Its media-buy targeting overlay represents administrative geography: countries, regions as ISO 3166-2 codes, metros in the nielsen_dma, uk_itl1, uk_itl2, eurostat_nuts2 or custom systems, and postal areas, each with a matching exclusion field. It represents geo_proximity as a radius, a travel-time isochrone or a buyer-supplied GeoJSON polygon, and store_catchments drawn from a synced store catalog [197, 221]. All of this is schema representation, which shows that a value can be expressed and nothing more [197, 221].

Identifier-based geo_places entered through PR #6093, merged on 15 August 2026, first published in the pre-release 3.2.0-beta.0 on 17 August and first stable in 3.2.1 on 1 October 2026 [203, 202, 597]. It keys each place by a stable ID in a declared system (geonames, google_ads, microsoft_ads or an owner HTTPS URI) and a place type drawn from administrative units plus airport. Its registered vocabulary therefore names cities and neighbourhoods, not store-level points of interest, and the PR defers place-level forecasting and delivery breakdowns [203, 202, 597].

DOOH has its own identifiers. Placements accept venue_id, screen_id and openooh_venue_type identifiers and authority-prefixed external IDs, with geopath:30961 as the schema's example. DOOH placements declare slot and loop durations, and the experimental static-OOH panel identifiers accept a Geopath Spot ID scheme [208, 211, 209, 597]. Delivery reporting has a top-level plays count and a dooh_metrics block of loop plays, screens used, screen time, share of voice achieved, calculation notes and a per-venue venue_breakdown [211, 212]. A measurement_source slug lists geopath and route among its examples, and an experimental ooh_metrics block tiers estimated impressions as currency_measured or seller_modeled ([Chapter 9](#dooh)) [211, 212].

Visits travel without their control design. The foot_traffic metric is defined as incremental visits over baseline and qualified by an attribution_methodology of panel_based, deterministic_purchase, probabilistic or modeled. No field declares the control design, on a search of the v3.2.6 schema source for holdout, ghost, control-group and randomisation terms [211, 212]. On this review's reading, a foot_traffic value received through AdCP cannot tell a buyer whether its baseline was randomised, held out or matched, which is the distinction H3 turns on ([Chapter 5](#measurement)) [211, 212].

### Cells earn no state

Hierarchical geographic cells are in no AdCP version current at the cutoff. Issue #5969 proposes geo_cells and geo_cells_exclude with an h3, s2, geohash or custom cell-system enum. The paper's author opened it on 21 July 2026, and it is open on milestone 3.4.0 with the labels rfc and needs-wg-review [205, 206, 207]. Its companions, #6013 on conversion and #6014 on cell-grain delivery reporting, are open on the same milestone. A GitHub search on 24 October 2026 found no geo_cells pull request, and neither v3.2.6 nor v3.3.0-beta.0 contains the field [205, 206, 207]. Under the brief's rules the proposal earns no state, whatever its standing in the working group the author co-leads [205, 206, 207].

### One connector, on an earlier beta

In the protocols, repositories and registry screened on 24 October 2026, the highest state any location signal reaches in open-protocol agent-driven buying is an implemented connector [198, 200, 201, 205, 598, 599]. The registry lists registered agents, not every AdCP deployment. Prebid Sales Agent v2.0.0, released on 1 July 2026, is built against AdCP spec 3.1.0-beta.3 (adcp Python SDK 5.7.0) rather than 3.2.6. It translates geo_countries, geo_regions and geo_metros, Nielsen DMA only, with their exclusions, into Google Ad Manager targetedLocations [198]. It rejects geo_postal_areas with an explicit error, "not implemented in GAM static mapping". The targeting manager read has no handling for geo_proximity, store_catchments or geo_places, an absence bounded to that one file [198]. The release predates geo_places by about seven weeks, since geo_places was first published on 17 August 2026 [203, 202, 597] [198].

Code shows a connector exists. It does not show use, and no source opened shows a production buy that exercised this mapping [198] [198, 200, 201, 205, 598, 599].

![No location signal has a public interoperability test; administrative geo reaches an implemented connector, on AdCP 3.1.0-beta.3, and geographic cells reach no state.](figures/fig-protocol-states.svg)

*Figure fig-protocol-states. **No location signal has a public interoperability test; administrative geo reaches an implemented connector, on AdCP 3.1.0-beta.3, and geographic cells reach no state.** Each class of location signal in the agentic protocols current at the cutoff, against the five states: schema representation, implemented connector, demonstrated interoperability test, documented production use and demonstrated transaction (stream J, corrected by its verifier). OpenRTB 2.6 is the non-agentic baseline: Google documents IP-derived, coarse geolocation in its live bid requests. H7 is unresolved under its registered rule. Units: state. Scope: AdCP 3.2.6 and 3.3.0-beta.0, AAMP components, ARTF 1.0, location-vendor MCP servers. Sources [197], [198], [199], [200], [201], [202], [203], [204], [205], [206], [207], [208], [209], [210], [211], [212], [213], [214], [215], [216], [217], [218], [219], [220]. Caveat: The author co-leads AdCP’s Signals & Measurement working group and wrote the geographic-cell proposal; neither earns a state. The registry is live data, read on 24 October 2026. Data: data/figure-data.json#protocols*

### No public test passes

A conformance storyboard pass is the nearest public analogue to an interoperability test, and none is public for location [200, 597] [198, 200, 201, 205, 598, 599]. In AAO's public registry on 24 October 2026, none of the 34 agents shows a passing geo or DOOH conformance test [200, 597]. The geo storyboards (geo_region_targeting, geo_region_all_values_support, geo_region_exclusion_support and geo_place_targeting) appear only on two agents graded on AdCP 3.2.6, goTom Sandbox and Publink, and both are untested at zero of zero steps [200, 597]. AdCP's sales_dooh specialism tests a seller exposing a venue_id and reporting plays and dooh_metrics, and no registered agent declares it [208, 211, 209, 597] [200, 597]. The 11 agents with a verified badge hold it in spec mode only, none in the live-traffic mode [200, 597]. These are public compliance records as read on 24 October 2026, the cutoff; they change as agents are regraded [200, 597].

### What vendors claim

The strongest public claims of location signals in agent-driven AdCP buys are vendor releases that record neither the signal nor the protocol version [201, 210]. PubMatic and Butler/Till, on 15 December 2025, describe an AdCP-enabled CTV campaign for Clubtails "delivering geo-targeted, premium CTV inventory" without naming a field or a geographic unit. Broadsign, on 27 May 2026, describes a Netherlands OOH campaign in which agents using AdCP informed audience and venue targeting, without naming identifiers or delivery metrics [201, 210]. Both are vendor assertions, and neither reaches documented production use of a named location signal under H7's rule [201, 210].

Outside the open protocols, proprietary DOOH seller agents reached over MCP, from VIOOH and Displayce, claim geographic and proximity handling. They name no fields and report no transaction [198, 200, 201, 205, 598, 599]. InMarket announced Maestrø in beta on 18 June 2026 as an agentic optimisation layer that feeds InMarket outcomes to AI agents, with an Open Access route "Built on MCP". No public MCP server or AdCP agent for InMarket was found, so it is an announcement, not a documented protocol implementation [273, 523].

### AAMP and ARTF stay coarse

IAB Tech Lab's AAMP components represent geography only as coarse codes at schema level [213, 214, 215]. OpenProposal 3.0 draft 1, published for comment on 22 September 2026, gives the targeting envelope a settable geo object with an allowed[] list of geo codes. The pre-release iab-agentic-primitives 0.5.1 gives packages geo_targets as ISO 3166-2 codes, and its linear-TV quotes use Nielsen DMA codes [213, 214, 215]. The reference seller-agent 2.4.2 carries geo_targets in its media kit. A grep of its ad-server clients found no mapping of geo_targets into Google Ad Manager geo targeting, although its GAM model declares a geoTargeting field [213, 214, 215].

ARTF 1.0, released on 12 November 2025, carries location only inside the embedded OpenRTB 2.x BidRequest that agents receive [216]. None of its eight mutation intents is specific to geography. An ARTF agent can read device.geo and could activate a location-derived audience segment by segment ID, but it has no standard intent to change or add a geographic field [216].

### The baseline is further along

The non-agentic baseline opened here is further along than the agentic evidence screened [219, 220]. OpenRTB 2.6 defines a Geo object (lat, lon, type, accuracy, lastfix, ipservice, country, region, metro, city, zip, utcoffset), DOOH venue types and a qty impression multiplier [219, 220]. Google Authorized Buyers documents production bid requests in which geo type is always IP (2) and accuracy reflects coarse IP-based geolocation. That is documented production use of coarse, inferred location, not device-observed location ([Chapter 1](#defining), [Chapter 2](#supply)) [219, 220]. It is one exchange's documentation of its own system, and other exchanges may pass GPS fixes [219, 220].

### Vendors' agent interfaces sit outside buying

Among location vendors, this review found agent interfaces only outside agent-driven buying [217, 218, 222, 200]. Foursquare publishes an MCP server, created in January 2025 and last pushed in April 2025, whose tools search places and snap coordinates to Foursquare place IDs. Mapbox publishes an MCP server (v0.16.0, 15 October 2026, with a hosted endpoint) for geocoding, point-of-interest search and isochrones [217, 218, 222, 200]. Radar runs two MCP servers. Its public documentation endpoint serves search and retrieval, and an authenticated server announced on 14 October 2026, in pilot for enterprise customers by request, gives AI assistants read-only tools to query a customer's own location records. Neither is an advertising-transaction interface [222, 223, 404]. Placer lists a remote MCP server, "Pi for Placer MCP", as a per-account enterprise entitlement among its data delivery routes; its documentation page was last updated on 22 October 2026, and its launch date is not established [280, 281, 282, 224].

For GroundTruth, Blis, InMarket and Vistar Media, the searches of 24 October 2026 found no MCP server or AdCP agent, and none of them appears among the 34 agents in AAO's registry [217, 218, 222, 200]. Private or unannounced agent interfaces would not appear in these sources [217, 218, 222, 200].

### H7, coded by the rule

H7 is unresolved at 24 October 2026 [198, 200, 201, 205, 598, 599]. Geographic cells reach no state. Proximity, store catchments, named places, venue identifiers, DOOH metrics and foot_traffic reach schema representation only, administrative geo reaches an implemented connector on AdCP 3.1.0-beta.3, and no interoperability test result is public [198, 200, 201, 205, 598, 599]. The supported rule needs production use or a transaction with the signal and version recorded, and the two vendor-claimed agentic AdCP campaigns record no location field, so it is not met [198, 200, 201, 205, 598, 599].

The contradicted rule needs a source documenting that a location signal was dropped, could not be carried, or failed in an agent-driven buying workflow or interoperability test, with no location signal in documented production use. The Prebid connector's refusal of postal areas is the nearest candidate [198]. A clarification dated 24 October 2026, made by the author after extraction, rules that "could not be carried" requires an observed attempted workflow or interoperability test that failed or dropped the signal, and that a connector refusing a field by design, seen only in code, does not meet it. No attempted workflow was observed, so the clarification leaves the verdict unresolved [198, 200, 201, 205, 598, 599]. Stream J read the refusal the same way, as a declared capability limit rather than a signal dropped or failed in a workflow [198, 200, 201, 205, 598, 599].

The rival framing is registered with H7: frequency and reach should be governed by place on a hierarchical cell grid. It comes from the author's essay and from #5969, and it is a design proposal [205, 206, 207]. The protocol-state evidence neither supports nor refutes it, and it moves no state [198, 200, 201, 205, 598, 599].

### What would move the verdict

Support needs a named location signal in documented production use or a transaction in an agent-driven buy, with the protocol, version and field recorded. A source showing that the PubMatic or Broadsign campaigns passed geography or venue identifiers as AdCP fields would be the shortest route [201, 210] [198, 200, 201, 205, 598, 599]. Earlier states would show first: a passing geo or DOOH storyboard, a registered agent declaring sales_dooh, a badge in live-traffic mode, or a connector built on 3.2 that handles proximity or places [200, 597] [198]. A documented attempt in which a location signal failed or was dropped in an agent-driven buy would meet the contradicted rule, provided no location signal reached documented production use.

A buyer offered an agent-driven buy by place can ask which field carried the geography, at which protocol version, and whether a foot_traffic number came with its control design. [Chapter 12](#next) turns those into questions for sellers.

The register lists the protocol, registry and platform surfaces screened, with each one's status and the observable behind it.

**AI and agentic capability map**

| Platform / product | Capability | Status | Date | Independently observable? | What the record supports | Src |
|---|---|---|---|---|---|---|
| AdCP 3.2.6 | Administrative geo targeting fields: geo_countries, geo_regions, geo_metros, geo_postal_areas (with exclusions) | live | 2026-10-23 | Schema representation at 3.2.6; no public interoperability test result (geo storyboards untested on the two agents listing them); no field-level production use | PubMatic (December 2025; protocol version not recorded) and Broadsign (May 2026) describe geo- and venue-targeted agentic campaigns without naming a field, version or delivery metric | [197, 221, 200, 201] |
| Prebid Sales Agent v2.0.0 (AdCP 3.1.0-beta.3) | Connector mapping geo_countries, geo_regions and Nielsen DMA geo_metros into Google Ad Manager targeting; postal codes rejected; no proximity, catchment or place handling | live | 2026-07-01 | Implemented connector in open-source code; built on the adcp SDK 5.7.0, which targets AdCP 3.1.0-beta.3, not 3.2.6 | The only implemented AdCP geo connector found; no deployment result with a geo field is public | [198, 199] |
| AdCP 3.2.6 | Point and polygon targeting: geo_proximity (radius, travel time, GeoJSON geometry) and store_catchments | live | 2026-10-23 | Schema representation only; absent from the Prebid GAM targeting manager |  | [197, 198] |
| AdCP 3.2.6 | Named places: geo_places and geo_places_exclude, keyed by geonames, Google Ads, Microsoft Ads or an owner URI | live | 2026-08-17 | Schema representation only; geo_place_targeting storyboard untested on the two agents listing it | Place types are administrative units plus airport, not store-level points of interest; place-level forecasting and delivery breakdowns deferred | [202, 203, 200, 204] |
| AdCP 3.2.6 and 3.3.0-beta.0 | Geographic cells (geo_cells) | announced | 2026-07-21 | No state: absent from the schema at 3.2.6 and 3.3.0-beta.0 | Proposed by the author (issue #5969) in a working group he co-leads: a declared interest, not evidence | [205, 206, 207] |
| AdCP 3.2.6 | Venue identifiers: venue_id, screen_id, openooh_venue_type, geopath:<id>; OOH panel id_type geopath | live | 2026-10-23 | Schema representation only; no registered agent declares the sales-dooh specialism | A vendor describes venue targeting through AdCP without naming a field | [208, 209, 200, 210] |
| AdCP 3.2.6 | Visit metric: foot_traffic with an attribution_methodology qualifier | live | 2026-10-23 | Schema representation only | No field declares the control design; 'modeled' covers MMM, MTA and incrementality testing alike | [211, 212] |
| AdCP 3.2.6 | DOOH delivery metrics: plays, dooh_metrics (loop_plays, screens_used, venue_breakdown), measurement_source | live | 2026-10-23 | Schema representation only; sales-dooh specialism defined, none declared |  | [211, 209, 200] |
| AAMP OpenProposal 3.0 draft 1 | Targeting envelope geo.allowed[] of geo codes | beta | 2026-09-22 | Schema representation in a draft only | No postal, DMA, lat/lon or venue terms in the document | [213] |
| AAMP iab-agentic-primitives 0.5.1 | Package.geo_targets (ISO 3166-2); linear-TV QuoteRequest DMA codes | beta | 2026-08-10 | Schema representation only | The README calls the library a pre-release | [214] |
| AAMP seller-agent 2.4.2 | Media kit geo_targets; GAM model geoTargeting (unmapped) | live | 2026-08-17 | Schema representation only; nothing writes geo_targets into GAM or FreeWheel targeting | Tag commit 2026-08-17; GitHub release 2026-08-18 | [215] |
| ARTF 1.0 | Geo fields inherited from the embedded OpenRTB 2.x BidRequest (device.geo, user.geo) | live | 2025-11-12 | Schema representation, inherited; no geo-specific mutation intent | ACTIVATE_SEGMENTS can carry a segment built from location data, though not a geo field | [216] |
| OpenRTB 2.6 (baseline, not agentic) | Geo object, DOOH venue type, Qty multiplier | live | 2026-04-01 | Documented production use: Google's exchange documents type=2 (IP), coarse geolocation in its bid requests | Comparator for the agentic protocols; Google documents its own exchange only | [219, 220] |
| Foursquare places MCP server | Place search and lookup tools (search_near, search_near_point, place_snap, place_details) | live | 2025-04-29 | Implemented connector outside agent-driven buying | Not a buying protocol | [217] |
| Mapbox MCP server v0.16.0 | Geocoding, POI search, isochrones (hosted MCP tools) | live | 2026-10-15 | Implemented connector outside agent-driven buying | Not a buying protocol | [218] |
| Radar MCP server 1.0.0 | Documentation search only (search_radar, query_docs_filesystem_radar, submit_feedback) | live |  | No location tools | Public documentation endpoint; Radar also announced on 14 October 2026 an authenticated MCP server, in pilot for enterprise customers, with read-only tools over a customer's own location records. Neither is a buying interface. | [222, 223] |
| Pi for Placer MCP | Enterprise MCP server over Placer.ai data (per-account entitlement); documentation updated 22 October 2026 | live | 2026-10-22 | documentation page; connector directory listings | Not a buying protocol | [224] |



## 11. The Ground Quotient


*The Ground Quotient (GQ) summarises what each location product documents in public, within its cohort. Across all five cohorts, the two weakest dimensions are the ones a buyer needs before treating a location number as fact: agreement with ground truth and independent assurance.*

> **In short.**
> 47 products are scored on 11 dimensions in five cohorts: 12 in visit measurement and attribution (A), 14 in location audiences and targeting (B), 10 in foot-traffic and visit analytics (C1), 7 in raw movement data (C2) and 4 in geofencing and visit-detection SDKs (D). The field GQ is 41 in A, 44 in B, 41 in C1, 38 in C2 and 54 in D, which is shown without bands by design. 1 product sits in the Most documented band, 18 in Well documented, 21 in Partly documented and 2 in Least documented. Independent assurance averages 1 in every cohort, and ground-truth validation is among the three lowest-scoring dimensions in A, B and C2. The GQ measures documented capability, not accuracy or performance.

### What the quotient does

Each product was scored from 0 to 5 on the dimensions that apply to its cohort, against anchors fixed before scoring and amended once, before any cell was scored, after the second outside review ([chapter 13](#method)). The quotient groups the cells into three. Depth is what the data and the number are: source transparency, the location fix and visit definition, ground-truth validation and, in cohort A only, causal design. Reach is where it works: coverage, data access, documented production volume and US availability. Standing is whether a buyer can trust and trade it: independent assurance, method disclosure and privacy and consent posture. A product's Raw score is 0.45 × Depth + 0.25 × Reach + 0.30 × Standing, each group the mean of its counted cells and renormalised if a group is missing. Cells that are n/e (no evidence) or n/a (outside the cohort's mask) are skipped, never counted as zero.

Raw is pulled toward the cohort's field mean in proportion to how thin the record is: Shrunk = (k × Raw + K × F) / (k + K), where k is the row's counted cells, F the mean Raw of the cohort's admitted rows and K, the prior strength, 4. The GQ is 100 × Shrunk / 5. K is an author-set heuristic, not a statistical estimate, and the tests below rerun the index without it. Bands come from the unrounded distance to the field mean, 20 × (Shrunk − F): Most documented at +12 or more, Well documented at +1 or more, Partly documented at −8 or more, Least documented below. A band needs at least 6 counted cells with every group represented, and a cohort shows bands only when 6 or more of its rows qualify. Cohorts A, B, C1 and C2 do. Cohort D has 4 rows and carries no bands by design.

Two rules shape the numbers and are the author's choices. Projection method, how panel visits are scaled to a population, is reported beside the matrix as a separate module and is not in the quotient: scoring it inside the GQ would have rewarded products that disclose nothing about their projections ([chapter 13](#method)). Privacy carries a risk cap: a product whose location data was the subject of a regulator order, entered judgment or confirmed breach in the three years before the cutoff scores at most 1, and its uncapped score for current controls is published beside it. Seven products carry the cap. Without it, their GQs would be 2 to 5 points higher, three would move up a band, and two uncapped products, Placer.ai among them, would move down as their cohorts' field means rose [130, 131, 132, 143, 128, 137, 142, 145].

The field mean here is the mean of the cohort's Raw scores. That differs from the author's Contextual and Native Quotients, which pooled group scores, so a GQ is not comparable with a CQ or an NQ, and a score in one cohort is not comparable with a score in another.

### The results

![Read the band, not the rank: in visit measurement and in audiences, no product clears the most documented line.](figures/fig-quotient.svg)

*Figure fig-quotient. **Read the band, not the rank: in visit measurement and in audiences, no product clears the most documented line.** The Ground Quotient for the banded products in cohorts A and B, one section per cohort; cohorts C1, C2 and D follow in the next figure, on the same axis. The Ground Quotient uses the band offsets of the author’s Contextual Quotient, but its field mean is the cohort’s mean Raw score rather than a pooled group mean, so the two do not read identically; neither construction validates it. Depth, Reach, Standing group means are combined 45/25/30 and shrunk toward the cohort’s field mean (the mean Raw score of its admitted rows) with a prior strength of 4; bands come from the unrounded distance to that field score, and only verified or qualified cells count. A band needs at least 6 counted cells with one in every applicable group, in a cohort with at least 6 such rows. Across the reweightings, positions move by up to 3 places within a cohort, and 34 of the 42 banded products in all cohorts keep their band under every weighting. GQ measures publicly documented capability within a cohort, not product performance. Units: GQ points on a 0–100 scale, pulled toward the cohort field mean; shown rounded, plotted unrounded. Scope: 26 banded products in cohorts A and B, of 47 in the capability matrix; scores as of 24 October 2026. Caveat: One initial scoring pass; 16 rows nearest a band line re-scored blind. The index inherits every limit of the dimension scores and is sensitive to the weights; the bands, not the positions, are the finding. Data: data/gq.csv*

![In foot-traffic analytics and raw movement data, one product clears the most documented line; developer tools carry no bands by design.](figures/fig-quotient-data.svg)

*Figure fig-quotient-data. **In foot-traffic analytics and raw movement data, one product clears the most documented line; developer tools carry no bands by design.** The Ground Quotient for cohorts C1, C2 and D, continuing the previous figure on the same axis and under the same rules. Cohort D is scored but not banded, by design; its rows are listed under the chart. GQ measures publicly documented capability within a cohort, not product performance. Units: GQ points on a 0–100 scale, pulled toward the cohort field mean; shown rounded, plotted unrounded. Scope: 16 banded products in cohorts C1, C2 and D, of 47 in the capability matrix; scores as of 24 October 2026. Caveat: One initial scoring pass; 16 rows nearest a band line re-scored blind. The index inherits every limit of the dimension scores and is sensitive to the weights; the bands, not the positions, are the finding. Data: data/gq.csv*

In cohort A the field GQ is 41; a product needs 42.2 for Well documented and 53.2 for Most documented. In cohort B the field is 44, with Well documented from 45.4. In cohort C1 it is 41, with Well documented from 42.4 and Most documented from 53.4. In cohort C2 it is 38. At these field means, a product scoring 5 on every applicable dimension would reach 84 in cohort A, not 100, because of the shrinkage.

The one product in the Most documented band is Placer.ai, in cohort C1, and it clears the line by less than a quarter of a point; under three of the four reweightings it is Well documented. Its band rests on documentation, not on accuracy: its US availability is backed by an independent record, and the one third-party comparison found for it, at a single trail site and with its funding not stated, put its estimate 30.5% above the counter [101]. That comparison scores 3, not 4, because a comparison with unstated funding is not independent under the rules. Placer.ai's population visit estimates are projections; it also exposes raw panel counts ([chapter 1](#defining)).

**The capability matrix** (0–5 evidence-bounded scores; n/e = insufficient evidence; n/a = outside the cohort’s mask; unverified cells are marked and counted as n/e; availability, evidence grade and verification status per cell are in data/capability-matrix-long.csv; one section per cohort; banded products ordered by GQ, the Ground Quotient, within their cohort, and unbanded rows listed A to Z)

*A · Visit measurement and attribution*

| Vendor | Product | Group | GQ (band) | Sources | Definition | Validation | Causal | Coverage | Access | Volume | US | Assurance | Disclosure | Privacy | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cuebiq | Cuebiq Footfall Measurement | specialist | 49 (Well documented) | 3 | 3 | 1 | 2 | 1 | 3 | 1 | 4 | n/e | 3 | 4 | 10/11 |
| Azira | Azira Measurement | specialist | 47 (Well documented) | 3 | 2 | 2 | 4 | n/e | 1 | 1 | 4 | n/e | 3 | 2 | 9/11 |
| GroundTruth | GroundTruth visit measurement | adtech | 45 (Well documented) | 3 | 2 | 2 | 1 | n/e | 3 | 2 | 4 | 1 | 3 | 3 | 10/11 |
| Reveal Mobile | Reveal OOH Measurement | specialist | 45 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 1 | n/e | 2 | 1 | 3 | 3 | 4 | n/e | 2 | 3 | 9/11 |
| Foursquare | Foursquare Attribution | specialist | 43 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 1 | 1 | 3 | 2 | 3 | 2 | 4 | n/e | 2 | 2 | 10/11 |
| Adsquare | Adsquare Footfall Attribution | specialist | 42 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 2 | 1 | 2 | n/e | 3 | n/e | 3 | 1 | 2 | 2 | 9/11 |
| InMarket | InMarket visit-lift measurement | specialist | 41 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 1 | 1 | 3 | 1 | 3 | 1 | 4 | n/e | 3 | 1 | 10/11 |
| Arrivalist | Arrivalist MediaInsights | specialist | 40 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 3 | 1 | 2 | 1 | 1 | 1 | 3 | n/e | 3 | 1 | 10/11 |
| Vistar Media | Vistar Media attribution | adtech | 39 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 1 | n/e | n/e | n/e | 3 | n/e | 3 | n/e | 1 | 2 | 6/11 |
| Blis | Blis Smart Holdout Groups | adtech | 36 (Partly documented) | 1 | 1 | 1 | 4 | 1 | 1 | n/e | 3 | n/e | 1 | 2 | 9/11 |
| Kochava | Kochava DOOH Advertising Measurement | adtech | 35 (Partly documented, provisional: Least documented with n/e = 1, Well documented with n/e = 4) | 1 | 1 | n/e | n/e | n/e | 1 | n/e | 3 | n/e | 2 | 2 | 6/11 |
| Precisely | Precisely PlaceIQ attribution | enterprise | 34 (Partly documented, provisional: Least documented with n/e = 1, Well documented with n/e = 4) | 3 (unconfirmed) | 1 (unconfirmed) | 1 (unconfirmed) | n/e | n/e | n/e | n/e | 1 (unconfirmed) | n/e | 1 (unconfirmed) | 2 (unconfirmed) | 6/11 |

*B · Location audiences and targeting*

| Vendor | Product | Group | GQ (band) | Sources | Definition | Validation | Causal | Coverage | Access | Volume | US | Assurance | Disclosure | Privacy | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Foursquare | Foursquare Audience | specialist | 52 (Well documented) | 3 | 3 | 1 | n/a | 4 | 4 | 1 | 4 | n/e | 2 | 4 | 9/10 |
| Cuebiq | Cuebiq Audiences | specialist | 48 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 1 | 1 | n/a | n/e | 3 | n/e | 4 | n/e | 2 | 4 | 7/10 |
| Reveal Mobile | Reveal Audiences | specialist | 48 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 2 | n/e | n/a | 3 | 3 | n/e | 4 | n/e | 2 | 3 | 7/10 |
| GroundTruth | GroundTruth Ads Manager | adtech | 47 (Well documented) | 3 | 3 | 1 | n/a | 4 | 3 | 1 | 4 | 1 | 2 | 3 | 10/10 |
| Foursquare | Foursquare Proximity | specialist | 47 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 3 | 1 | n/a | 1 | 4 | n/e | 4 | n/e | 2 | 3 | 8/10 |
| Adsquare | Adsquare ID-Precise Audiences | specialist | 46 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 3 | 1 | n/a | 3 | 4 | n/e | 3 | 1 | 2 | 3 | 9/10 |
| Arity | Arity Audiences | enterprise | 45 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 1 | n/e | n/a | n/e | n/e | 1 | 4 | n/e | 2 | 3 | 6/10 |
| Simpli.fi | Simpli.fi Addressable Geo-Fencing | adtech | 45 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 3 | 1 | n/a | 3 | 3 | n/e | 3 | n/e | 2 | 2 | 8/10 |
| Azira | Azira Audience | specialist | 43 (Partly documented) | 3 | 2 | 1 | n/a | 3 | 3 | 1 | 3 | 1 | 2 | 3 | 10/10 |
| Vistar Media | Vistar Media audience targeting | adtech | 43 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 2 | 1 | n/a | 1 | 3 | n/e | 3 | n/e | 2 | 2 | 8/10 |
| Precisely | Precisely Audiences | enterprise | 41 (Partly documented) | 3 | 1 | 1 | n/a | 3 | 3 | n/e | 4 | 1 | 1 | 2 | 9/10 |
| InMarket | InMarket Audiences | specialist | 41 (Partly documented) | 3 | 2 | 1 | n/a | 3 | 3 | 1 | 4 | 1 | 2 | 1 | 10/10 |
| Blis | Blis audiences | adtech | 39 (Partly documented) | 3 | 1 | 1 | n/a | 3 | 3 | 1 | 3 | 1 | 2 | 2 | 10/10 |
| Mobilewalla | Mobilewalla audiences | specialist | 36 (Least documented) | 3 | 1 | 1 | n/a | 1 | 3 | 1 | 4 | 1 | 1 | 1 | 10/10 |

*C1 · Foot-traffic and visit analytics*

| Vendor | Product | Group | GQ (band) | Sources | Definition | Validation | Causal | Coverage | Access | Volume | US | Assurance | Disclosure | Privacy | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Placer.ai | Placer.ai | specialist | 54 (Most documented) | 3 | 3 | 3 | n/a | 4 | 4 | 1 | 5 | 1 | 2 | 4 | 10/10 |
| Unacast | Unacast foot-traffic data | specialist | 51 (Well documented) | 3 | 3 | 3 | n/a | 4 | 4 | 2 | 4 | 1 | 3 | 1 | 10/10 |
| Advan | Advan Patterns+ | specialist | 51 (Well documented) | 3 | 2 | 3 | n/a | 3 | 4 | 1 | 5 | 1 | 3 | 3 | 10/10 |
| Foursquare | Foursquare Visits feeds | specialist | 50 (Well documented, provisional: Well documented with n/e = 1, Most documented with n/e = 4) | 3 (unconfirmed) | 3 (unconfirmed) | n/e | n/a | 2 (unconfirmed) | 4 (unconfirmed) | 1 (unconfirmed) | 3 (unconfirmed) | n/e | 2 (unconfirmed) | 3 (unconfirmed) | 8/10 |
| Mapbox | Mapbox Movement Data | enterprise | 40 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 3 | 1 | 1 | n/a | n/e | 3 | 1 | 3 | n/e | 2 | 2 | 8/10 |
| Huq | Huq Industries | specialist | 36 (Partly documented, provisional: Least documented with n/e = 1, Partly documented with n/e = 4) | 1 | 1 | 2 | n/a | n/e | 3 | n/e | 3 | 1 | 1 | 1 | 8/10 |
| Precisely | Precisely Visits | enterprise | 36 (Partly documented) | 1 | 2 | 1 | n/a | 2 | 3 | n/e | 4 | n/e | 1 | 1 | 8/10 |
| PassBy | PassBy | specialist | 34 (Partly documented, provisional: Least documented with n/e = 1, Partly documented with n/e = 4) | 1 | 1 | 1 | n/a | n/e | 3 | n/e | 3 | 1 | 1 | n/e | 7/10 |
| Zartico | Zartico | specialist | 30 (Least documented, provisional: Least documented with n/e = 1, Partly documented with n/e = 4) | 1 | 1 | 1 | n/a | n/e | 1 | 1 | 3 | n/e | 1 | n/e | 7/10 |
| Arity | Arity Geosight and Roadway Insights | enterprise | 38 (thin record) | 2 | n/e | n/e | n/a | n/e | n/e | 1 | 3 | n/e | 1 | n/e | 4/10 |

*C2 · Raw movement data*

| Vendor | Product | Group | GQ (band) | Sources | Definition | Validation | Causal | Coverage | Access | Volume | US | Assurance | Disclosure | Privacy | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Quadrant | Quadrant | specialist | 45 (Well documented) | 3 | 3 | 1 | n/a | 1 | 4 | 1 | 4 | n/e | 2 | 3 | 9/10 |
| Complementics | Complementics Location Data Feeds | specialist | 41 (Well documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 2 | n/e | n/a | 1 | n/e | 1 | 4 | n/e | 2 | 3 | 7/10 |
| Veraset | Veraset Movement | specialist | 40 (Well documented) | 3 | 2 | 0 | n/a | 2 | 4 | 1 | 4 | 1 | 2 | 3 | 10/10 |
| Unacast | Unacast device-level location data | specialist | 37 (Partly documented) | 3 | 2 | 1 | n/a | 1 | 3 | 1 | 4 | 1 | 2 | 1 | 10/10 |
| Outlogic | Outlogic Observation Panel | specialist | 37 (Partly documented) | 3 | 1 | 1 | n/a | 1 | 3 | n/e | 4 | 1 | 2 | 1 | 9/10 |
| Collective Data Solutions | Collective Data Solutions data feeds | specialist | 33 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 2 | 1 | n/a | 1 | n/e | 1 | 4 | n/e | 1 | 1 | 8/10 |
| Tamoco | Tamoco | specialist | 33 (Partly documented, provisional: Partly documented with n/e = 1, Well documented with n/e = 4) | 2 | 1 | 1 | n/a | n/e | n/e | 1 | n/e | n/e | 1 | 3 | 6/10 |

*D · Geofencing and visit-detection SDKs and APIs*

| Vendor | Product | Group | GQ (band) | Sources | Definition | Validation | Causal | Coverage | Access | Volume | US | Assurance | Disclosure | Privacy | Counted |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Bluedot | Bluedot | developer | 48 (cohort without bands) | 2 | 2 | 1 | n/a | 4 | 4 | n/e | 3 | n/e | 2 | 2 | 8/10 |
| Foursquare | Foursquare Movement SDK | developer | 59 (cohort without bands) | 3 | 4 | 2 | n/a | 3 | 4 | n/e | 3 | n/e | 3 | 3 | 8/10 |
| Infillion | Gimbal | developer | 52 (cohort without bands) | 3 | 3 | 1 | n/a | 3 | 3 | n/e | n/e | n/e | 3 | 2 | 7/10 |
| Radar | Radar | developer | 56 (cohort without bands) | 3 | 4 | 2 | n/a | 4 | 4 | 1 | 4 | 1 | 3 | 3 | 10/10 |

### Where the field is weak

Independent assurance averages 1 in every cohort, the floor for a scored cell. No product holds a current accreditation or an independent audit of its location metrics. On 24 October 2026 none of MRC's eleven lists carried a visit or location metric, and both visit data sets MRC once accredited, Foursquare's and GroundTruth's, have lost accreditation [171, 172, 173, 174, 175, 176, 177, 178, 179, 180]. The products that score anything here score it for self-attestation or for a pending or historical record, which the rules treat as a flag, not a score.

Ground-truth validation is the lowest-scoring dimension in cohort C2, level with the lowest in B, and third-lowest in A, above only independent assurance and coverage. In the sources screened, no product documents an independent comparison of its visit counts with door counts or transactions across many places [98, 99, 102]. Where a vendor publishes an accuracy figure, it publishes no places, period or held-out design, so the cell stays at the vendor-claim anchor ([chapter 4](#validation)).

Documented production volume is the third weak dimension. Panel sizes are stated in incompatible units, often without a period, and bidstream reach is not a panel, so most volume cells are low or n/e ([chapter 2](#supply)).

The strongest cells are US availability in every cohort and data access outside cohort A: vendors document where they sell and how a buyer gets the data, in detail. That is what the caps are for. A detailed help page is vendor documentation, so it earns a 4 at most, and only two cells in the matrix score 5, both for US availability backed by an independent record.

### How stable the bands are

The index was recomputed under four other weightings, three other prior strengths and every single-cell deletion.

Under the four reweightings (equal thirds; Depth-led 60/20/20; Reach-led 30/45/25; Standing-led 30/20/50), recomputing Raw and the field mean each time, the largest position change is 3 places in cohort A, 3 in B, 3 in C1 and 1 in C2; 34 of 42 banded products keep their band under all four.

Removing the shrinkage (K = 0, which reports Raw) changes 3 bands in cohort A, 1 in B, 6 in C1 and 1 in C2.

Deleting one counted cell at a time, with every other row and the field mean held fixed, changes a band in 33 of 104 cases in cohort A and 26 of 121 in B. Leaving out a low score raises a quotient: missing evidence is not neutral. The largest single-cell effect in cohort A comes from deleting Arrivalist's privacy and consent posture cell, a 1, which would raise its GQ from 40 to 44. That cell is the risk cap.

The missing-evidence scenarios test the opposite case. Each row's n/e cells are filled with 1, then 4, one row at a time with every other row and the field mean frozen. These are scenarios, not bounds. 23 products change band under one of them and are marked ± in the matrix as provisional. The rows nearest a band line were also scored a second time, blind ([chapter 13](#method)); a band that the second scorer's scores would move is marked † in the matrix. Rows so marked: 1.

### What the quotient is not

The GQ says how much of a product's capability is documented in public, and how well. It does not say whose visit counts are most accurate, which is the question chapter 4 could not answer from public evidence, and it does not say whose lift is causal. It rests on one initial scoring pass under published rules, a cell-by-cell verification of every counted cell by a separate AI verifier, and a blind second pass on the rows nearest a band line. No human second scorer was used.

### The profiles

The paper profiles 8 companies in depth: five whose main row is in cohort A (Foursquare, InMarket, GroundTruth, Blis and Vistar Media), two in cohort C1 (Placer.ai and Unacast) and one in cohort D (Radar). Each profile has a SWOT in which every line cites its sources and is tagged as record, vendor documentation, vendor claim or inference.

### Blis (active; acquired (completed 3 March 2025))

*Blis Holdco Limited (England and Wales, company 13974159; formerly Project Balloon Bidco Limited, renamed 8 January 2024), holding company of Blis Group Limited and its trading subsidiaries; principal trading company and data controller Blis Global Ltd (company 06455773; formerly Blis Media Limited 2012-2019 and Breeze Tech (UK) Ltd 2007-2012)*

One of two holdout lift designs in cohort A, geographic and not documented as random; owned by T-Mobile since 2025.

**SWOT (2026-10-24).** Blis, owned by T-Mobile since March 2025, sells a location-led DSP whose zip-code holdout measurement withholds ads by design, but it publishes no visit definition, sample rules or independent validation.

- **Strengths.** Smart Holdout Groups withholds ads in control zip codes with buffer zones, a geo holdout rather than an exposed-only or synthetic comparison. (vendor documentation) [19, 123, 124] T-Mobile US completed the acquisition on 3 March 2025, transferring US$180 million (US$174 million consideration at fair value; the rest settled preexisting relationships); Blis became wholly owned. (record) [36, 225] Audited accounts show GBP 82.9m turnover in the year to June 2024, GBP 37.9m in the Americas, with an operating loss of GBP 7.7m. (record) [226] The only omnichannel DSP, by its own claim, combining T-Mobile's US app data with movement and transaction data, live since 10 September 2025. (vendor claim) [227, 228] The geo design measures DOOH, audio and CTV, where device-level controls are hard to build. (vendor documentation) [19, 229]
- **Weaknesses.** No visit definition, footfall data source, minimum sample, significance rule or uncertainty is published for footfall lift. (vendor documentation) [19, 123, 124, 229] Control areas come from matched pairs; Blis never states that designation within a pair is random. (vendor documentation) [123] Not on MRC's accredited or under-review digital lists; no independent audit of the holdout method found. (record) [171] Last panel figure is 370m 'opted-in' devices, global and undefined, from August 2021; no current US panel. (vendor claim) [230] The only third-party location-data verification found (2019) was a commercial engagement, an 18-month deal, and the verifier later alleged data manipulation, which Blis denied. (record) [231, 232, 233]
- **Opportunities.** UK app-usage lift on mobile-network data is offered; a page last modified April 2025 says 'coming soon' to the US, where the parent is a carrier. (inference) [19, 36] With Vistar Media under the same parent since February 2025, a shared geo holdout for DOOH buys is a plausible next step. (inference) [234, 36] Publishing the pair-assignment rule, including whether designation is random, and sample thresholds would let buyers judge whether lift reads as causal; Blis claims independent verifiability. (inference) [229, 123]
- **Threats.** Registered California data broker declaring precise-geolocation collection; DROP deletion requests must be processed from 1 August 2026. (record) [163] Its privacy notice lists precise GPS data and names SSPs and exchanges among its sources, a route state precise-geolocation laws are narrowing; the share is not stated. (inference) [235] A listed verifier's 2021 regulatory announcement still records an unadjudicated data-manipulation allegation against Blis. (record) [232] Parent is a large US advertiser and data supplier; with no independent audit, buyers may question neutrality of in-platform lift. (inference) [234, 227, 171]

- **What is sold.** An omnichannel demand-side platform built around geography rather than identifiers: planning (Audience Explorer, launched 2021, and the Blis AI planning assistant, June 2025), buying (Dynamic Audience Targeting, managed or self-service, on display, native, audio, online video, CTV and DOOH inventory) and measurement (Smart Holdout Groups: brand lift, footfall lift in the US only, app-usage lift in the UK only, sales lift and custom lift, all computed by comparing machine-balanced target and control zip codes or postal districts). Data layered in the platform includes T-Mobile Advertising Solutions' US app ownership and app engagement data (since 10 September 2025), other telco data (UK and Europe), movement data and transaction data.
- **Who buys it.** Advertisers and media agencies (Blis: 'working with major agencies and their brands'; operates in 'more than 40 markets'). The buyer pays for media bought through the DSP, with measurement attached to Blis-bought campaigns. Named clients in vendor material are not reproduced (paper rule). Its parent is itself a large US advertiser; any conflict that creates is a reading, not a documented issue.
- **How it earns.** Media buying through the DSP. Blis says 'We do not mark up media' and its data fees are 'capped to not scale up for premium content' (vendor). The audited accounts recognise revenue 'on a monthly basis, based upon the level of impressions generated from each individual client campaign' and do not say whether turnover is gross media or net fees. Smart Holdout Groups pricing is not published; a free trial for a limited time was offered in the UK in October 2024.
- **Supply and data dependencies.** Location observations: the company privacy notice lists precise GPS-sourced latitude/longitude and IP-derived geolocation, historic location profiles and residential-IP household linking, received from Blis's own cookie or from 'data partners, supply side platforms (SSPs), exchanges and clients' (so bid requests are a documented route); the product pages instead speak of 'first-party movement data' without saying how it is collected (the privacy centre lists SDKs among its tracking technologies without naming them or tying them to location). Precision class: precise inputs (GPS; California registry: collects 'precise geolocation') activated as 'approximate geography' at postal-district level. Consent route: IAB TCF, GPP and IAB CCPA Framework participation; sensitive data processed only where Blis, clients or data providers obtained consent; no product-level consent description. Other inputs: T-Mobile app data (US), telco partnerships in North America, the UK and Europe (the technology page names telco 'location history' among telco-data inputs, unquantified), Vodafone quadkey data, census data, transactions; location-attribution vendors are listed among 'Ad Effectiveness & insights' partners (Cuebiq, Foursquare, InMarket, Kochava, Lifesight, Samba TV and others) without saying which, if any, supplies Smart Holdout Groups' footfall outcomes. Disclosure: Samba TV, listed among those partners, is the author's employer.
- **Differentiation.** Measurement by geographic holdout: ads are withheld in control zip codes with buffer zones, rather than a matched or synthetic device control, and the same design is applied to DOOH, audio and CTV where device IDs are absent (LOC-BLIS-S07, S08, S09). Unit of balancing is the zip code or postal district rather than the DMA or state, which Blis contrasts with earlier geo tests (LOC-BLIS-S09). Owned by a US carrier and documented as the DSP where T-Mobile's US app ownership and engagement data is activated, a telco input most location vendors screened do not show (LOC-BLIS-S01, S14, S16). Filed, audited group accounts give a revenue scale (GBP 82.9m, year to June 2024) that most private location vendors do not disclose (LOC-BLIS-S04). Registered as a California data broker with an explicit 'Yes' to collecting precise geolocation, a clearer public statement of precision than its marketing's 'geography, not identity' framing (LOC-BLIS-S13, S10).
- **Risks.** Regulatory record (dated): no FTC order, state attorney-general action or UK ICO action against Blis was found in web searches logged on 24 October 2026 (not direct searches of the enforcement registers; bounded, not proof of absence); the only adverse public record found is Location Sciences' 2020-2021 allegation, which Blis denied. Blis Global Ltd is on California's 2026 data broker registry (last updated 29 July 2026) declaring precise-geolocation collection, which brings it under the Delete Act's DROP mechanism: data brokers must process DROP deletion requests from 1 August 2026 (LOC-BLIS-S13). Bidstream location dependence (inference): the privacy notice lists precise GPS and IP-derived location among the data it receives and names SSPs and exchanges among its sources, without stating the share arriving that way; state laws restricting sale of precise geolocation (workstream H) could limit that route; no source documents an effect on Blis (LOC-BLIS-S11). Verification history: a listed verifier's regulatory announcement (18 March 2021) records its allegation that Blis manipulated data in 2020; Blis denied it; the allegation remains on the public record unadjudicated (LOC-BLIS-S24, S25). Parent conflict (inference): T-Mobile is a large US advertiser and a telco data supplier to the platform; buyers of Blis measurement have no documented independent check of the holdout results (LOC-BLIS-S05, S16, S26). Brand and integration ambiguity: the About page still says 'Independent DSP' after the March 2025 acquisition, and the measurement page was last modified in April 2025; whether the product roadmap ('clients will soon be able to select their own control factors', October 2024) shipped is undocumented (LOC-BLIS-S07, S09, S15).
- **Best fit.** US advertisers buying omnichannel media through Blis who want an in-platform geo holdout read on store visits across channels without device IDs (DOOH, audio, CTV) and accept a vendor-run, unaudited method (LOC-BLIS-S07, S09, S10). Buyers who want to activate audiences built from telco app data and coarse geography rather than device IDs, inside one DSP (LOC-BLIS-S10, S16).
- **Poor fit.** Buyers who need an accredited, independently audited or published-method visit-lift metric, or significance and uncertainty reporting (LOC-BLIS-S07, S26). Measurement of media bought outside Blis, or data buyers who need raw visit data, segments exported to other DSPs or an API (LOC-BLIS-S07, S10, S29). Uses that require a documented consent route for the location data and sensitive-place suppression at product level (LOC-BLIS-S11, S13).
- **Limits.** Smart Holdout Groups publishes no visit definition, footfall data source, minimum sample, significance rule or uncertainty for its footfall lift (LOC-BLIS-S07, S08, S09, S10). The control is algorithmically balanced and built from matched pairs of areas; random designation within pairs is not documented, so the design is a holdout but not a documented randomised experiment (LOC-BLIS-S08, S09). Whether footfall lift counts observed panel devices or projected visits is not disclosed (Q1 'not disclosed') (LOC-BLIS-S07, S08). No current US panel size; the last figure (370m 'opted-in' devices, global, undefined) is from August 2021 (LOC-BLIS-S21). Source mix for audiences is named without proportions, and the collection route of Blis's 'first-party movement data' is undocumented; bid requests are a documented route at company level (LOC-BLIS-S10, S11). No MRC accreditation or listing, no independent validation of visits or segments; the only third-party location-data verification found (2019) was a commercial engagement (an 18-month deal) and was later disputed by the verifier (LOC-BLIS-S22 to S26). Results and segments are documented only inside Blis's own platform (plus PMG Alli for activation); no API or export found (LOC-BLIS-S07, S10, S29). Footfall lift is US-only and measures campaigns bought through Blis; it is not sold as standalone measurement of other buyers' media in any source screened (LOC-BLIS-S07, S10).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | The design pages name one input: Blis's 'first-party movement data' (Our Technology) or 'Blis' movement data' (July 2024 release), used to model how people move between zip codes and to place buffer zones (S08, S10). The source of the footfall outcome itself (whose devices are counted as store visitors, collected how) is not documented on any Smart Holdout Groups page (S07, S08, S09, S10). At company level only, Blis's advertising privacy notice says its location data 'includes precise GPS-sourced latitude/longitude data and geolocation data derived from IP addresses' and that data reaches it from its own cookie or from 'data partners, supply side platforms (SSPs), exchanges and clients' (S11), and the privacy centre lists location-attribution vendors as 'Ad Effectiveness & insights' partners that may verify 'whether a user visited a store after seeing an ad' (S12). No proportions, no per-source precision class, and no statement tying any of these to the product's footfall metric. [123, 229, 235, 342] |
| Location fix and visit definition | live | C | Not documented for the product: no polygon or radius type, dwell rule, accuracy threshold, de-duplication rule or attribution window was found on the product page, the two launch releases or the technology page (S07, S08, S09, S10). The units of the design are zip codes ('zip code or postal district level', S09), not places. The last public description of location-data filtering is from 2021 marketing ('Smart Pin' discards 'about 80% of the raw data', and a 'Smart Places' database of 'over 100 million global locations', S21), which predates the product and is not tied to it. [19, 123, 124, 229, 230] |
| Ground-truth validation | live | D | No comparison of the product's footfall-lift or visit numbers with door counts, transactions or official visitor statistics was found. The only third-party check of Blis location data found is Location Sciences' 2019 'Verify' test (a '99%' score, no method; S22), which was commissioned (Location Sciences described an 18-month deal and Blis as one of Verify's key customers, S23, S24), predates the product, concerns location-signal quality rather than visit counts, and was followed in 2020 by Location Sciences' public allegation of 'data manipulation', which Blis denied (S24, S25). A 2026 Blis-reported study with Circana retail sales data measured incremental sales for one advertiser without saying whether Smart Holdout Groups was the design (S31); it is an outcome source, not a validation of visit counts. Coded: accuracy claims only ('unprecedented accuracy', S07). [19, 343, 231, 232, 233, 344] |
| Causal design | live | C | Default design is a geographic holdout at zip-code (US) or postal-district level: an algorithm scores each audience area on 'seven independent data points, including demographics, household characteristics, and income', 'matches pairs of similar areas, with one designated for the control group and one for the targeted group', and inserts buffer zones 'where ads are not served or measured' to limit cross-over from commuting (S07, S08, S09, S10). Ads are withheld in control zip codes, but the documents describe algorithmic balancing and matched pairs and never state that the designation within a pair is random, so it is coded as a matched-pair geo holdout: reported lifts are matched-control comparisons, correlational under this paper's rules. No minimum number of zip codes or exposures, significance rule, confidence interval or uncertainty is documented ('statistically robust' and 'Statistical reliability and accuracy' asserted, S07, S09). October 2024: 'clients will soon be able to select their own control factors' (S09); not confirmed as live. [19, 123, 124, 229] |
| Coverage | live | D | No US place base is documented for the footfall metric (no count of US places, polygons or chains measured). The design operates over zip codes; how many US zip codes or stores a study can cover is not stated (S07, S09). The 2021 'Smart Places' figure (over 100 million global locations in over 75 markets, S21) is global, pre-2024, undefined as to polygons, and not tied to the product. [19, 124, 230] |
| Data access and integrations | live | C | Results are produced inside the Blis platform ('at the click of a button', S07); the platform is offered 'either as a managed or a self-service platform' (S10). No API, log-level export, clean-room route, MMP or ad-server integration for measurement results was found in the sources screened. Measurement is described as working across Blis-bought channels and also 'non-programmatic channels, such as linear TV and OOH' (S10), without a documented ingestion route for non-Blis exposure data. [19, 229] |
| Documented production volume | live | D | No production volume is disclosed for the product: no campaigns or studies measured per year, and no current US panel. The only panel figure found is the 2021 press kit's '370m opted-in global planning and measurement panel' (global, no period or definition, dated before 2024, S21). Parent-level context: Blis Holdco's audited turnover for the year to 30 June 2024 was £82.9m, of which £37.9m 'AMERICAS' (S04); revenue is not split by product. [226, 230] |
| US availability and coverage | live | D | Footfall lift is stated as 'Only available in the US' (S07); the July 2024 release says 'The offering is now available in the US' and describes one US campaign measured with it (S08). US observation source for the footfall metric is not documented; US production use rests on vendor releases only. The app-usage-lift variant is UK-only ('Coming soon to the US!', S07). [19, 123] |
| Independent assurance | live | A | No accreditation. Blis does not appear (whole-word search) on MRC's Accredited Services: Digital list, its Services Under Review: Digital list or the accreditation landing page on 24 October 2026 (S26). Blis's own statement that 'All measurement is done transparently and can be independently verified' (S10) names no auditor and is self-attestation. No independent audit of the holdout design was found. [171, 229] |
| Method disclosure | live | C | Of the seven elements, the public record carries one in summary and one partly: (6) control design is summarised (seven balancing variables, matched pairs, buffer zones, zip-code unit; S08, S09, S10); (1) sources are named only at company level (precise GPS and IP-derived location via SSPs, exchanges, data partners, clients; S11), not for the product. Not published: (2) visit definition, (3) panel size and consent route, (4) projection, (5) validation statistics, (7) update cadence and uncertainty. No white paper, methodology document or patent publication was found (searches logged). [123, 124, 229, 235] |
| Privacy and consent posture | live | A | Company-level only. Blis Global Ltd's advertising privacy notice (effective 17 November 2025) names location data ('precise GPS-sourced latitude/longitude data and geolocation data derived from IP addresses', historic location profiles, residential-IP household linking), its sources (own cookie, data partners, SSPs, exchanges, clients) and purposes including 'measuring footfall attribution'; it states deletion or anonymisation 'within 13 months', participation in IAB Europe TCF, GPP and the IAB CCPA Framework, and that sensitive personal data is processed only where 'we, our clients or data providers' obtained consent (S11). Blis Global Ltd is on California's 2026 data broker registry and answers 'Yes' to 'collects consumers' precise geolocation' (S13). The notice also states Blis does 'not knowingly collect, store, share or use health data' or other special-category data and refers to a Content and Advertising Standards Policy whose link returned HTTP 404 on 24 October 2026. No product-specific consent route and no sensitive-place suppression (health, worship, shelters) is documented. Not found on the NAI member page (S27). No FTC, state attorney-general or UK ICO action against Blis was found in web searches logged on 24 October 2026 (the enforcement registers were not searched directly; bounded, not an absence proof). [235, 163, 345] |

Performance evidence: Smart Holdout Groups launch case (US, 2024): 'a total store visit uplift of 67% and a 160% increase in likelihood to buy across all campaign channels' (five channels per the product page). (vendor assertion, sponsor Blis (vendor release; the agency is quoted)) [123, 19]; US restaurant case (published 12 September 2019): '55% Store visit uplift (control vs exposed)', a 0.76% store-visit rate against a 'benchmark 0.5%', and 36% of visits on the day of exposure. (vendor assertion, sponsor Blis (vendor case study)) [13]; UK case on the product page: '56% YoY uplift in monetizable views' across CTV, online video and display. (vendor assertion, sponsor Blis) [19]; Location Sciences' 2019 Verify test: Blis's data 'achieved near perfect scores (99%) in our tests'. (commissioned measurement, sponsor Blis (Location Sciences described an 18-month deal with Blis and Blis as one of Verify's key customers)) [343, 231, 232, 233]; 'Independent tests show our geo-driven campaigns can double performance and halve costs'. (vendor assertion, sponsor Blis) [229]; 'Compared with leading DSPs, Blis has driven 1.5x more iOS conversions, 29% lower CPA and higher incremental sales, and 61% more conversions across all channels.' (vendor assertion, sponsor Blis) [346]; Sales-lift study with Circana retail sales data for a confectionery advertiser: every channel (DOOH, online video, audio) showed positive sales uplift and the combination the largest, per Blis's summary of a WARC article. (vendor assertion, sponsor unknown (Blis and Circana; funding not stated)) [344]

### Foursquare (acquired by Infillion (announced 18 September 2026; completion reported by participants; closing date unverified))

*Foursquare Labs, Inc.*

The first location data set MRC accredited (2020), revoked in 2022; lift rests on a modelled control; now part of Infillion.

**SWOT (2026-10-24).** Foursquare sells measurement, targeting, visit data and an SDK on one documented visit pipeline, but its lift is model-controlled, its MRC accreditation was revoked in 2022, and Infillion's acquisition, announced 18 September 2026, is reported complete by participants with no verified closing date.

- **Strengths.** Publishes the most detailed attribution pipeline seen in this review: tenure filters, block-group census weights, place weights, matched twin cohorts and impression-based projection. (vendor documentation) [236, 120] Audience activates through 26 listed platforms spanning DSPs, social, CTV, audio and DOOH, with self-serve CPM pricing published (US$1.50 Audience, US$0.75 Proximity). (vendor documentation) [237, 238] Research Visits Feed documents a US first-party visit panel with SAG normalization weights, a two-minute dwell rule and history from July 2020. (vendor documentation) [239, 240, 241, 242] NAI precise-location Enhanced Standards signatory (March 2025 list) with a published 13-category sensitive-place exclusion list. (record) [243, 244]
- **Weaknesses.** Attribution's default control is modelled or matched; no randomised or holdout option is documented, so its 'incremental lift' is not causal under this paper's rules. (vendor documentation) [119, 120, 236] MRC revoked the Visits data set's accreditation in December 2022 after an October suspension; Attribution was never audited or accredited. (record) [167, 245] No numeric minimum sample, significance rule, accuracy threshold or detection error is published; reports release when 'statistically viable', undefined. (vendor documentation) [246, 247, 248, 249] Device figures change definition each time: 100M US monthly devices (2019), 25M opted-in panel (2020), 250M US devices of aggregated coverage (2026). (record) [30, 38, 52]
- **Opportunities.** Infillion pairs Foursquare visits with Catalina purchase data in one platform, which could support sales-and-visit measurement at scale as the businesses integrate. (inference) [52, 51] No location lift metric appears on MRC's Digital accredited list at 24 October 2026 (Out-of-Home list unreadable); a re-accredited visit metric would be rare in the market. (inference) [171, 167] Proximity's documented NY, WA and CT radius rule shows Foursquare can productise state-law compliance, a selling point as more states restrict precise location. (inference) [237, 250]
- **Threats.** Infillion is evaluating Swarm and Superlocal's future after City Guide's December 2024 retirement; these apps supply Foursquare's first-party location and check-in ground truth. (inference) [52, 64, 236] AdExchanger, citing Axios, reports Foursquare will sell to third parties 'for now, at least'; Infillion said the same of Catalina, whose data would eventually become exclusive. (record) [51] Foursquare already attributes raw panel step changes to OS updates such as iOS 13; further permission changes could shrink supply without warning. (vendor claim) [240] California registry lists Foursquare as a data broker collecting precise geolocation, exposing it to Delete Act deletion requests processed from August 2026. (record) [251]

- **What is sold.** Four buyer-facing lines on location data, plus place data. (1) Attribution: store-visit and sales-impact measurement of campaigns across digital, social, audio, linear and connected TV and OOH, reporting projected visits, 'incremental' or 'behavioral' lift against a modelled control, and sales metrics from card transactions (S19, S21, S23). (2) Audience: visitation-based segments for activation in 26 listed platforms (S30, S34). (3) Proximity: real-time radius geofences around Foursquare POIs delivered to DSPs (S31, S32). (4) Visits data feeds: a US-only, first-party Research Visits Feed with normalization weights for investors and researchers, a Standard Visits Feed of chain or category visits, and device-level Offline Conversions for partner attribution (S38, S39, S43); the documentation is live but no product named Visits appears on the 2026 menu (S12, S36). (5) Movement SDK for app developers (S45). Listed, not scored: Places API and Places POI data (100M+ POIs; open-source and Pro/Premium tiers), the Geospatial Intelligence Platform (Spatial Workbench, H3 Hub, Desktop, Studio), Business Listings, and the Swarm and Superlocal consumer apps (S12, S44).
- **Who buys it.** Advertisers, agencies and media sellers measuring store visits (Attribution; Infillion says Foursquare 'serves all agency holding companies' and measures for 'hundreds of brands', S11); advertisers buying through DSPs, social, CTV, audio and DOOH platforms (Audience, Proximity; S34); hedge funds, consultancies, asset managers, government and academic researchers (Research Visits Feed, S38); brands and agency data platforms (Standard Visits Feed, S39); app developers (Movement SDK, S45). No buyer is named here.
- **How it earns.** Self-serve CPM data fees for targeting: Audience US$1.50 CPM and Proximity US$0.75 CPM 'in most scenarios', with a 2M impressions a month minimum for managed service (S33). Attribution, Visits feeds, Places and SDK pricing are not published in the sources opened (Attribution report tiers are sold through sales, S21). Revenue is not disclosed in any filing; the only revenue figure found is the parties' 2020 merger claim of more than US$150M combined (S17).
- **Supply and data dependencies.** Location observations: opt-in precise and background location from Foursquare's own apps (now Swarm and Superlocal; City Guide retired December 2024) and data licensed from SDK and API partners (S08, S18); third-party GPS data calibrated against overlapping first-party users (S23). The Research feed is first-party only; the Standard feed mixes first- and third-party data (S38, S39). Movement SDK stops send lat/long, Wi-Fi and beacon scans and AdID to Foursquare (S48). Consent route: OS location permission and in-app opt-in for Foursquare's apps; for partner SDKs the end-user consent sits with the partner app and is not audited publicly. Precision class: precise (GPS-level) inputs snapped to venues; no numeric accuracy threshold is published. Bidstream: not named as a visit source in any page opened; Proximity delivery relies on the buying platform matching 'the lat/long of the impression' to the fence (S31). Other inputs: card transactions (2B+ a year, 450+ US chains), TV viewership data joined to ad run-logs, partner exposure logs (S23). Places: crowd-sourced and verified POI data (S44). Key dependencies: the consumer apps whose future Infillion is reviewing (S11), partner SDK supply, and Apple and Google permission rules (Foursquare names iOS 13 among causes of panel step changes, S40).
- **Differentiation.** First location data set MRC accredited under its Location-Based Advertising Measurement Guidelines (July 2020; revoked December 2022); MRC accredited GroundTruth's visitation data sets in July 2021 (announced in August) (S01, S02, S05). Visit detection by a snap-to-place model trained on human check-ins, rather than radius or polygon fences, plus a large global POI base (100M+ POIs) shared across products (S21, S23, S44). Most detailed public description of an attribution pipeline among the cohort A documents seen by this stream: tenure filters, block-group census weights, place weights, matched twin cohorts, multi-touch credit and impression-based projection (S23). Sales Impact layers card transactions (2B+ a year, 450+ US chains) on visits in the same report (S21, S23). Products in all four cohorts plus open-source and licensed place data, and wide activation (26 platforms for Audience, including DOOH) (S34, S12).
- **Risks.** Ownership: Infillion announced the acquisition on 18 September 2026 and completion is reported by participants; closing date unverified; leadership joins Infillion after closing; Foursquare to keep selling to third-party customers 'for now, at least' per AdExchanger citing Axios (S11, S13). Supply: City Guide retired 15 December 2024; Infillion is evaluating the future of Swarm and Superlocal, the first-party sources of check-ins and opt-in location (S18, S11). Assurance: MRC suspended (October 2022) and revoked (December 2022) the Visits data set's accreditation, with no published reason or re-accreditation (S02). Regulatory record: no FTC or state attorney-general action found by 2026-10-24; Proskin v. Foursquare Labs (D. Mass.), a putative class action alleging sale of Movement SDK location data without consent, filed 2023-07-24 and voluntarily dismissed without prejudice by the plaintiff on 2023-10-26 before any ruling (S53). State law: Foursquare already splits national geofences for New York, Washington and Connecticut with a 3,000-foot minimum radius and issues a Washington and Nevada consumer-health-data notice (S34, S10); more state restrictions on precise location would bear on Proximity and Visits. Platform permissions: Foursquare attributes panel step changes to OS updates such as iOS 13, without quantifying them (S40).
- **Best fit.** Buyers who want one vendor to target, geofence and measure store visits across many channels, and who accept a modelled control and projected visits (S19, S34). Analysts who need a US first-party visit panel with documented weights and a three-year history at venue level (Research Visits Feed), and who will treat it as a projected index, not a count (S38, S40). App developers who need venue-level visit detection and geofence events with marketing-platform integrations (S45, S50).
- **Poor fit.** Buyers who need causal lift with a randomised or holdout design by default, or published significance rules (S19, S24). Buyers who require a currently MRC-accredited visit or lift metric (S02, S03). Buyers who need disclosed source proportions or a stable, defined panel size over time (S11, S16, S17, S37).
- **Limits.** Visit counts in Attribution and the Research feed are projected (census and place weights, SAG scores), not observed, and no per-estimate uncertainty is published (S23, S40). Attribution's default control is modelled or matched; no randomised or holdout option is documented, so its lift is not causal evidence under this paper's rules (S19, S22, S23). No numeric minimum sample, significance rule or definition of 'Behavioral Lift' is published; release is gated on an undefined 'statistically viable' test (S20, S24, S25). No published accuracy threshold or detection error statistic for visits; Foursquare acknowledges 'inaccurate snaps' without a rate (S41, S46). MRC accreditation of the Visits data set was revoked in December 2022 and nothing has replaced it; Attribution was never accredited (S02, S05). Source proportions (first-party apps, partner SDKs, other third parties) are not disclosed for any product, and device figures change definition each time they are published (S11, S12, S16, S17). Proximity's delivery depends on the impression coordinate the buying platform holds, whose accuracy Foursquare does not document (S31). Ownership has changed: Infillion's acquisition was announced on 18 September 2026 and completion is reported by participants, with no verified closing date, and the consumer apps that supply first-party data are under review (S11).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Foursquare says it processes 'GPS pings from a combination of 1st party and carefully vetted 3rd party sources' and applies its models to generate visits (S21, S23). The 2024 method post names the inputs: GPS signals from 'our own consumer apps and vetted third-party partners', with each third-party provider calibrated by 'comparing the pings of overlapping users with our first-party quality data' (S23); the 2022 post lists Bluetooth, beacon, Wi-Fi, compass, accelerometer, barometric and GPS signals (S22). Exposure comes from partners by pixel or SFTP/S3, is inferred for TV by joining viewership data with ad run-logs, and for OOH by joining Foursquare movement data with billboard locations (S23); TV viewing-data suppliers are not named. Transactions (Sales Impact) come from 'leading transaction data providers', over 2 billion a year, covering 450+ US chains (S23). The consumer privacy policy adds that SDK/API partners 'may license data to us' (S08). No proportions by source and no precision class per source are published; the bidstream is not named as a source in any page opened. [28, 120, 236, 244] |
| Location fix and visit definition | live | C | Visits are detected by 'Stop Detection' (clusters of GPS signals indicating dwell) and a 'Snap-to-Place' model trained on check-ins from Foursquare's own apps, which Foursquare contrasts with a proximity radius it calls 'inherently flawed' (S23, S21). No numeric dwell rule, horizontal-accuracy threshold, de-duplication rule or detection error statistic is published in the Attribution pages opened; the precision class is not stated numerically. Attribution rules: an advertiser-defined conversion window 'from a few days to a few weeks' (S23), a documented example of 14 days plus 7 days of reconciliation (S20), and equal fractional ('even-weighting, multi-touch') credit across impressions inside the window (S22, S23). Visit-definition details documented for the separate Visits data feeds (S37) are not inherited here. [246, 28, 120, 236] |
| Ground-truth validation | live | D | The only validation described for Attribution is Foursquare's own trend check: scaled transaction trends are compared with 'data in public financial statements' and visitation-panel trends with transaction trends (S23); no places, period, sample or statistic is published. The product page's 'Third-party verified' Forrester item (S21) is a Forrester Consulting Total Economic Impact ROI study commissioned by Foursquare (June 2022), not a validation of visit counts or lift. Foursquare's 2020 release says Attribution 'has not been audited or accredited by the MRC' (S05). This review found no independent comparison of Foursquare Attribution's visit counts or lift with door counts, transactions or official visitor statistics in the sources screened by 2026-10-24. [236, 28, 245] |
| Causal design | live | C | Default design: a modelled or matched control. Current docs: 'We create a modeled control group'; the 'expected visit rate' uses '10+ core attributes, such as age, gender, household income, past visitation behavior, and geography' (S19, updated 2026-02-12). 2022 method: regression adjustment building 'a synthetic control group based on a collection of static features', then a difference-in-differences step so 'lift in the pre-exposure time frame is zero' (S22). 2024 method: each exposed user gets 'a modeled twin cohort' from unexposed users matched on demographics, geography and past chain and category visitation (S23). No randomised, ghost-ad, PSA or geo-holdout option appears in the documentation opened. Minimum samples: a feasibility score (Excellent, Good, Consult Foursquare) computed from estimated impressions, with no numeric threshold (S24); reports appear 'once the results are statistically viable' (S20); no significance rule or uncertainty is published. The output is called 'incremental lift' on the docs page (S19) and 'Behavioral Lift' in reports (also 'our proprietary visitation behavioral lift', S23), which no opened page defines; Foursquare's line 'we cannot offer a direct view into the calculation' (S25) refers to the star ratings that compare partners' metrics, not to the Behavioral Lift calculation. Under this paper's rules a matched or synthetic control is not causal evidence. [119, 120, 236, 247, 246, 248] |
| Coverage | live | C | No US place count for Attribution is published in the sources opened. Global figures: '100M+ POIs across 200+ countries' (S12) and 'Over 100 million POIs across 247 countries and territories' (S44); '83 B+ Visits per year' (S21, no geography or definition). Measurement geography is set as National, Regions, States or DMAs, one country per campaign; 'store level reporting is currently unavailable' in the feasibility calculator (S24), while the docs page lists 'store-level visits' among UI outputs (S19). Places are snapped to POIs; no polygon count is documented (S28 has no polygon field). [347, 348, 28, 247, 119, 349] |
| Data access and integrations | live | C | Documented routes for results: the Attribution UI with daily updates and report sharing/exports (S19, S20); the Measurement API (MAPI) with a Data API ('Pull a custom dataset from any existing measurement report') and a Reports API, each with an API reference (S26). Exposure intake by pixel, SFTP or S3, the latter only when exposures match a MAID (S20, S23); automated OOH proof-of-play ingestion and a Spotify integration are claimed (S21). '550+ integrated partners' are media partners, not buyer access routes, and none is documented as a clean-room or MMP route in the pages opened (S21). [119, 246, 350, 236, 28] |
| Documented production volume | live | D | No US monthly panel figure with a definition and period is published for Attribution. Figures found, each on its own definition: 'aggregated coverage of 250 million U.S. devices' (Infillion release, 2026-09-18, S11); '9B+ monthly visits from 500M+ devices worldwide' (homepage, S12); '83 B+ Visits per year' (S21); 'measured U.S. audience of more than 100 million monthly devices' (as reported in May 2019, S16, pre-2024). Campaigns measured per year are not disclosed; Infillion says Foursquare 'powers location measurement for hundreds of brands' (S11). City Guide was retired in December 2024 (S18); no source says whether or how this changed the first-party panel. [52, 347, 28, 30, 64] |
| US availability and coverage | live | C | Generally available in the US: campaigns are configured nationally or by US Regions, States or DMAs, with census weighting to the US population (S24, S22, S23); transaction coverage is '450+ chains across the United States' (S23); the feasibility tool allows one country per campaign (S24). US production use is shown in vendor case material with unnamed US brands (S27) and in Infillion's statement that Foursquare 'serves all agency holding companies' (S11). [247, 120, 236, 351, 52] |
| Independent assurance | live | A | MRC record: 'Foursquare (formerly Placed) Visits data set' accredited for 'location information reported in Visits data set' (July-September 2020 communication, S01); accreditation 'previously suspended in October 2022' and revoked in December 2022 (S02). Foursquare's own 2020 release says Visits is the foundation of Attribution but that Attribution 'has not been audited or accredited by the MRC' (S05). On 2026-10-24 MRC's Digital accredited and under-review lists contain no Foursquare entry (S03); the Out-of-Home list page returned no readable list (S04). Whether the revocation of the Visits data set's accreditation counts as a withdrawal of 'the product's visit metric' for Attribution is a scoring question; the record itself is unambiguous that Attribution's lift metric was never accredited. [166, 167, 171, 173, 245] |
| Method disclosure | live | C | Of the seven elements: (1) sources named without proportions (S23); (2) visit definition described qualitatively (stop detection, snap-to-place) without thresholds (S23); (3) consent route described company-wide (S08) but no Attribution panel size; (4) projection documented (S22, S23); (5) no validation statistics; (6) control design documented as modelled or matched (S19, S22, S23); (7) daily update cadence documented (S19, S20), no uncertainty reporting. Roughly four to five elements are published in part; no data or code is available. [119, 246, 120, 236, 244] |
| Privacy and consent posture | live | C | Company-level consent route: opt-in precise and background location from Foursquare's apps; SDK/API partners 'may license data to us'; Attribution described as matching ad impressions to actions, including associating 'devices that we believe are related to each other' (S08, updated 2026-02-20). Sensitive places: NAI Precise Location Enhanced Standards signatory with a 13-category list (S08, S06 list dated March 2025), and the Privacy Center says Foursquare keeps 'an inventory of sensitive location categories', naming 'women’s reproductive health clinics' (S09); Washington and Nevada consumer-health-data notice (S10). California data-broker registry: Foursquare Labs registered, declaring it collects precise geolocation (S07). No Attribution-specific consent audit is documented. Regulatory record: no FTC or state attorney-general action against Foursquare was found in the sources screened; a putative class action (Proskin v. Foursquare Labs, D. Mass.) alleging that Movement SDK location data from third-party apps (Uber named) was sold without consent was filed 2023-07-24 and voluntarily dismissed without prejudice by the plaintiff on 2023-10-26, before any ruling; it falls inside the three-year window but is not an order, judgment or finding, and the allegations were never adjudicated (S53). [244, 243, 352, 250, 251, 353] |

Performance evidence: Linear-TV QSR case: 1.32% behavioral lift, 3.4M+ incremental store visits, 58.9M total attributed store visits; '80% of visits came from loyal, high-value consumers'. (vendor assertion, sponsor Foursquare) [351, 119]; Product-page cases (brands and media platforms not named here): 9.2% behavioral lift in visitation (retailer, social-video platform campaign), 2.5% more people to a QSR location and 7.8% behavioral lift for Foursquare audiences (QSR, social platform campaign), CVR 2.11% and CPSV US$0.68 (QSR agency case). (vendor assertion, sponsor Foursquare (with the media platforms involved)) [28]; 'A 23% lift in in-store visits among exposed users versus the control group' (national QSR, mobile in-app, $1 offer). (vendor assertion, sponsor Foursquare) [28]; '500%+ with FSQ Attribution' (homepage badge). (vendor assertion, sponsor Foursquare) [347]; Snap-to-Place 'performed over five times better at placing a device at the correct venue in dense cities than polygons'. (vendor assertion, sponsor Foursquare) [354]; Ground-truth validation: scaled transaction trends 'aligned with data in public financial statements'; visitation trends aligned with transaction trends. (vendor assertion, sponsor Foursquare) [236]; MRC accreditation (2020) of 'location information reported in Visits data set' after a CPA audit; revoked December 2022. (direct record, sponsor not stated in the sources opened (Foursquare was the audited party)) [166, 167, 245]; Check-ins cover about 10% of GPS-detected visits and about 75% of check-ins do not match a GPS visit (244 users, 2013). (synthesis, sponsor unknown (academic; no funding statement found)) [355]

### GroundTruth (active; ZeroToOne.AI announced on 25 March 2026 that it had acquired GroundTruth)

*xAd, Inc. (doing business as GroundTruth, per California's data-broker registry)*

Observed and Projected Visits carry no control and no lift claim; an MRC accreditation that ended with withdrawal; acquired by ZeroToOne.AI (announced 25 March 2026).

**SWOT (2026-10-24).** GroundTruth sells self-serve location media with built-in, exposed-only visit counts on a polygon place base, but its MRC accreditation ended by withdrawal and its projections and accuracy claims carry no published validation.

- **Strengths.** Satellite-traced three-layer polygons (in-store, on-lot, retail block), 4.8 million across the US and Canada, underpin targeting, visits and DOOH viewsheds; unmappable places fall back to 12-meter circles. (vendor documentation) [252, 253, 254, 255] Documents observed and projected visits as separate metrics, with the 14-day window, last-touch credit, 7-day verification and projection thresholds stated. (vendor documentation) [6, 18] Publishes three protected-location classes with category lists and states which targeting and measurement features each state rule switches off. (vendor documentation) [256] Self-serve Ads Manager reaches six channels, including DOOH via Place Exchange and direct mail, with a $0 minimum in its own comparison. (vendor claim) [122, 257]
- **Weaknesses.** MRC revoked accreditation of its Location, Place, and Visit data set after GroundTruth withdrew; no MRC listing at 24 October 2026. (record) [169, 170, 171, 177] Observed and Projected Visits are exposed-only with no control; a separate DOOH claim of incremental visits documents no control design. (vendor documentation) [6, 122, 257] Projected Visits name no weighting variables, report no uncertainty and show no validation against door counts or transactions. (vendor documentation) [18] Accuracy evidence is vendor-commissioned receipt-panel audits from 2017-2019 (10,000-panelist sample stated) and an undated 99% claim, with no method or report published. (vendor claim) [258, 259] No current US panel size; the only source proportion dates from 2018 and the bid-stream is an acknowledged source. (vendor documentation) [260, 261, 262]
- **Opportunities.** Re-entering MRC audit could restore the independent standing it held from July 2021 until its September 2023 hiatus. (inference) [168, 169, 170] ZeroToOne.AI's March 2026 acquisition adds predictive audiences; a published holdout test of them would turn bare accuracy claims into evidence. (inference) [47, 263, 264] DOOH attribution launched October 2025 through Place Exchange gives a measurement offer in a channel where buyers ask for outcomes. (inference) [257]
- **Threats.** Five state restrictions dated 1 October 2025 to 1 October 2026 already switch off visit tracking there; more states would widen gaps. (inference) [256] Registered as a California data broker collecting precise geolocation, and saying it 'may sell your sensitive personal data', it sits where state laws are tightening. (inference) [251, 265, 256] Selling media, defining the visit and billing per visit without independent audit invites buyer challenge to Cost Per Visit invoices. (inference) [258, 266, 170]

- **What is sold.** Location-targeted media across mobile, CTV and OTT, desktop, audio, DOOH and direct mail, sold self-serve through Ads Manager and through a managed service, with visit measurement attached: Observed Visits (exposed users later detected inside a Blueprint polygon within 14 days, last-touch) and Projected Visits (an impressed-device panel scaled to all impressed devices). Buying models are CPM and a managed Cost Per Visit model. Targeting layers include geotargets, real-time on-premise targeting, Neighborhoods, deterministic visit audiences, ZeroToOne.AI predictive audiences and Datonics third-party segments (S19, S20, S27, S38, S42, S46).
- **Who buys it.** Brands, agencies and small and mid-sized businesses buying local and national media, self-serve or managed; the acquirer claims 'more than 2,000 enterprise customers globally' (S11, S42, S08). No client is named in this profile.
- **How it earns.** Media margin on CPM buys and pay-per-visit pricing under Cost Per Visit; a flat US$1.50 CPM upcharge for Datonics Premium Audiences; measurement is delivered with GroundTruth-served media rather than sold as a stand-alone fee in the sources opened (inference; S27, S38, S46). No revenue figure is published.
- **Supply and data dependencies.** Location observations: owned-and-operated WeatherBug app, direct app partnerships, SDK integrations and the bid-stream, with a 2018 statement that 90% of signals came from the first three (S25, S26); privacy-policy source categories add publishers and consumer data resellers (S43). Consent route: OS location permission in partner apps, with opt-out through OS settings and DAA AppChoices plus state-law rights to opt out of sale, limit sensitive-data processing and withdraw consent; no app-level consent audit described (S43). Precision class: device latitude and longitude matched to building, lot and block polygons (precise), with coarse ZIP, DMA and state targeting also sold (S21, S38). Bid-stream use: acknowledged as a source; centroids and carrier-IP locations filtered (S24, S26). Other dependencies: TapAd IP-to-device matching for CTV visits (2021, S31); Place Exchange for DOOH supply (S49); Datonics and ZeroToOne.AI for audience data (S40); POI business databases for sensitive-place classification (S34).
- **Differentiation.** Three-layer polygon place model traced from satellite imagery (in-store, on-lot, retail block) used for targeting, measurement and DOOH viewsheds, with a 12-meter radial fence for places that cannot be Blueprinted (S21, S28, S44). Pay-per-visit buying (Cost Per Visit), managed, priced on its own verified visits (S27, S46). Documents both an observed and a projected visit number, and labels them as such (S19, S20). Publishes product-level sensitive-location classes with category lists, and feature-by-feature withdrawals by state (S34). Self-serve platform with a $0 minimum in the vendor's comparison table and six channels including direct mail (S42).
- **Risks.** Assurance risk: MRC accreditation granted July 2021, hiatus September 2023, revoked after withdrawal (row 'February', communication 26 July 2024); vendor pages still describe the accreditation (S01, S02, S05, S08, S52). Regulatory exposure: five state restrictions take effect between 1 October 2025 and 1 October 2026 (vendor-stated dates) and already remove visit tracking there; more states would widen the gaps (S34). Inference: further state precise-geolocation laws are the main product risk. Regulatory record: no FTC or state order against GroundTruth or xAd found in the searches logged (2026-10-24); the company is registered as a California data broker collecting precise geolocation and says it 'may sell your sensitive personal data' (S09, S43). Ownership and strategy change: ZeroToOne.AI announced on 25 March 2026 that it had acquired GroundTruth, using debt financing; marketing now leads with AI prediction claims that carry no published method (S11, S38). Seller-as-measurer conflict: GroundTruth sells the media, defines the visit and bills Cost Per Visit on its own count, with no independent audit since the MRC withdrawal (S05, S27, S46).
- **Best fit.** Local and multi-location advertisers who want self-serve geotargeted media with a built-in visit count and accept exposed-only, vendor-defined visits (S19, S38, S42). Advertisers who want to pay per visit and accept that the seller counts the visits (S27, S46).
- **Poor fit.** Buyers who need causal incrementality or an independent, accredited visit metric (S05, S19). Cross-media measurement of media bought elsewhere; the metric covers GroundTruth-served ads only (S19). Campaigns focused on Maryland, Oregon, Virginia, New Jersey or Connecticut, where visit tracking and on-premise targeting are switched off (S34).
- **Limits.** Visit measurement is exposed-only: Observed Visits and Projected Visits count or project visits by exposed devices with no control group, so neither is lift (S19, S20). Projected Visits are model outputs scaled from an undisclosed panel with no weighting variables, uncertainty or validation (S20). No current MRC accreditation: the Location, Place, and Visit data set was revoked after GroundTruth withdrew (reported 26 July 2024); it is on no MRC list at the cutoff (S05, S06, S07). Measures only GroundTruth-served media; no export of visit data to MMPs, ad servers or clean rooms was found (S19, S30, S42). No numeric dwell, accuracy or de-duplication threshold and no detection error statistic is published (S19, S24, S48). Visit tracking is unavailable for locations in five states (MD, OR, VA, NJ, CT) on GroundTruth's own account, so national campaigns carry measurement gaps (S34). No current US panel size; the most recent panel-type figures date from 2016 and 2017 (S16, S18).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Sources are named in vendor help articles, with one stale proportion and no per-source precision class. A help article whose body was last edited 12 April 2018 (still published) says '90% of our location signals' come from 'owned & operated (WeatherBug), direct data partnerships with apps, and SDK app integrations' (S25). A second article (body edited 2 April 2024) says 'we collect location data from apps and through bid-stream' and that 'the majority of our data comes from pings from apps and not just ad requests' (S26). Filters named: centroid ad requests are removed and 'GroundTruth does not accept location information coming from Carrier IPs' (S24). Projected Visits is built on Observed Visits 'including first-party data from WeatherBug' (S20). The privacy policy (effective 2 April 2026) lists sources of geolocation information as consumers, customers, 'Ad inventory supply partners (Publishers)', service providers, 'Consumer data resellers' and other third parties, without proportions (S43). Vistar Media's partner page describes GroundTruth's audience data as first-party, third-party partnerships and 'bid stream data' (S50). The 2021 release refers to 'owned & operated properties, along with 3rd party mobile location data' (S08). No current proportion, no per-source accuracy class and no statement of which sources feed visit measurement versus targeting were found. [356, 18, 357, 260, 261, 265, 191] |
| Location fix and visit definition | live | C | Polygon place model and a visit window are documented; no numeric dwell, accuracy or de-duplication rule and no detection error statistic. Places are 'Blueprints', polygons 'created by tracing the store or place boundary based on satellite images' in three layers (in-store, on-lot, retail block) (S21); a visit is 'recorded when a device that was previously exposed to an ad is detected within a Blueprinted area' (S28). An Observed Visit is a user visiting the brand or 'drive-to' location 'within 14 days after being exposed to a GroundTruth ad', credited by last-touch, with a 7-day verification period before 'verified visits' appear (S19). Which Blueprint layer qualifies as a visit is not stated. The verification article lists the tests applied to each signal and visit (GPS-sourced, derived or stale, outlier, fraudulent device, inside a Blueprint, during open-store hours, 'typical employee behavior', drive-by speed) without thresholds (S24). A 2019 post names dwell-time, speed and horizontal-accuracy thresholds without values (S48). Precision class: device latitude and longitude inside a building or lot polygon, i.e. precise; a POI that 'cannot be accurately Blueprinted' is automatically assigned 'a standard 12-meter radial fence' for targeting and measurement (Glossary, updated 6 January 2026; S28). [6, 252, 357, 255, 259] |
| Ground-truth validation | live | D | Vendor-commissioned comparisons with a receipt panel, reported as percentages with sample size and reference stated but no method or error statistic. A 2019 post reports the 'third annual Accuracy Audit' by Numerator (formerly InfoScout), comparing GroundTruth's in-store foot traffic with Numerator receipt data 'for a sample of key retailers', finding '94.4%* visit accuracy' (S48); the footnote states the audit 'analyzed 10,000 matched panelists who shopped at a sample of key retailers, consistently submitted receipts' and used GroundTruth location services, but period, retailer list, method, error statistic and the treatment of unconfirmed visits ('couldn't be 100% verified') are not published and the report was not found. A help article edited March 2024 claims '99% accurate as verified by Numerator' with no year or method (S27; also S52). The 2017 rebrand release says 'GroundTruth commissioned InfoScout' to verify its data at select retailers and cites 'more than a 90% accuracy rate' (S16). What is measured is the share of GroundTruth-recorded visits a receipt panel confirms (a precision check on observed visits); none of these compares visit counts or Projected Visits with door counts or transactions, and none reports recall. MRC's 2021-2024 accreditation (see assurance) is the only independent record touching visit determination. Data vintage: 2017-2019 audits predate the 2020-2021 permission changes; the 99% figure is undated. [56, 258, 259, 358] |
| Causal design | live | C | Default design is exposed-only, with no control. Observed Visits counts exposed users' later visits with last-touch credit (S19); Visitation Rate is Visits / Impressions and Store Visitation Rate Visits / Reach (S19, S28); Projected Visits scales exposed-device visitation (S20). No minimum sample, significance rule or uncertainty for lift is documented; GroundTruth says it has no 'standard match rate for visitation' (S32). 'Incremental' is the vendor's term, not lift: the '89% Incremental Visits' card on the Ads Manager page (S42) is defined in its case study as visits from newly acquired customers, and 'more than half of those visits being incremental' for a national retailer in the DOOH release states no comparator; that release also offers 'third-party lift studies' without a design (S49). The only comparative metric documented, Market Share, compares the brand's share of all visits in the targeted geography 90 days before the campaign with campaign start to 30 days after, shown 'only' as a directional trend with the caveat 'Other factors may contribute to changes in market share' (help article updated 15 November 2025); it is not a control. No GroundTruth document opened describes a holdout, ghost-ad or matched control. [6, 18, 255, 359, 122, 257] |
| Coverage | live | C | 4.8 million polygons across the US and Canada, US count not separated. 'GroundTruth currently has 4.8M Blueprinted locations across the United States and Canada' (S22, body edited 2023-09-18); 'a database of 4.8M Blueprints' adding 'thousands of new Blueprints each month' (S21, 2026-02-25). Other vendor figures conflict: 'over 5 million Blueprints' (S45), '5 million Blueprints globally' (S27), '6 million Blueprints globally' (S52, 2023). Brands are 'pre-blueprinted'; custom places are Blueprinted on request, and users cannot draw their own (S21, S29). Places in protected categories are excluded from Blueprinting or visit measurement (S34). Places that cannot be Blueprinted are measured with a standard 12-meter radial fence instead, so the measured place base exceeds the polygon base by an unstated number (S28). Under the anchors the documented base is at least 1,000,000 places with polygons if the US share is a majority, but the US share is not stated. [252, 253, 258, 360, 256, 361, 358] |
| Data access and integrations | live | C | Visit metrics are delivered inside GroundTruth's own Ads Manager for GroundTruth-served media, with reporting export routes: an 'external reporting API' with documentation at reporting.groundtruth.com/api (S37; only the page title readable, S51); 'Reporting Integrations' to TapClicks, NinjaCat and Agency Analytics (S42); PDF and tabular export (S30). No MMP, ad-server, clean-room or log-level visit export route was found. Visits are defined on exposure to 'a GroundTruth ad' (S19), so the product does not measure other sellers' media in the sources opened; DOOH exposure is measured on DOOH bought through GroundTruth (S44, S49). [6, 362, 363, 122, 254, 257, 364] |
| Documented production volume | live | D | No current US panel figure. Dated vendor figures, each with its own definition: a 'mobile panel - users who have opted-in to share their location data many times a day' expected to reach 50 million after the WeatherBug deal (November 2016, geography not stated; S18, S17); WeatherBug 'over 20 million unique users per month' (2016, S17); '95 million active monthly users and 17 million active daily users' (June 2017, geography not attached; S16), in the same release as claims that GroundTruth 'sees two out of three people who own a smartphone in the U.S.' and 'over one billion physical visits to places globally each month' (S16); 'over 30 Billion visits annually' (2021, S08; repeated in S27); 'more than 2,000 enterprise customers globally' (March 2026, S11). No count of campaigns measured per year was found. Every panel figure predates 2024 (capped under the rules) and none is a US monthly panel on the brief's definition. [356, 47, 56, 365, 262, 258] |
| US availability and coverage | live | C | Sold in the US with documented state-level gaps. Ads Manager targets 'USA' and 'Canada' (Quebec excluded following Quebec's privacy law, effective 22 September 2023, per an article updated 15 November 2025; GroundTruth's own exit date is not stated) (S33); Blueprints cover the US and Canada (S22); in-store visit attribution uses 'GroundTruth's verified blueprint database of commercial locations across the United States' (S29); vendor case studies report US campaigns (S47, S49). GroundTruth's own guidelines (updated 28 September 2026) list 'Visit Tracking' as unavailable for locations in Maryland (from 1 October 2025), Oregon (1 January 2026), Virginia (1 July 2026), New Jersey (8 July 2026) and Connecticut (1 October 2026), and say foot-traffic visits to Healthcare-Protected places in WA, NY, CT, NV and CA 'are not tracked or attributed' (S34). [253, 360, 366, 256, 367, 257] |
| Independent assurance | live | A | Accreditation granted, then withdrawn, documented by the accreditor. MRC granted accreditation 'to GroundTruth's Location, Place, and Visit data set' in July 2021 (S01); granted a 'Six-month accreditation hiatus period' at GroundTruth's request in September 2023, 'The service is not accredited during the hiatus period' (S02); and recorded 'Accreditation revoked ... after being in hiatus, due to withdrawal from the MRC accreditation process' in a row labelled 'February' in its April-June 2024 communication of 26 July 2024 (S05). The October-December 2023 and January-March 2024 communications carry no GroundTruth row (S03, S04), so the effective revocation month is unconfirmed. GroundTruth appears on neither MRC's accredited nor its under-review digital list at 24 October 2026 (S06, S07). The accreditation covered a data set, not a visit-lift metric. GroundTruth's 2021 release announcing it remains online (page modified 4 February 2026) with no hiatus or revocation notice (S08), and a comparison page last modified 9 November 2023, after the hiatus began, states in present tense that GroundTruth is 'Accredited by the Media Rating Council (MRC) for our Place, Location, and Visitation datasets' (S52). [168, 169, 368, 369, 170, 171, 177, 356, 358] |
| Method disclosure | live | C | Partial summary method. Of the seven elements: (1) sources named, one stale proportion (S25, S26); (2) polygon and window documented, no thresholds (S19, S21, S24); (3) no current panel size; consent route described as OS settings, DAA AppChoices and state-law opt-out, limit and consent-withdrawal rights (S43); (4) projection summarised without variables (S20); (5) validation stated as audit percentages with a 10,000-panelist sample and no method (S27, S48); (6) control design disclosed as none: exposed-only, last-touch, with a pre/post Market Share trend formula and timing published (S19); (7) daily panel refresh, 7-day verification lag and 21-day post-flight reporting stated, no uncertainty (S19, S20, S30). About four to five elements are published in summary form; no data, code or report is available. [6, 18, 252, 357, 260, 261, 258, 362, 265, 259] |
| Privacy and consent posture | live | C | Product-level sensitive-place rules and named state laws are documented by the vendor; consumer choices are opt-out (DAA AppChoices, OS settings, state-law opt-out and universal opt-out signals), with consent withdrawal referenced for sensitive data where state law requires consent, and no app-level consent audit described; registered as a California data broker; no regulator order found. Guidelines updated 28 September 2026 define three classes that govern targeting, audience building and measurement: Consumer-Protected (1-mile minimum; 'Individual device-level location data is not collected'), Healthcare-Protected (0.4-mile minimum for healthcare advertising in WA, NY, CT, NV and CA; visits 'not collected or attributed') and Prohibited (no location targeting; visits 'cannot be measured or attributed'), with category lists, plus 1-mile minimums and no visit tracking in MD, OR, VA, NJ and CT, a ban on GroundTruth pixels on sensitive health sites, and the rule that 'the most restrictive requirement always takes precedence' (S34). An older article allows medical locations to be marked 'semi-sensitive' on request so visits can be tracked (S35, body edited 2021). A 2025 article says physical therapy locations are in the Consumer-Protected category, so GroundTruth does not Blueprint, track or build audiences from them, while 'aggregate visit reporting is available in connection with drive to campaigns' (S36). The privacy policy states 'NOTICE: We may sell your sensitive personal data', names the DAA's AppChoices and OS location settings as the choices specific to precise location, lists state-law rights to opt out of sale, limit or opt out of sensitive-data processing and withdraw consent previously given, honours universal opt-out mechanisms, and reports 29,209 opt-out-of-sale requests 'in the previous calendar year', all complied with in whole or in part, the same figure California's registry gives under 2024 column headers (S43, S09). CalPrivacy's registry lists 'xAd, Inc.' doing business as 'GroundTruth', collecting 'precise geolocation' ('Yes') and not reproductive health care data, with no sale to the federal government, law enforcement, foreign actors or GenAI developers in the past year (registrant's answers, S09). This review found no FTC or state order against GroundTruth or xAd in the searches logged. No independent audit of the suppression rules was found; vendor statements are self-attested. [251, 256, 370, 371, 265] |

Performance evidence: '94.4%* visit accuracy' in the third annual Numerator accuracy audit (2019), up '4.4%' since 2017 (commissioned measurement, sponsor GroundTruth ('we annually appoint an independent auditor'); the post also says 'Commissioned by Numerator') [259]; '99% accurate as verified by Numerator' (vendor assertion, sponsor GroundTruth) [258, 358]; 'more than a 90% accuracy rate' (2017 rebrand release; an audit GroundTruth says it commissioned from InfoScout) (commissioned measurement, sponsor GroundTruth ('GroundTruth commissioned InfoScout')) [56]; Radial geofences measure visits 'with up to 75% inaccuracy'; Blueprints measure 'with complete accuracy' (vendor assertion, sponsor GroundTruth) [252, 253, 254]; Predictive Audiences 'over 90% predictive accuracy' / '90% accuracy' (vendor assertion, sponsor GroundTruth / ZeroToOne.AI) [263, 122]; ZeroToOne deployments across 'a subset of GroundTruth's customers': '70 percent reduction in media costs', '45 percent increase in ROI', 'over 85% prediction accuracy' (vendor assertion, sponsor ZeroToOne.AI (acquirer)) [47, 49, 372]; '89% Incremental Visits' for a specialty retailer using mobile and audio (vendor assertion, sponsor GroundTruth) [122]; National retailer on Place Exchange DOOH: 'over 42,000 in-store visits from exposed consumers', 'more than half' incremental (vendor assertion, sponsor GroundTruth and Place Exchange) [257]; Snack brand omnichannel (DOOH plus mobile) campaign: '221,000 physical, in-store visits' to a national retailer in one month at a '3.44% visitation rate' (vendor assertion, sponsor GroundTruth) [257]; Auto dealers' association Cost Per Visit test: 'over 1,200 visits' in a month, paid 'only for verified visits' (vendor assertion, sponsor GroundTruth) [367]; Tourism campaign: 'Over 7,500 visits' from on-premise mobile ads at a sports event and CTV retargeting (vendor assertion, sponsor GroundTruth) [254]; MRC accreditation of the Location, Place, and Visit data set (July 2021) (direct record, sponsor GroundTruth (audit client); MRC (accreditor)) [168, 169, 170, 356]

### InMarket (active, independent)

*InMarket Media, LLC (Delaware limited liability company; DBA InMarket)*

The only one of five FTC-ordered companies whose order bans selling precise location outright; synthetic-control lift.

**SWOT (2026-10-24).** InMarket sells visit and sales lift built on its own apps and partner SDKs, and location audiences, under a 2024 FTC order that bans selling precise location; every lift design it documents is matched or synthetic, with no accreditation or published validation.

- **Strengths.** Own reward and list apps, a partner SDK network and unnamed third-party sources supply location; InMarket holds item-level receipts and claims closed-loop visit and sales attribution. (vendor documentation) [134, 267, 268] An undated LCI glossary listed numeric minimums, a standard error and a confidence rule; the console now requires sign-in, so these are unverifiable publicly. (vendor documentation) [17] Measurement reaches buyers through platform partners: TikTok, Reddit, Yahoo DSP (daily observed visits) and Basis, announced between October 2024 and May 2026, plus Vistar Media (undated partner page). (vendor claim) [11, 269, 23, 270, 271] One vendor for visit and sales lift: Retail Sales Lift uses a partner's card transactions from a subset of the same exposed and control users; CPG lift, receipts. (vendor documentation) [11, 272]
- **Weaknesses.** Every documented lift control is matched or synthetic (1:1 matched devices; a synthetic control model); no randomised or holdout default was found, despite 'causal' marketing. (vendor documentation) [17, 121, 11, 273] The detailed method is an undated NinthDecimal-era console page; the dated 2026 description withholds the 'proprietary mechanics'. (vendor documentation) [17, 121] No MRC accreditation or MRC-listed review on 24 October 2026, and Comscore validation of visits is asserted without any published record. (record) [171, 177, 274, 40] Scale figures (about 200 million 'consumers', 110M monthly devices, 100M devices a year) use incompatible, undated definitions; no US monthly panel is disclosed. (vendor documentation) [275, 267, 274, 273] Visit geometry, dwell and accuracy rules are undocumented for both measurement and audiences. (vendor documentation) [17, 268, 276]
- **Opportunities.** Maestrø (beta, 18 June 2026) offers MCP access to outcomes for agent-driven buying; if it reaches production use it would be an early location-outcome signal for agents. (inference) [273] Publishing the current lift design, visit rules and a ground-truth comparison would separate InMarket from peers whose methods are equally thin. (inference) [17, 121] Circana's acquisition of NCSolutions, which InMarket dates to mid-2025, opens CPG sales-lift buyers to its receipt-based incrementality, per InMarket's March 2026 pitch. (vendor claim) [277, 272]
- **Threats.** FTC order C-4803 (29 April 2024; at least 20 years) bars selling precise location and targeting its identified sensitive locations; alleged breaches could reach federal court. (record) [130, 132] Its June 2026 policy reports state-law 'sale or share' of precise geolocation while denying plain-sense sale; regulators applying state sale bans could read this against it. (inference) [134, 130] InMarket gives a Texas address in California's data-broker registry, yet no Texas registry entry was found on 24 October 2026; the legal requirement is unassessed. (record) [251, 278] Selling media and guaranteed iROAS while measuring the outcome invites buyer challenge to its claim of 'independent measurement free from media buying incentives'. (inference) [279, 273, 276]

- **What is sold.** A real-time marketing and measurement platform with four lines: Audiences (InMarket Audiences / Audience Suite: 10,000+ syndicated location, intent and commerce segments, custom audiences, PMPs, GeoType personas, RMN and Lift Optimized Audiences); Activation (managed omnichannel media and the GeoLink self-serve platform with geo-audiences, geofencing and retargeting); Measurement (Campaign Incrementality for incremental visit and sales lift, still sold as LCI on other current pages; no opened source says Campaign Incrementality is LCI renamed, so the equivalence is this profile’s inference; CPG Sales Lift incrementality on receipt data; Visit Stream daily observed-visit feeds; Unified Measurement, an MMM and scenario-planning layer calibrated on LCI results); and InSights. Maestrø (beta, June 2026) packages outcomes for agentic optimisation, including a guaranteed-iROAS managed offer (S08, S09, S10, S19, S25, S26).
- **Who buys it.** Brands and agencies buying US retail, QSR, dining, auto and CPG media (the vendor’s own case studies), plus media platforms that offer InMarket measurement to their advertisers (TikTok, Reddit, Vistar Media, Yahoo DSP, Basis) (S10, S21-S25). No client is named in this profile.
- **How it earns.** Not disclosed in filings. The sources show media sales (managed activation and GeoLink self-serve), audience segments activated through DSPs and PMPs, measurement studies and dashboards, and a guaranteed-iROAS managed offer (S08, S19, S26). Pre-order, the FTC recorded revenue each time an advertiser used InMarket audiences on real-time bidding platforms and revenue shares paid to SDK app developers (S02 paragraphs 4, 11). No price or fee level was found.
- **Supply and data dependencies.** Location observations: SDK and first-party apps (CheckPoints, ListEase, Key Ring, Subscription Stopper, Out of Milk) and a partner SDK app network, plus unnamed ‘trusted & certified third-party sources’ (S05, S07); proportions not disclosed. Consent route: OS location permission in InMarket and partner apps, with reward incentives in CheckPoints (S05); the FTC alleged inadequate consent before the order and now requires Affirmative Express Consent records for own apps and an SDK Supplier Assessment Program for partner apps (S01, S02). Precision class: precise (SDK latitude/longitude, S02); targeting offered ‘down to the block level’ (S19). Bidstream: GeoLink scores the location in ad requests before bidding (S19); no bidstream use is documented for measurement. Other data: item-level receipts and card transactions (‘$197 billion+’; Retail Sales Lift via an unnamed sales data partner, S08, S21), surveys and shopping-list intent (S07). Distribution depends on platform partners (S21-S25). Historic dependencies: Thinknear (2019) and NinthDecimal (2020) assets (S27, S29).
- **Differentiation.** Owned-and-operated consumer apps plus a partner SDK network, alongside unnamed third-party sources, give InMarket permissioned location supply that most bidstream-reliant targeting vendors lack (S05, S07). Visits and item-level receipts in one platform: Retail Sales Lift uses card transactions, through a sales data partner, from a subset of the same exposed and control users, and CPG incrementality uses $197B+ of item-level receipt data (S15, S21). An undated LCI glossary read once on 24 October 2026 lists numeric minimum samples and a confidence rule (10 MM / 25 MM impressions; ‘>95%’), but the console now redirects to a sign-in page, so those thresholds are not verifiable from a public route (S11; LOC-C-F24). Broad platform distribution of its measurement (TikTok, Reddit, Vistar, Yahoo DSP, Basis) (S21-S25). The only FTC location-order company whose order bans selling or licensing precise location outright, which shapes its offer toward in-house activation and measurement rather than data licensing (S01).
- **Risks.** Regulatory: FTC Decision and Order C-4803 (complaint and proposed order 18 January 2024; issued 29 April 2024; finalisation announced 1 May 2024; term of 20 years from issuance, extended if a federal-court complaint alleging a violation is filed) bans sale or licence of precise Location Data and targeting on Sensitive Location Data tied to the sensitive locations its own programme identifies, and requires consent records, supplier assessments and deletion; an alleged breach could be pursued in federal court (S01, S03, S04). Disclosure tension: the June 2026 privacy policy reports state-law ‘sale or share’ of precise geolocation to Third Party Partners while denying plain-sense sale or licence; a regulator could read targeted-advertising sharing differently (S05, S01). State law: several states now ban the sale of precise geolocation (which states and from when is documented in workstream H, not re-read here); InMarket discloses state-law ‘sale or share’ of precise geolocation and no InMarket page opened says how it handles those states (S05; inference). Registration: InMarket gives a Texas address in California’s registry but no entry was found in Texas’s registry on 24 October 2026; the legal requirement was not assessed (S30, S31). Method currency: the public method is undated and NinthDecimal-era; buyers cannot tell which control design produced a given lift number (S11, S14). Arbiter conflict: InMarket sells media and guaranteed iROAS (Maestrø Outcomes, GeoLink) while measuring the outcome, despite marketing ‘Independent measurement free from media buying incentives’ (S16, S26). Supply: dependence on its own reward and list apps and on partner SDK apps whose consent the order now requires it to verify annually; unverifiable data must not be used (S01 Provision VI).
- **Best fit.** US brands in retail, QSR, dining and CPG that want visit and sales outcomes from one vendor, read as modelled incremental estimates, and that buy media on platforms where InMarket measurement is already integrated (S10, S15, S21-S25). In-flight optimisation on daily observed-visit signals where directional reads are acceptable (Visit Stream, Yahoo DSP) (S10, S24).
- **Poor fit.** Buyers who need a randomised or holdout lift design by default, an MRC-accredited metric, or a published validation of visit counts (S11, S14, S32, S33). Buyers who want to license raw precise location data: the order bars InMarket from selling or licensing it (S01). Buyers needing audited segment accuracy at the cutoff: the only audit found is a commissioned 2020 one (S34, S35).
- **Limits.** The visit-lift control is matched or synthetic by every documented description; no randomised or holdout default was found, so InMarket’s ‘causal’ and ‘incremental’ lift is a modelled comparison under this paper’s rules (LOC-INMARKET-S11, S14, S21). The detailed method (control matching, minimums, uncertainty, projection) sits on an undated, NinthDecimal-era console page, while the dated 2026 description is a one-paragraph synthetic-control summary (S11, S14). No published ground-truth validation of visit counts; Comscore validation is asserted without a document (S07, S12, S27). No MRC accreditation or MRC-listed review for any visit or lift metric (S32, S33). No dated US monthly panel; scale figures use incompatible units (S02, S07, S12, S23, S26). Visit geometry, dwell and accuracy rules are undocumented for both rows (S11, S08, S19). Audience activation routes are mostly unnamed (‘any DSP’); Basis is the one named external DSP (S08, S25).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Current vendor documentation names three source types without proportions: ‘InMarket owned and operated first-party app’, a ‘Comscore-verified partner SDK app network’ and unnamed ‘Trusted & certified third-party sources’ (S07); precise coordinates come ‘via mobile applications operated by InMarket and Third Party Partners’ (S05); the undated NinthDecimal sell sheet says ‘100% SDK Data’ (S12) and Vistar’s page ‘SDK-based location data’ (S23). The FTC complaint (an official record of allegations, January 2024) describes SDK collection of precise latitude/longitude with timestamp and device ID in own apps and 300+ third-party apps, plus consumer information purchased from other sources (S02). Precision class: precise (SDK GPS-class fixes) per S02; no per-source precision for the unnamed third-party sources. No bidstream use is documented for measurement; GeoLink’s ad-request scoring (S19) concerns activation. The FTC alleged that consent was not properly obtained (F07, S02 paragraph 12), which bears on the 0 anchor’s ‘consent claims’ for the pre-order period. [267, 134, 274, 23, 275, 276] |
| Location fix and visit definition | live | C | Documented: measured places are ‘Client-specified locations’; control devices are drawn from visits to the same locations with employees removed; the lift window is two weeks after versus two weeks before exposure (S11, undated). Not documented for the product: polygon or radius geometry, dwell rule, horizontal accuracy threshold or de-duplication. The FTC complaint records that InMarket processes location data to determine how long a device stays at a location (dwell exists; no rule stated) (S02 paragraph 7). A 2020 Location Sciences audit of InMarket audiences (not this row) counted a visit only inside store boundaries (S35). Precision class sold: precise (S02, S07 ‘Accurate and precise location data’). [17, 275, 373, 267] |
| Ground-truth validation | live | D | Accuracy claims only: ‘Comscore-Verified Visit Methodology’ (S12, undated), ‘LCI Attribution Comscore validated for visit metrics accuracy and aptness’ (S27, 2020), ‘Data verified using multiple truth sets’ (S07), and a customer’s own unpublished analysis quoted on the TEI page (S18). No comparison with door counts, transactions or official visitor statistics, and no error statistic, was found in the sources screened by 2026-10-24 (F09). [274, 40, 267, 374] |
| Causal design | live | C | Default design: matched_or_synthetic. Current (2026-07-29): ‘a synthetic control that estimates what visits would have looked like without the campaign’, described ‘without getting into proprietary mechanics’ (S14). Older console (undated): 1:1 audience-matched control on eight traits from devices at the same locations, difference in two-week pre/post visit changes, standard error, ‘>95%+’ confidence rule, minimums of 10 MM / 25 MM impressions per campaign and per report dimension, waivers by feasibility assessment (S11). TikTok (2024): ‘similar control users’ (S21). No randomised, PSA or geo-experiment design is documented as default or on request in the sources opened. The vendor calls its approach causal (S14, S26); under this paper’s rules it is not. Which of the two documented designs is current at the cutoff is not settled (F02 vs F03). [121, 17, 11, 273] |
| Coverage | live | C | No US place base is documented: measured locations are ‘Client-specified’ per campaign (S11). The console’s general population refers to ‘110+ locations including residential, retail, government and commercial locations’ (S11), which reads as location types, not a place count. The CPG sales product covers ‘nearly all major U.S. retailers’ through receipts (S15), a sales, not visit, coverage. Coding: bespoke or campaign-specific places only. [17, 272] |
| Data access and integrations | live (Maestrø beta) | C | Named routes: TikTok (partner page, 2024-10-14, S21); Reddit, all US advertisers (2024-11-12, S22); Vistar Media DOOH, Live in USA (S23); Yahoo DSP, daily observed-visit stream into In-Flight Outcomes, US and Canada (2026-04-23, S24; Campaign Incrementality itself ‘available outside of the Yahoo DSP’); Basis (2026-05-27, S25); Visit Stream daily feed ‘to your activation platform of choice’ (S10); console dashboard (S11); dashboard, PowerPoint and ‘all data exports’ for the CPG product (S15); Maestrø MCP access in beta (S26). The Maestrø release boilerplate also says InMarket is available on Snapchat and Pinterest, a vendor claim with no partner page opened (S26). No published API reference or log-level export documentation was found. Partner routes are announced by interested parties. [11, 269, 23, 270, 271, 375, 17, 272, 273] |
| Documented production volume | live | A | Documented production volume: no US monthly device panel with a dated definition. Figures: about 100 million unique devices a year sent InMarket location data, 2016 to January 2024 (FTC allegation; annual, geography not stated, S02); ‘200 million+ Location-based consumers’ (current, undated, S07, S10); ‘200M+ opted-in consumers’ (2026-06-18, S26); ‘nearly 200M+ permission-based users’ (S23); 110M monthly active unique devices (undated NinthDecimal sell sheet, S12); a ‘US Only’ NinthDecimal general population of 110M unique smartphone devices over 13 months of data (1B devices in total; undated console, S11); ‘over 100M+ consumers’ reached (2026, undefined, S16). Campaign counts: 126 TikTok campaigns measured (period not stated, S21); 109 campaigns analysed (2026, S16). None meets the scale definition as stated. [275, 267, 375, 273, 23, 274, 11, 279] |
| US availability and coverage | live | C | Generally available in the US: Vistar ‘Live in: USA’ (S23); Reddit advertisers ‘across the U.S.’ (S22); Yahoo DSP in the US and Canada (S24); US observations: the console defines the general population as ‘US Only’ (S11) and the CPG product as spanning major US retailers (S15). US production use: vendor case studies of US brands on S07, S09 and S10 (not named here). Both US observations and US production use are documented in vendor sources. [23, 269, 270, 17, 272, 267] |
| Independent assurance | live | A | No MRC accreditation or MRC-listed audit for any InMarket or NinthDecimal visit or lift metric on MRC’s Accredited Services: Digital or Services Under Review: Digital lists on 2026-10-24 (S32, S33). Vendor claims of Comscore verification/validation (S07, S12, S27) have no published record; the Forrester TEI (July 2021) is a commissioned economic study, not an assurance of the metric (S10, S18). No self-attestation of MRC location-guideline conformance was found. [171, 177, 267, 274, 40, 375, 374] |
| Method disclosure | live | C | Seven-element count; elements (2), (4) and (7) rest partly or wholly on the legacy LCI console glossary (S11), read once, undated and since behind a sign-in, so not reproducible from a public route at the cutoff: (1) sources named without proportions (S07, S05) - partial; (2) visit definition: window and employee removal only (S11) - partial; (3) panel size and consent route: undefined counts (S07, S26), consent via OS location permission and app incentives described in the privacy policy (S05) - partial; (4) projection formula (S11) - yes, thin; (5) validation statistics - none; (6) control design (S11, S14) - yes; (7) uncertainty (standard error, confidence rule, S11) and cadence (daily feed S10, sub-weekly in-flight S26) - yes. About four to five elements in summary form, much of it from the undated console; no data released. [267, 134, 17, 273, 121, 375] |
| Privacy and consent posture | live | A | Regulatory record inside the three-year window: FTC Decision and Order C-4803, issued 2024-04-29 (S01), finalised 2024-05-01 by a 3-0-2 vote (S03, S04), settling allegations that InMarket did not obtain informed consent in its own apps and did not verify consent in third-party SDK apps (S02 paragraph 12). The order bans sale or licence of precise Location Data for any valuable consideration, bars products that categorise or target consumers on Sensitive Location Data tied to the sensitive locations its own programme identifies (nine categories, with a carve-out for compliance use), excludes coarse data at a radius of at least 1,850 feet, and requires a Sensitive Location Data Program, an SDK Supplier Assessment Program, consent records and deletion (S01). Vendor statements since: no sale or licence of precise geolocation ‘in the plain language sense’ alongside a state-law sale/share disclosure (S05); sensitive-location suppression in Ethics Principles with one example (S06); NAI member (S05, S09); precise location retained 37 months (S05). Registered in California with ‘collects precise geolocation: Yes’ (S30); no entry found in the Texas registry (S31). [130, 132, 131, 275, 134, 376, 133, 251, 278] |

Performance evidence: 121 of 126 TikTok campaigns measured with LCI showed statistically significant lift in retail and dining; TikTok at 1.2x the LCI Incremental Lift benchmark. (vendor assertion, sponsor InMarket and TikTok Marketing Science (both interested)) [11]; National dining chain: 26% incremental visit lift versus a 23% benchmark, 285,000 incremental visits; casual dining: 10% incremental visit lift and 17% incremental sales lift, US$197K incremental sales. (vendor assertion, sponsor InMarket) [121]; Restaurant chain CTV campaign measured with Campaign Incrementality: 40% incremental visit lift and 48% ‘Incremental Sales Visit’. (vendor assertion, sponsor InMarket) [267]; Home-page outcome figures: 52.6% and 31.5% incremental sales lift, 2.5% transaction size lift, 4.6M incremental visits, +$11.23 iROAS. (vendor assertion, sponsor InMarket) [133]; 11x improved media effectiveness (InMarket and CMO Council, June 2023); 25% more repeat store visits and $40 return on investment spend (Forrester TEI of LCI, July 2021). (commissioned measurement, sponsor InMarket (co-author of the CMO Council study; TEI studies are vendor-commissioned)) [375, 374, 269]; 109 campaigns analysed: those in the bottom quintile not yet optimising wasted more than 60% of media spend on average. (vendor assertion, sponsor InMarket) [279]; AI InSights recommendations ‘could help increase overall performance’ by up to +28%. (vendor assertion, sponsor InMarket) [377]; LCI visit methodology ‘Comscore-Verified’ / ‘Comscore validated for visit metrics accuracy and aptness’. (vendor assertion, sponsor InMarket (Comscore’s engagement terms not stated)) [274, 40]; Several InMarket audiences rated ‘Optimal’ by Location Sciences audits (H1 2020). (commissioned measurement, sponsor InMarket (first customer of the certifier’s Verify Audience platform; the announcement’s ‘paid for tests’ were across multiple agencies and suppliers, without saying who paid for InMarket’s audits)) [378, 373]; GeoLink: 4.6x higher CTR and 38% cost savings against an industry-leading DSP in four campaigns; Audiences outperform Wordstream benchmarks by 3.7x. (vendor assertion, sponsor InMarket) [276, 268]

### Placer.ai (active, independent)

*Placer Labs Inc. (Delaware corporation, per Form D and EDGAR)*

The most-quoted foot-traffic dashboard: its population visit estimates are projections from a panel it also shows raw, with one third-party check found, funding not stated.

**SWOT (2026-10-24).** Placer sells the most widely used US foot-traffic projections through broad documented delivery routes, but publishes no error statistic, holds no data-quality assurance, and has one third-party check on record, with its funding not stated.

- **Strengths.** Delivery is broadly documented: a published API reference, CSV/Parquet feeds, Snowflake private sharing, Bloomberg Terminal distribution and an enterprise MCP server documented by October 2026. (vendor documentation) [280, 281, 282, 224] The API and dashboard show raw panel visits beside projected visits, so buyers can see how thin the observed base is for a place. (vendor documentation) [3, 283] Visit rules are documented: geofenced POIs, one visit per person per POI per day in the visits metric, dwell from one minute, per-category default duration segments. (vendor documentation) [284, 285, 283] Sensitive-place suppression is documented per product and tied to named laws, Washington's My Health My Data Act and Nevada SB370. (vendor documentation) [286, 287]
- **Weaknesses.** Its population visit estimates are projections from raw panel visits it also shows; no weighting targets, per-estimate uncertainty or projection validation is published. (vendor documentation) [3, 285, 288] The only third-party check (funding not stated), at one Minnesota site in 2021, put Placer 30.5% above partial counters; no significant difference. (record) [101] Panel size appears as 20M+, 25M+ at about 9%, about 30M at about 8%, and about 7% of the population, mostly without period or definition. (vendor claim) [3, 289, 290, 291] No MRC listing and no independent data-quality audit; its certifications cover information security, and the CSA STAR entry is a self-assessment. (record) [171, 177, 292]
- **Opportunities.** An MCP server documented by October 2026 puts Placer analytics inside AI assistants, though nothing yet connects it to agent-driven media buying. (inference) [224] Publishing raw-to-projected scale factors and validation statistics would turn "+92% accuracy" into evidence no rival screened here offers. (inference) [283, 293, 102] Advertising pages and Placer's commissioned 2026 radio-campaign measurement show demand for campaign readings; a controlled design would make them causal evidence. (inference) [294, 291]
- **Threats.** The 2022 Senate inquiry over Planned Parenthood data shows how sensitive-place exposure draws scrutiny; state health-data laws already withhold Washington and Nevada locations. (record) [295, 296, 286] An unconfirmed ransomware leak-site claim from 12 October 2026 could become a documented breach if confirmed. (inference) [297] De-biasing changes, usually at most yearly, can restate past figures (Data Version 2.0 restated all history to 2017), undercutting buyers who quote Placer series in valuations or reports. (inference) [102, 285]

- **What is sold.** Subscription access to projected foot-traffic analytics for US places: visits, visitors, dwell time, hourly and daily patterns, visitor origins and True Trade Areas, inferred demographics and psychographics (licensed overlays), visitor journeys and cross-shopping, chain and industry rankings, retail-sales estimates, vehicle traffic at "traffic pins", migration trends, chain visit forecasts and CMBS feeds; delivered through a dashboard, a REST API, bulk feeds (including Snowflake private sharing), embeddable widgets, Bloomberg Terminal distribution and, as an enterprise entitlement, an MCP server. A freemium tier shows limited data. Advertising is one solution area (planning and campaign "lift" reading), with no separately named measurement product, though Placer has been commissioned to measure a campaign's foot-traffic impact.
- **Who buys it.** Retailers and restaurant brands, commercial real-estate owners and brokers, investors and finance firms, consumer-goods companies, civic bodies (economic development, tourism, parks) and agencies, as listed by Placer; US public-sector use is documented on public records in Utah and California (LOC-PLACER-S15, S16, S23, S25). Paid customers are vetted for an incorporated brick-and-mortar interest; no paid access for law enforcement or advocacy groups, per Placer (S19).
- **How it earns.** Subscriptions only (no one-time reports), priced by quote; unlimited properties within the subscription; a Customer Success Manager included (LOC-PLACER-S21). One public proposal (Utah, Sept 2024) priced a three-year economic-development subscription at US$34,000 rising to US$37,485 a year (S23). Revenue is not in any filing; Placer announced US$100 million ARR on 20 February 2024 (S20, S26).
- **Supply and data dependencies.** Location observations: background location from partner apps that integrate the Placer SDK ("hundreds" of apps, "500" in one deck, from "dozens" of SDK partners), iOS 55% / Android 45%, plus auxiliary Bluetooth, Wi-Fi beacon, battery and network data; TechCrunch reports unnamed "other third-party sources" (LOC-PLACER-S01, S19, S23, S26). Consent route: partners represent compliance and obtain "affirmative opt-in" (2024 deck and 2026 city sheet only) or "opt-in or opt-out consent" as required by applicable law (Placer's help centre, Privacy FAQ and 2026 Trust Center); identifiers stripped on partners' premises before receipt (S04, S17, S25). Precision class: not stated by Placer; inputs are GPS/Wi-Fi-class fixes matched to building polygons (inference), outputs are aggregated with a 50-device floor and approximate dots (S17, S18). Bidstream use: none documented and none excluded in the pages opened. POIs from third-party providers, public data and manual review (S03); demographic overlays from third parties (S23, S25).
- **Differentiation.** Exposes raw panel visits beside projected visits in its API and dashboard, which few foot-traffic vendors document (S01, S08). Documents product-level sensitive-place suppression tied to named state laws (Washington MHMDA, Nevada SB370) (S05). Broad documented delivery: dashboard, published API reference, bulk feeds, Snowflake private sharing, Bloomberg Terminal distribution (per Placer) and an MCP server (S11, S12, S13, S14). Sells aggregated place analytics only and says it is "not a data broker"; it is absent from California's data-broker registry (S04, S34).
- **Risks.** Privacy history: a 2022 press finding and a 14-senator inquiry (17 May 2022) over Planned Parenthood location data, with Placer admitting 2020-2022 access and a weekly feed (S31, S32, S33). No regulator order or court finding found through 2026-10-24. Unconfirmed security claim: a ransomware leak-site listing first observed 12 October 2026; no company disclosure at 2026-10-24 (S41). State-law exposure: Placer already withholds health-care locations in Washington and Nevada (S05); further state rules on precise geolocation could narrow its place base (inference). Supply dependency: the panel rests on partner apps whose consent wording Placer describes inconsistently ("affirmative opt-in" in a deck and a city sheet vs "opt-in or opt-out" in its help centre, Privacy FAQ and Trust Center) (S04, S17, S19, S23, S25). Restatement risk for buyers who use Placer series in reporting or valuation: de-biasing changes, usually no more than once a year, can alter past figures, as Data Version 2.0 restated everything back to 2017 (S02, S07).
- **Best fit.** Relative comparisons of US places, chains and categories over time (rankings, year-on-year trends, trade areas) where projection bias is shared across the compared units (S01, S11). Site selection, leasing and civic planning that need visitor origins and cross-shopping at scale without device-level data (S16, S23).
- **Poor fit.** Uses that need an observed or audited count of visits at one place, such as billing, rent clauses or official visitor statistics (S24, S35). Causal campaign measurement: no named product, no documented control design (S16, S39). Very small, dense or multistory places and low-traffic rural sites, where data can be flagged as partial or withheld (S06, S18).
- **Limits.** Placer's population visit, visitor and retail-visit estimates are projections from a device panel; it also exposes raw, unextrapolated panel counts (panelVisits, panelVisitors) (LOC-PLACER-S01, S08, S17). Panel size is stated in several mostly undefined forms between 2022 and 2026 (20M+; 25M+ and 30M at ~9%; ~30M at ~8%; ~7% of the population) (S01, S23, S25, S31, S39). No published error statistic, held-out validation or per-estimate uncertainty; accuracy claims ("+92%", "92-96%") carry no method (S02, S15, S25). History is restated when the model changes (Data Version 2.0 restated all data to 2017), so archived Placer numbers may not match today's (S02, S07). The only third-party check found, with its funding not stated, is one Minnesota site in summer 2021, where Placer was 30.5% above a counter-based visitor profile covering only part of the site, a difference the author found not statistically significant (S24). Small, dense or multistory places can be flagged as partial, and places under 50 unique devices show no data (S06, S18). No MRC accreditation and no independent data-quality audit; certifications are for information security only (S17, S35, S36, S37).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Sources named, without proportions: background location from "hundreds of partner mobile applications" that integrate the Placer SDK (S01, S23), iOS 55% / Android 45% (S01, S23), auxiliary Bluetooth and Wi-Fi beacon, battery and network data (S19); TechCrunch (Aug 2024) adds unnamed "other third-party sources" (S26); demographic overlays licensed from third parties (S23, S25). No proportions by source, no precision class per source, and no statement on bidstream use either way in the pages opened. The 2024 deck counts both "500 Mobile Apps w/ SDK" and "Dozens of mobile app partners" (S23), different units (apps and partners), not a contradiction. No regulator or court record on Placer's sources was found. [3, 379, 289, 380] |
| Location fix and visit definition | live | C | Geofence (polygon) POIs across the US, up to 200 square miles, no lower size limit (S06); custom POIs by polygon or ring radius (S08); visit = cluster of signals matched to a POI using POI metadata and visit time and duration (S03), counted once per person per POI per day in the visits metric, though breakdown endpoints (by day, hour, dwell) can count a device more than once a day and traffic-pin impressions count every pass (S08); dwell: short visits from 1 minute since Data Version 2.0, 31 March 2024 (S07, S42), default segment per category of all visits or 10 minutes or longer (S07, S10); dense-area algorithm with tighter boundaries and flagged POIs (S03, S06); "proprietary venue detection logic" counts visits "both inside and outside the geofence" (S06). Not documented: horizontal accuracy threshold, minimum fixes, detection error statistic, precision class sold (outputs are aggregated; display floor of 50 devices, S18). [284, 287, 285, 283, 381, 382] |
| Ground-truth validation | live | B | Third-party, funding not stated (so not independent under the rules): one single-site grey-literature comparison (University of Minnesota Extension, summer 2021, published April 2024): Placer 13,233 visits vs 10,138 in a visitor profile built from counters on the cross-country trails only, while Placer's polygon covered the whole area (+30.5%; "slightly higher", no statistically significant difference per the author), weekend/weekday ratio 7x vs 2.1x (not like for like), demographics significantly different from an intercept survey (S24). Vendor: validation against MLB attendance, transactions, store revenues, conference attendees and vehicle counters named with no result, sample or statistic (S02); "+92% accuracy" (S15) and "92-96% accuracy", correlations "consistently exceeding 90%" relayed in a city fact sheet (S25), all without method; Data Version 2.1 is said to better align "with external ground truth sources", unquantified (S09). Interested party: chain-level directional agreement in 9 of 12 retailer-quarter comparisons vs SEC-reported figures (S40). Vintage: the third-party check's visit period predates Data Versions 2.0 and 2.1 (2024), and the study does not state which Placer data version it used. No replicated or out-of-sample independent validation found. [102, 293, 101, 290, 383] |
| Causal design | n/a | E | n/a: cohort C mask (causal design applies to cohort A only). Recorded for context: the advertising page offers to "measure lift or incrementally" with no control design (S16); a radio network commissioned Placer.ai itself to measure a February 2026 national campaign and reported "+34% lift" with no control, baseline or formula described (S39), and the home page features a case study, "Audacy Ad Campaign Boosts Restaurant Chain Visits By 20%" (S15). No separately named measurement product and no cohort A row exists; whether commissioned measurement warrants one is a register question. [294, 291] |
| Coverage | live | C | "Millions of POIs across the U.S." with polygons from third-party providers, public data and manual review, thousands added weekly (S03, S06); all 50 states, territories excluded (S01); "Every county in the U.S." (S23). No dated numeric count of US places with polygons found in the pages opened. [3, 284, 287, 289] |
| Data access and integrations | live | C | Documented live routes: dashboard; REST API with published reference (PAPI v1.2.1; weekly quotas) (S11, S08, S22); bulk feeds in CSV or Parquet, daily/weekly/monthly (S12); Snowflake private sharing into the buyer's account (S13, updated 2026-01-05); remote MCP server as an enterprise entitlement, listed in Claude's connector directory and the ChatGPT store (S14, page last updated 2026-10-22; launch date not established); distribution inside the Bloomberg Terminal's {ALTD} alternative-data platform since September 2023, with Data Entitlements from 17 February 2026 (per Placer's release); embeddable widgets and a Data Marketplace of partner datasets inside the platform (S15). No public Snowflake, AWS Data Exchange or Databricks marketplace listing found in the searches logged. [283, 280, 281, 282, 224, 293, 384] |
| Documented production volume | live | D | Documented production volume, US: "tens of millions" of devices, about 7% of the US population (S01, undated); "30M Active Devices" and "25M+ devices representing ~9%" (S23, Sept 2024, no period); "over 20 million monthly active users" (S25, a city fact sheet relaying Placer, Feb 2026: the only monthly definition found); "roughly 30 million mobile devices" (S39, commissioned, Sept 2026, representing about 8% of the U.S. population); "over 20 million active devices" in May 2022 (S31). All are vendor statements or relays; definitions of "active" are not given; not comparable as a series. [3, 289, 290, 385, 291] |
| US availability and coverage | live | C | Sold in the US, built on a US-only panel (50 states, territories excluded) and US POIs (S01, S06); US production use documented on public records: a September 2024 proposal on a Utah public-meeting record (S23) and a California city's February 2026 system fact sheet (S25). International breadth not documented and not scored. [3, 287, 289, 290] |
| Independent assurance | live | A | No MRC accreditation or audit in progress (not on MRC accredited or under-review digital or OOH lists, 2026-10-24; S35, S36). No independent audit or certification of visit data quality, methodology or accuracy found. Security only: ISO/IEC 27001:2022 and SOC 2 Type II as stated by Placer (S17), CSA STAR Level 1 self-assessment listed 2026-07-21 (S37); these earn nothing here and are recorded under privacy. [386, 171, 177, 292, 387] |
| Method disclosure | live | C | Seven elements: (1) sources: named without proportions (S01, S19); (2) visit definition: polygon, dwell and de-duplication documented, accuracy filter not (S03, S07, S08); (3) panel size and consent route: stated in inconsistent, undefined units; consent wording varies (S01, S17, S23, S25); (4) projection method: summary only (S01, S09); (5) validation statistics: none published; (6) control design or accuracy method: none (validation references named only, S02); (7) update cadence and uncertainty: daily refresh (S16), model adjustments usually no more than once a year, with restatements (S02, S07), no per-estimate uncertainty. Roughly four of seven partly published; no data, code or replicable dataset for researchers found (free tier only, S21). [3, 102, 284, 285, 283, 288, 294, 386, 379, 388] |
| Privacy and consent posture | live | C | Consent route: partners' representations and an undescribed three-stage vetting; "affirmative opt-in" (only in the S23 2024 deck and the S25 2026 city sheet) vs "opt-in or opt-out consent" as required by applicable law (Placer's help centre S04, Privacy FAQ S19 and Trust Center S17); identifiers (MAIDs) stripped on partners' premises by a "differential privacy service" (S04); K-anonymity of 50 on displays (S17, S18); no data from known under-18s (S04). Sensitive places: suppression documented with named laws (Washington MHMDA, Nevada SB370) (S05, S06). Regulatory record in the window 2023-10-24 to 2026-10-24: no regulator order, court finding or documented breach found (S34 registry: not listed; searches logged); an unconfirmed ransomware leak-site claim of 12 Oct 2026 (S41) is not a documented breach. Before the window: 2022 press finding and senators' inquiry over Planned Parenthood location data, with Placer's admission of 2020-2022 access and a weekly feed (S31, S32, S33). Security certifications as stated by Placer (S17, S37). No independent privacy audit scoped to location data found. [389, 286, 386, 289, 290, 385, 295, 296, 251, 297] |

Performance evidence: Placer estimated 13,233 visits against 10,138 in a counter-based visitor profile covering only the cross-country trails (+30.5%; "slightly higher") at one Minnesota recreation area, summer 2021; no statistically significant difference per the author; weekend/weekday ratio 7x vs 2.1x; demographics significantly different from an intercept survey. (synthesis, sponsor University of Minnesota Extension Tourism Center (no funder stated)) [101]; "+92% accuracy" (home page); "92–96% accuracy" and ground-truth correlations "consistently exceeding 90%" (relayed in a city fact sheet). (vendor assertion, sponsor Placer Labs) [293, 290]; Visitation estimates are tested against customers' first-party data and public references (MLB attendance, transactions, store revenues, conference attendees, vehicle counters). (vendor assertion, sponsor Placer Labs) [102]; A US network radio campaign "generated a +34% lift in foot traffic" across 48 DMAs in February 2026, with 10-30% increases in medium and heavy markets and 0-10% in light ones. (commissioned measurement, sponsor The publishing US radio network, which commissioned Placer.ai itself to measure the campaign) [291]; A direct-mail campaign planned with Placer trade-area and competitor-overlap data reported a 7.1% response rate and a "$9 Return on ad spend". (vendor assertion, sponsor Placer Labs (case study on its advertising page; full study gated)) [294]; Placer visit trends agreed in direction with SEC-reported traffic or transaction figures in 9 of 12 chain-quarters (six chains, 2025-2026), median gap about 2.2 percentage points. (commissioned measurement, sponsor GrowthFactor (interested party: site-selection platform selling foot traffic as a data layer)) [383]

### Radar (active, independent)

*Radar Labs, Inc.*

Geofencing infrastructure for app developers that says it does not sell location data.

**SWOT (2026-10-24).** Radar sells well-documented geofence, place and trip detection to app owners about their own users and says it sells no data, but publishes no detection error rates, no defined volume and no independent assurance.

- **Strengths.** Documents detection at parameter level: geometry limits, accuracy-based confidence, buffers, a 140-second, 70-metre stop rule in its tracking presets, dwell thresholds and server-side event generation. (vendor documentation) [298, 299, 300] Eight documented SDK platforms plus a REST API, with Apache-2.0 client source and iOS, Android and React Native releases in October 2026. (vendor documentation) [301, 302, 303] Privacy notice and DPA commit Radar to use SDK data only to serve its customers and not to sell it, with consent and client-directed sharing exceptions. (vendor documentation) [304, 305] Short default retention for raw location updates (7 days) with per-project settings and immediate API deletion. (vendor documentation) [306]
- **Weaknesses.** No false-positive or false-negative rates for detection; published accuracy figures are generic ranges without sample, devices or period. (vendor documentation) [307, 308] Volume is stated only as undefined per-day and installed-device counts (1B+ API requests a day, 300M+ installed devices), with no monthly or US figure. (vendor claim) [309, 310] Hospitals and places of worship are selectable place categories; whether they are detected is the customer's choice, and the only further control is a customer-set per-state blocklist. (vendor documentation) [299, 311] GLI and BMM certifications are claimed without a public certificate, scope or version; SOC 2 report is available only on request. (vendor claim) [310, 306, 312]
- **Opportunities.** As US rules tighten on selling precise location, a documented no-sale processor model lets Radar sell detection rather than data to app owners. (inference) [304, 305, 313] Regulated gaming adds licensed demand: an active Kentucky licence from February 2025 and claimed geo-compliance work for four of seven top US sportsbooks. (inference) [314, 315] Publishing detection error rates under stated test conditions would let Radar lead a cohort that, in the sources screened, publishes none. (inference) [307]
- **Threats.** Engage competitor-visit audiences, and home and work inference surfaced as at-home and at-work states since July 2026, move Radar toward the sensitive uses state laws target. (inference) [316, 317, 318] US$14.5M of TriplePoint venture debt (plus a US$1.5M undrawn commitment) matures 1 September 2028; no priced equity round has been announced since February 2022. (record) [319, 320, 309] Contract drift: the 2023 Terms describe a free tier the pricing page drops, and pair a broad SDK-data use right with deference to a narrower privacy notice. (vendor documentation) [321, 304, 322] OS permission rules and Google Play's bar on geofencing foreground services limit background tracking; Radar's geofencing presets avoid foreground services, so trips and navigation feel it most. (vendor documentation) [300, 323]

- **What is sold.** Developer location infrastructure sold to app and web owners about their own users: SDKs and APIs for geofencing (circle, polygon and isochrone geofences with entry, exit and dwell events), place visit detection against Radar’s POI database for the chains or categories the customer enables (Enterprise plan), trip tracking with live ETAs and arrival detection, beacons, region (country and state) detection, fraud and spoofing checks, user insights (inferred home and work locations and, since July 2026, at-home, at-work, travelling and commuting states, enabled through Radar customer success), maps, geocoding, search and routing APIs, and the Reveal fraud-risk API. Since October 2025 these are packaged as three solutions: Protect (location verification and geo-compliance), Optimize (operations) and Engage (location-triggered messaging and first-party audiences). The scored row covers the geofencing, place detection and trip tracking SDK and APIs; Maps, Reveal and the solutions are recorded as context and not scored.
- **Who buys it.** Product, engineering, operations, fraud and marketing teams at consumer apps and enterprises (the vendor names retail, restaurants, gaming, financial services, travel and logistics as industries), and regulated online gaming operators for geo-compliance. Vendor pages quote named customers; this profile does not name them as clients.
- **How it earns.** Software subscriptions on annual commitments with usage billing: list prices start at US$0.04 per monthly tracked user for Protect and Optimize and US$0.02 for Engage, US$0.50 per 1,000 Core Maps calls and US$2.00 per 1,000 Premium Maps or Reveal calls, with negotiated volume discounts and no free tier (S04). Radar says it does not sell the data it collects (S29, S31, S32). No revenue figure is public.
- **Supply and data dependencies.** Location observations come from the customer’s own app or site through Radar’s SDKs under the OS location permission (precise or approximate), or from coordinates the customer posts to the track API. Signals fused on the device: GPS, Wi-Fi, IP, Bluetooth and cellular, as delivered by the OS. Consent route: the OS permission prompt run by the customer’s app; Radar acts as processor or service provider (DPA). Precision class: precise when precise permission is granted; requested iOS accuracy defaults to the hundred-metre class. Bidstream use: none documented. Third-party dependencies: POI data from unnamed open and commercial datasets; hosting on AWS and MongoDB (DPA sub-processors); OS location services (Apple Core Location, Android location APIs), whose permission rules bound what the SDK can collect.
- **Differentiation.** Detection rules documented at parameter level (geometry limits, confidence inputs, buffers, stop rule, dwell, de-duplication), more than any cohort B or C vendor profiled shows for its visit definition (LOC-RADAR-S06, S07, S13). Business model sells detection to the app owner about its own users and states it does not sell data, the opposite of the data-seller model behind the FTC location orders (LOC-RADAR-S29, S31, S32; inference). Server-side event generation lets geofence counts, polygons and isochrones exceed native OS geofence limits (LOC-RADAR-S06). Same SDK serves engagement, operations and regulated geo-compliance, with a regulator licence record in Kentucky (LOC-RADAR-S37, S46). Open-source client SDKs (Apache-2.0) with frequent releases and a machine-readable iOS privacy manifest (LOC-RADAR-S42, S43).
- **Risks.** Regulatory record: none found in the sources screened (California data broker registry and FTC case search, 2026-10-24; LOC-RADAR-S47, S48). Exposure grows as Engage builds audiences from location history, including competitor visits, and user insights surface inferred home and work as at-home and at-work states (since July 2026) (LOC-RADAR-S09, S38; inference). State precise-geolocation laws (for example Maryland’s and Oregon’s bans on sale) bind Radar’s customers; Radar’s no-sale commitment limits its own exposure, but sensitive-place detection is left to customer configuration (LOC-RADAR-S07, S29; inference). Contract-text drift: Terms last updated 24 May 2023 describe a free version the pricing page drops, and pair broad SDK-data use rights with deference to the narrower 2025 privacy notice (LOC-RADAR-S29, S30, S04). Platform dependency: Apple and Google permission and background-location rules cap what the SDK can observe; Radar’s Android docs already steer customers away from foreground services for geofencing under Google Play policy, and its geofencing presets never start one, so the limit bites hardest outside trips and navigation (LOC-RADAR-S13). Leverage: US$14.5M of venture debt to TriplePoint funds matures 1 September 2028, plus a US$1.5M unfunded TPVC commitment; no priced equity round has been announced since February 2022, though both funds acquired Radar preferred stock (cost about US$0.8M) on 30 September 2025 (LOC-RADAR-S02, S44, S45). AI-assistant access: the data MCP server sends coordinates and identifiers to third-party AI providers, whose retention Radar says it cannot control (LOC-RADAR-S27).
- **Best fit.** App owners who need geofence, place-visit or arrival events about their own consenting users, with configurable confidence, stop and dwell rules and server-side delivery to their stack (LOC-RADAR-S06, S07, S22). Regulated operators needing state-level location verification and spoofing checks, subject to their own review of the claimed GLI and BMM certifications (LOC-RADAR-S10, S39, S46).
- **Poor fit.** Buyers of third-party audiences, panels or foot-traffic datasets: Radar sells no observations of other companies’ users (LOC-RADAR-S29, S31). Uses needing a published detection error rate or an independent accuracy benchmark (LOC-RADAR-S33; F08). Programmatic geofenced ad buying: no DSP, SSP or bidstream route is documented (LOC-RADAR-S22; F20).
- **Limits.** No published false-positive or false-negative rate for geofence, place or trip-arrival detection, and no independent benchmark found (LOC-RADAR-S33, S03; F08). Accuracy figures are generic location-fix ranges and best-case marketing ("down to 5 meters"), while the SDK’s default requested iOS accuracy is the hundred-metre class (LOC-RADAR-S03, S13, S33). Volume figures carry no period, definition or geography (per-day API requests, installed devices, installed apps); reading them as run-rates and cumulative installs is this paper’s classification; no monthly device or request count (LOC-RADAR-S01, S02, S39, S40). The place database is "millions of places" from unnamed sources with no US count or polygon share (LOC-RADAR-S07). Sensitive place categories are detected if the customer enables them; no default suppression is documented beyond that choice and a per-state blocklist (LOC-RADAR-S07, S25). The confidence function behind event levels 1 to 3 is described qualitatively, not published (LOC-RADAR-S06, S07). Data-use commitments are vendor statements with a consent exception, client-directed sharing routes, a usage-data carve-out in the DPA and a broad Terms clause; no independent audit of them (LOC-RADAR-S29, S30, S31).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Sources are the customer’s own end users’ devices: Radar’s SDKs collect location under the OS location permission in the customer’s app or website, and customers may also post coordinates with an accuracy value to the track API (S23). Signal types named: "GPS, Wi-Fi, IP, and more" (S01); "sensor fusion across GPS, Wi-Fi, Bluetooth, and cellular" (S03); beacons for "micro-geofences, accurate down to a few meters" (S11); the web SDK uses the browser’s HTML5 geolocation API (S16); IP geocoding returns city, state and country (coarse, S16). Precision class: precise when the user grants precise permission (privacy notice: "Precise location (e.g., GPS-level) data", S29); Android approximate location is supported (S15); the SDK’s default requested iOS accuracy is kCLLocationAccuracyHundredMeters, "high" requests kCLLocationAccuracyBest (S13). Place data for visit detection comes from unnamed "open and commercial datasets" (S07). No proportions by signal, no per-source accuracy class and no bidstream or third-party location supply were documented in the sources opened (F20). [390, 308, 299, 391, 300, 323, 392, 302, 304] |
| Location fix and visit definition | live | C | Detection method, accuracy handling, dwell and processing location are documented; no detection error statistic. Geofence events are generated server-side from location updates (S06); the SDK uses OS region monitoring, visit monitoring and significant-change services as wake-up triggers and can sync nearby geofences to the device (S13); iOS client-side geofence notifications run on-device with no background location collection (S26). Geometry: circle radius 10-10,000 m, polygons of 4-2,000 coordinates, isochrones (S06). Accuracy filter: every event carries confidence 1-3 from device-reported accuracy and geofence geometry; optional buffer entries and buffer exits both enlarge the boundary by the reported accuracy, giving an earlier entry and a later exit (S06); fraud checks fail a location whose accuracy exceeds a threshold (S10). Stop rule: preset stopDuration 140 s and stopDistance 70 m (iOS Responsive and Continuous; all Android presets), configurable (S13); stop detection can require a stop before an entry (S06); places require a stopped or foreground update and a place matching the chains or categories the customer has enabled (S07). Dwell: user.dwelled_in_geofence after a per-geofence threshold in minutes (S06). De-duplication: "A device must exit a geofence before a subsequent entry into that same geofence" (S06). Place confidence adds place footprint, area density and popularity (S07). Trips: approaching (configurable ETA threshold) and arrival on destination-geofence entry (S08). Thresholds per confidence level and false-positive or false-negative rates are not published (F05, F08). [298, 299, 393, 394, 300, 395] |
| Ground-truth validation | live | D | Vendor accuracy figures with stated conditions but no method: a 2025 blog gives average observed accuracy by environment (5-10 m open outdoor, 10-30 m urban outdoor, 5-20 m indoor retail, 30-100 m rural, 10-50 m large venues) "based on data from thousands of app sessions and field tests", with no sample, devices, period or reference, and describes location-fix accuracy for geofencing generally, not Radar detection rates (S33); product pages claim "accuracy down to 5 meters" and "~5 meters" (S03, S38). Events can be accepted or rejected after check-ins to improve future confidence (S07), a feedback loop, not a published validation. GLI and BMM testing is claimed for geo-compliance (S39) without a public report. No independent detection-accuracy benchmark of Radar was found in the searches listed. [308, 299, 307, 317, 310] |
| Causal design | n/a | E | n/a: the cohort D mask excludes causal design (cohort A only).  |
| Coverage | live | C | Eight client SDK platforms each with a documentation page: iOS (iOS 15 and higher), Android, React Native, Web, Cordova, Capacitor, Flutter and MAUI (S12, S14-S21), plus a server-side REST API (S23). Recent releases: iOS 3.44.0 (2026-10-15), Android 4.0.0 (2026-10-16), React Native 4.37.0 (2026-10-22) (S43). The Flutter repository’s latest GitHub release is 3.12.4 (2025-03-17) while the docs pin ^3.12.4 (S18, S43); the Cordova plugin’s latest GitHub release is 3.9.1 (2024-04-08) and Capacitor’s is 4.1.1 (2026-10-07). Geo-compliance pages add desktop apps for Mac and Windows (S41). Distinct platform categories under the rule (iOS, Android, web, cross-platform frameworks, server-side): all five documented and live. [301, 396, 323, 392, 397, 398, 399, 400, 401, 302, 402, 303] |
| Data access and integrations | live | C | Routes documented with an API reference and per-integration pages: SDK callbacks and a REST API with publishable and secret keys (S23); webhooks and 16 named event integrations (analytics, CDP, messaging and marketing-automation platforms) (S22); geofence sync from HTTP or AWS S3 and three geofence integrations; one trip integration (S06, S22); Engage segment export to marketing platforms (S38); a manual per-user dashboard export (a user with their events and locations, from the user detail page, framed for data-access requests; not a bulk export) (S24); since October 2026, a pilot, read-only MCP server for querying a project’s location records from AI assistants, with paged queries and temporary JSON downloads (S27, S34). These are routes for the customer’s own data; no resale or marketplace route exists in the sources. [298, 403, 302, 306, 223, 404, 317] |
| Documented production volume | live | D | Documented production volume: no monthly device or monthly request figure with a stated period. Figures are attributes under the unit rules: "1B+ API requests per day" (a per-day figure; run-rate is this paper’s classification, not a vendor definition) and "300M+ installed devices" (read here as cumulative installs, likewise this paper’s classification of an undefined label) and "1,000+ installed apps" (S02); "1 billion API calls per day and installed on hundreds of millions of devices" (S01); "peaking at 20K+ per second" (S39); October 2023: "over 100 billion API calls per year from over 100 million devices" (annual total, S40). Radar defines monthly tracked users (S04, S24) but publishes no total. Geography is not stated for any figure. [390, 309, 322, 306, 310, 405] |
| US availability and coverage | live | C | Sold in the US: USD list prices (S04), US-hosted services ("hosted and operated in the United States", S29) and a New York headquarters (S29, S35). US production use: a Kentucky regulator record shows an active information-services-provider licence for Radar Labs, Inc. (issued 2025-02-25, expiring 2026-12-31, 2026 renewal paid), and the record links Radar Labs, as service provider, to a licensed Kentucky sportsbook operator, an independent record of US production use (S46); the vendor says it powers geo-compliance for "4 of the top 7 US sportsbooks" (S35) and claims gaming vendor licences in Colorado, Arizona and West Virginia from 2023 (S40). US observations: the place catalogue lists "thousands of U.S. and international chains" (S07) and US state detection and blocklists (S07, S10). The register’s "unconfirmed" US entry can be updated on this evidence. [322, 299, 394, 304, 315, 405, 314] |
| Independent assurance | live | D | No completed assessment of detection accuracy with a public scope and result was found. Claims: "GLI and BMM tested and certified" and mobile SDKs "audited and certified by Gaming Labs International (GLI)" (S39, S40); Mac and Windows apps "passed a comprehensive GLI audit" (S41), all vendor statements for geo-compliance with no assessor record, scope or version read. A gaming licence (Kentucky, S46) is a suitability approval, not a test of the metric. SOC 2 Type II (S24) is security and belongs to privacy under the rules. Trust portal unreadable (S51). [306, 310, 405, 402, 314, 312] |
| Method disclosure | live | C | Elements published (cohort D reading): (1) sources: named signal types, no proportions (S01, S03); (2) detection definition: detailed (S06, S07, S13); (3) how end-user permission is obtained: OS foreground and background permissions, prompt guidance and permission-state statistics documented (S14, S15, S24); (4) detection algorithm or configuration model: configuration parameters, server-side processing and confidence inputs documented, the confidence function itself is not (S06, S07, S13); (5) validation statistics: none for detection, generic accuracy ranges only (S33); (6) detection-accuracy method: not published; (7) update cadence and uncertainty: sync intervals and confidence levels documented (S06, S13), analytics refreshed nightly (S08, S24). Client SDK source code is public under Apache-2.0 (S43); server-side detection logic is not. Count read: four to five of seven elements, no data. [390, 308, 298, 299, 300, 396, 323, 306, 307, 303] |
| Privacy and consent posture | live | C | current_controls evidence: consent route documented for the product: collection only under the OS location permission, end-user opt-out by revoking it, and a statement that SDK data is used only to provide the services and not sold or made accessible to third parties without consent, while the notice lists three sharing routes (to the client, to the end user at the client’s request, and to integration partners on the client’s behalf) and says its use of SDK Information is governed by each client agreement (S29, S24, S32); DPA commits Radar, as CPRA service provider and VCDPA processor, not to "Sell or Share" customer personal information, except account and usage data it holds as an independent controller (S31); retention defaults (location updates 7 days, trips 90 days, users and events 1 year) and immediate deletion (S24); terms bar eligibility uses (S30); EU-U.S. DPF self-certification stated (S29). Sensitive-place handling: no default suppression found; sensitive POI categories (for example hospital, medical lab, nursing home, church, mosque) are detected if the customer enables those categories or chains, with a customer-configured, per-state category blocklist (which requires Regions) as the only further documented control (S07, S25; F19). Named legal frameworks: CCPA/CPRA, VCDPA, the Colorado, Connecticut and Utah privacy acts, GDPR and UK GDPR, Swiss nFADP and DPF (S29, S31); no state precise-geolocation statute (for example Maryland or Oregon) and no NAI standard is addressed in the sources opened. SOC 2 Type II attestation (security; S24). Tensions: Terms give Radar "the right to use and store the SDK Information to the maximum extent permitted by law" (S30), a right the Terms tie to Radar’s Privacy Notice; privacy notice lists mobile advertising IDs among SDK data while the iOS privacy manifest declares no tracking (S29, S42). Registry and regulator record: not on California’s data broker registry; no FTC case found (S47, S48; F15). Event flag: none found. Active legal obligations: none found. Risk cap: not triggered on the evidence screened. [299, 306, 311, 304, 321, 305, 313, 406, 163, 407] |

Performance evidence: "57% uplift in conversion rate from nearby notifications" and "4x uplift in open rates from location-based notifications" (vendor assertion, sponsor Radar (vendor)) [317]; Brands using nearby notifications "have seen conversion rate lifts of 25 to 57% compared to generic campaigns" (vendor assertion, sponsor Radar (vendor)) [308]; Users who engage with Store Mode convert at 2.5x the rate of those who do not (vendor assertion, sponsor Radar (vendor)) [308]; Average observed geofencing accuracy of 5-10 m (open outdoor) to 30-100 m (rural); "between 5 and 50 meters" in most consumer apps (vendor assertion, sponsor Radar (vendor)) [307]; "Accuracy down to 5 meters" from sensor fusion (vendor assertion, sponsor Radar (vendor)) [308, 300]; Background tracking with the efficient and responsive presets uses "only 1-2% battery per day" for most users (vendor assertion, sponsor Radar (vendor)) [306]; API latency p50 50 ms, p90 150 ms (vendor assertion, sponsor Radar (vendor)) [306]; Apps following best practices "can expect 70-80% of users to grant location permissions, with 40-50% of users granting background" permission on iOS (vendor assertion, sponsor Radar (vendor)) [306]; Mobile SDKs "audited and certified" by GLI; "GLI and BMM tested and certified" (vendor assertion, sponsor Radar (vendor); assessors GLI and BMM named, payment not stated) [310, 405, 402]

### Unacast (active)

*Gravy Analytics, Inc. (Delaware), doing business as Unacast; affiliates Unacast, Inc. (Delaware, 2016) and Unacast AS (Norway, org. no. 913820746); subsidiary Venntel, Inc.*

Owner of Gravy Analytics: an FTC order, a confirmed 2025 breach, and precise geolocation still sold.

**SWOT (2026-10-24).** Unacast sells precise device-level movement data and on-demand foot-traffic APIs; the dwell- and confidence-documented ML foot-traffic datasets it built moved with its Aggregated Location Insights business to Pine59 by management buyout announced 1 September 2026; Gravy (dba Unacast) operates under a 2025 FTC order and disclosed a January 2025 breach.

- **Strengths.** The ML foot-traffic datasets, now operated by Pine59, publish a numeric dwell rule (multiple pings within 50 metres over 60 seconds), visit-length buckets and per-estimate confidence scores. (vendor documentation) [106, 324, 325, 326] Device-level feed carries per-signal forensic flags on origin and accuracy, so buyers can filter signals; delivered by API or batch with up to three years of history. (vendor documentation) [327, 328, 329] State registrations disclose precise-geolocation collection, sharing or sale to federal and other state governments (California), and one breach with a short narrative (Texas). (record) [251, 278]
- **Weaknesses.** No independent US validation; accuracy evidence is an undocumented R-squared of 91.6% against one unnamed retailer and a 2021 self-comparison at five plants and two stadiums. (vendor claim) [106, 14] Volume is stated only in incompatible global units (1 billion monthly devices, 70 billion daily signals, 600 million MAUs); no US monthly device count is published. (vendor claim) [328, 330] Sources are named only as SDK aggregators, WiFi and beacons, without proportions; FTC's complaint alleged suppliers may obtain data from other suppliers, the mobile advertising marketplace or apps. (record) [137, 135]
- **Opportunities.** The order's consent-record and supplier-assessment duties could give buyers an auditable consent route, if Unacast publishes evidence of compliance. (inference) [135] With aggregated foot traffic moved to Pine59 on 1 September 2026, Unacast can concentrate on device-level data, identity linkage and APIs it still sells. (inference) [326, 331, 332]
- **Threats.** Order C-4810 runs 20 years from 13 January 2025, or longer after any federal-court violation complaint, and bans selling or using Sensitive Location Data, with exceptions. (record) [135] Unacast's Texas filing says an intruder copied data from Gravy's AWS buckets in January 2025 and briefly posted a small subset on a dark web forum. (record) [278, 333] Buyers face unclear provenance: documentation now redirects to Pine59 while Unacast pages still sell the same foot-traffic packages, so contract counterparties may be ambiguous. (inference) [325, 332, 326]

- **What is sold.** Device-level location data ("Global Location Data": timestamped coordinates with MAIDs, IP and forensic flags, by API or batch); foot-traffic outputs (on-demand Location Intelligence APIs for custom polygons: Area Visitors unique device counts, Observations, Trade Areas, Personas; machine-learning Foot Traffic Datasets and the Location Insights platform, which Pine59 now claims); consumer data enrichment (demographics, personas, MAIDs and hashed emails); Data Linkages (MAID to hashed email and to IP). Venntel sells location intelligence to government and risk buyers (context, not scored). Sources: S04, S05, S06, S08, S34.
- **Who buys it.** Described by Unacast as retail, real estate, supply chain, insurance, telco, consultancies, government, software and ad-tech buyers, and investors (S03, S06); its policy lists advertising, attribution, market research, fraud detection, law enforcement and national security among customer uses (S02). No customer is named here.
- **How it earns.** Data licensing scoped "to your exact specifications" and API access (S06); Datarade lists usage-based pricing from "$1 / API call" (S33). No revenue figure is disclosed.
- **Supply and data dependencies.** Location observations are licensed from app owners and aggregators "who primarily receive data from software development kits", plus WiFi routers and Bluetooth beacons; Unacast runs no SDK of its own (S02, S10). Supplier count stated as 15+ or 12+ (S04, S06). Consent route claimed as opt-in (S03, S05); the FTC complaint alleged that consent was not confirmed and that suppliers may draw on the mobile advertising marketplace (S25), and the order now requires consent records and a Supplier Assessment Program (S25). Precision class: precise latitude/longitude; the OS decides GPS, cell or WiFi (S02). Bidstream use: not documented by Unacast; alleged as a possible upstream source in the complaint (S25); reporting on the breach sample found much location inferred from IP (S29). The ML foot-traffic datasets use aggregated person counts from a first-party SDK partner (800k-1.5M daily active users) plus contextual features (S16).
- **Differentiation.** A published numeric dwell rule (50 m, >60 s, multiple pings) and visit-length buckets for the ML visits, a product now operated by Pine59 (S16, S18, S23). Per-prediction confidence scores in the ML foot-traffic schema since Q4 2025, a product now operated by Pine59 (S15, S19, S23). Per-signal forensic flags on origin and accuracy in the device-level feed (S03, S10). Unusually full self-disclosure in state registries, including the breach and government sales (S30, S31). Bundles device-level location with MAID-to-email and MAID-to-IP linkage (S01, S06).
- **Risks.** Regulatory: FTC Decision and Order C-4810 against Gravy Analytics, Inc. and Venntel, Inc., complaint and proposed order announced 3 December 2024, issued 13 January 2025, finalisation announced 14 January 2025; runs at least 20 years from issuance; bans selling or using Sensitive Location Data (narrow exceptions), requires consent records and supplier assessment, and requires deletion of Historic Location Data within 60 days unless, within 90 days, consent records are obtained or the data are deidentified or rendered non-sensitive; no admission (S25, S26, S27). Security: unauthorised AWS access discovered 4 January 2025, processing restored 9 January 2025; a subset of copied data briefly posted on a dark web forum (company statements to Datatilsynet and in its Texas registration of 9 January 2026) (S28, S31). State law: the September 2026 policy reports sale of precise geolocation without stating how sales are limited in states that ban it (S02; the state bans are recorded by workstream H, not re-opened here). Consistency: privacy policy, Terms of Service and California registry answers describe law-enforcement use differently (S02, S13, S30). Continuity: the Pine59 buyout separates the ML foot-traffic business from Unacast; contracts reportedly "continue as before" with Pine59 (S23).
- **Best fit.** Data teams that want raw, device-level US and international movement observations with forensic flags and will run their own visit logic and compliance review (S04, S05, S10). Analysts who accept projected visit estimates with a documented dwell rule and per-estimate confidence scores, and will check them against their own sales data (S15, S16).
- **Poor fit.** Buyers who need an accredited or independently validated visit count (S37, S38, S16). Buyers who must exclude any supplier under a recent FTC order or breach, or who operate in states that ban precise-geolocation sales (S25, S28, S02). Buyers seeking campaign lift measurement: no named attribution product with a method was found (S04).
- **Limits.** The ML visit numbers are projections from aggregated first-party counts and contextual features, not observed visits (S15, S16). The headline accuracy statistic (R-squared 91.6%) has no method, period or sample; the only other published check is a 2021 self-comparison at five plants and two stadiums using 2018-2019 data (S16, S11). No US monthly device figure; every volume claim is global and in a different unit (S04, S26, S33). No proportions or per-source precision for the device-level feed; fix-acceptance thresholds and forensic-flag definitions are not public (S02, S10). No MRC accreditation or other product-scoped assessment (S37, S38). Since 1 September 2026 the ML foot-traffic datasets are delivered and invoiced by Pine59, per its FAQ, while Unacast pages still list them (S06, S23). Sensitive-location suppression is asserted without a category list (S03, S06).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Source types named, no proportions and no per-source precision class. Two layers are documented. (a) The machine-learning "Visitations" datasets (documentation on docs.unacast.com, which now says it "has moved to Pine59"): the FAQ says the data are "sourced by our first party data partners via SDKs embedded within multiple phone applications, with opt-in consent from all users" and "passed to us as anonymized and aggregated person counts", that "GPS is not the only data source feeding our model", and that 1st-party GPS aggregates enter the model per H3 hexagon alongside autoregressive, time, polygon-geometry and land-use features (S16, S15). (b) Unacast's current Foot Traffic Data page (Location Intelligence APIs) says "We license pseudonymous location signals from aggregators representing SDKs across tens of thousands of mobile apps" (S05); the company-level policy names partner sources as app owners or "data aggregators who primarily receive data from software development kits", plus WiFi routers and Bluetooth beacons, and says neither Unacast nor its partners determine whether a fix came from GPS, cell triangulation or WiFi (S02). The FTC complaint (an allegation; no admission) says Gravy's suppliers "may themselves obtain the location data from other data suppliers, the mobile advertising marketplace, or mobile applications" (S25). [137, 331, 325, 106, 135] |
| Location fix and visit definition | live | C | Polygon type, dwell rule and a resolution floor documented for the ML datasets; unique visitors are counted per day, and weekly and monthly figures sum daily counts, so a repeat visitor is counted once per day visited (Pine59 documentation of the same product); no detection error statistic. Visits "estimates the amount of unique people with a stay at a location in a day", including workers and residents (S15). Dwell: "Multiple pings from the same identifier", "Within a 50m radius", "Lasting longer than 60 seconds", applied identically to every polygon (S16). Precision: signals are never assumed "to be of higher resolution than 10-20m" (S16). Places are polygons with a Polygon Confidence Score, or customer polygons (S17, S20). Visit Length uses non-resident non-worker visitors in four buckets (<=10, 10-45, 45-90, >90 minutes); locations larger than 70,000 sqm get no result (S18). Unacast's current Area Visitors API takes "a polygon and a date range" plus chosen forensic flags and returns "unique device counts" (S05), with no dwell rule on that page. Precision class sold: polygon-level visits derived from precise fixes; the ML product receives fixes pre-aggregated to H3 hexagons (S16). [331, 325, 106, 408, 324, 409] |
| Ground-truth validation | live | C | Vendor comparisons only; no independent US comparison found in the sources screened. FAQ: "When validated against ground truth data, such as the foot traffic to a popular nationwide clothing retailer, Unacast's models recorded an R-Squared of 91.6% or higher" (no places, period, sample or holdout; S16). A 2021 post (updated 2024) compares visits with stadium capacity in Kansas City and Boston (October 2018 to February 2019), with 2019 employee counts at five US manufacturing plants ("All other data matched within 20%", Everett the outlier) and with college-town seasonality; per-site results are chart images (S11). The FAQ also describes internal loss, exit-criteria and "story based" checks (S16). Pine59's documentation of the same product says the model is trained and calibrated against ground-truth visitation data, with no statistics or reference sets named. A UK study summary (CDRC, 2021) compared Unacast device counts with London footfall sensors in the 2020 lockdown period, with an industry partner and a method summary only (S39); it predates the Gravy merger and is not about the US product. Data vintages: 2018-2019 (post), 2020 (UK), undated (R-squared). [14, 106, 410] |
| Causal design | n/a | n | Not applicable: causal design sits only in the cohort A mask. No named Unacast attribution product with a method was found (register row "Unacast footfall attribution" is listed).  |
| Coverage | live | C | "In total, Visitations products come with around 4 Million places, with the exact count subject to variation with each monthly release", history from January 2019, monthly map updates and backfill (S17); the geography of that count is not stated, while the FAQ says area features are mapped "across the United States" (S16) and pricing labels the Foot Traffic Datasets and Location Insights packages "US only" (S06). Custom polygons are accepted (S20). Unacast's current Area Visitors API covers "locations of interest across the globe" (S05). Pine59's documentation names the US place layer as SafeGraph Places (third-party POIs) and states the product's coverage after the buyout as the US plus four Nordic countries. No dated US-only place count with polygons was found. [331, 332, 106, 408, 409] |
| Data access and integrations | live | C | Three route types documented, with named destinations. ML datasets: API (standard locations), DaaS file delivery to AWS S3, Google Cloud Storage, Google Cloud BigQuery or Azure Blob Storage (CSV, GZIP, partitioned folders, checksum-validated success files) and the SaaS Insights platform (S21). Current Unacast page: four on-demand Location Intelligence APIs and scheduled "Location Intelligence DaaS" delivery (S05); pricing: "APIs or batch data delivery" (S06). No public API reference was opened here; Pine59 says it now hosts the API with open documentation at console.pine59.com (S23) and no cloud-marketplace listing of the foot-traffic datasets was confirmed; the Datarade listing found covers the mobility data (S33). [331, 332, 411, 330] |
| Documented production volume | live | C | The FAQ answer to "What percentage of the US population do you observe" says "The 1st Party GPS aggregates at present are based on between 800k-1.5M daily active users" (undated page, copyright 2026; S16): a daily figure, so at most a lower bound on monthly uniques for the same population. Unacast's current APIs rest on device-level supply marketed as "more than 1 billion monthly devices across 180+ countries from 15+ suppliers" (S04), global and undefined. No US monthly device figure with definition and period was found. [328, 106] |
| US availability and coverage | live | C | Sold in the US with US observations: pricing headline "Buy location data & insights for the US", Foot Traffic Datasets and Location Insights "US only" (S06); "foot traffic datasets for the U.S." (S09); US sites in the 2021 validation post (S11); US headquarters Ashburn, Virginia (S01). US production use is not documented with named campaigns in the sources read (logo walls are not used as evidence). Since 1 September 2026 the ML datasets and Insights platform are delivered and invoiced by Pine59 (Pine59, Inc., New York), per Pine59's FAQ, with existing contracts continuing, while Unacast pages still list the packages (S22, S23, S06). [125, 332, 412, 14, 413, 326] |
| Independent assurance | live | A | No independent assessment found. No Unacast or Gravy entry on MRC's Accredited Services: Digital or Services Under Review: Digital (text search, 24 October 2026; S37, S38). The privacy policy says Unacast "is a member of the Network Advertising Initiative (NAI) in good standing, and undergoes annual reviews" (S02); NAI carries a Unacast member page dated 24 July 2021 (S32). Membership is not a product-scoped assessment with a public scope and result. [137, 414, 171, 177] |
| Method disclosure | live | C | ML datasets: six of seven elements at summary level, without data: (1) sources (first-party SDK partner aggregates plus contextual features), (2) dwell and visit definition, (3) panel size (800k-1.5M daily active users) and consent route ("opt-in consent from all users"), (4) estimation method (Pine59's documentation of the same product adds home-area market-share extrapolation, updated continuously, and a per-device signal-density correction), (5) one validation statistic (R-squared 91.6%, no method), (7) cadence (monthly data by the 5th, weekly on Wednesdays, daily since Q4 2025) and a per-estimate confidence score; (6) no detection- or segment-accuracy method (S15-S19). This documentation now carries a "moved to Pine59" banner (S15). Unacast's current Foot Traffic Data page documents only sources in outline, the polygon/date/flag inputs and delivery modes (S05). [331, 325, 106, 408, 324, 415] |
| Privacy and consent posture | live | A | current_controls evidence: the ML product says signals are aggregated at the partner, so "we do not receive precise individual locations" (S16), removes identifiers living in low-resident hexagons (S16) and excludes places "blacklisted by our Location Privacy check" (S17); company-level PrivacyCheck says sensitive-location data are "never used, shared, or resold" (S03) and "All products remove data from sensitive locations" (S06), with no category list on the pages read; opt-out form; registered data broker in California and Texas (S30, S31). Event flags: FTC Decision and Order C-4810 against Gravy Analytics, Inc. and Venntel, Inc., issued 13 January 2025 (S25); breach discovered 4 January 2025, confirmed in the company's notice and Texas registration (S28, S31). Both fall inside the three-year window. Scoping note for the scorer: the ML datasets are documented as built on aggregates from a first-party partner (S16), while Unacast's current Foot Traffic Data APIs (Area Visitors, Observations) run on the device-level signals and forensic flags that the order and breach concern (S05, S25, S28). Active obligation: the 20-year order (S25). [327, 331, 332, 106, 408, 135, 333, 251, 278] |

Performance evidence: "R-Squared of 91.6% or higher" against foot traffic to "a popular nationwide clothing retailer" (vendor assertion, sponsor Unacast) [106]; Stadium visitation "matches the approximate capacity of the stadiums on game days" (Kansas City and Boston, October 2018 to February 2019) (vendor assertion, sponsor Unacast) [14]; Workday foot-traffic estimates at five US plants "matched within 20%" of 2019 employee counts, Everett excepted (vendor assertion, sponsor Unacast) [14]; Processing removes fraudulent, problematic and duplicate data "by up to 65%"; "an average of 45% of location data is flawed or duplicative" (vendor assertion, sponsor Unacast) [327, 329]; Unacast mobility data "comparable to many alternative mobility data sources"; 70-100% decline in activity by early April 2020 (London) (synthesis, sponsor ABC Research Programme (ESRC, UBEL); industry partner Geolytix; independence from Unacast not established) [410]

### Vistar Media (active, subsidiary)

*Vistar Media Inc. (10-K; the privacy notice uses 'Vistar Media, Inc.')*

The DOOH platform where telco data now meets screens: T-Mobile owned, with partner-run visit studies and exclusive network segments.

**SWOT (2026-10-24).** Vistar, owned by T-Mobile since February 2025, sells DOOH screen targeting on partner and exclusive carrier audiences and partner-run visit studies, with no published method, validation or accreditation of its own.

- **Strengths.** Since May 2025 Vistar has offered T-Mobile carrier-derived audience segments in its DSP; an enhanced set with location insights, announced 5 October 2026, is DSP-exclusive. (vendor documentation) [334, 335] Vistar lists 18 USA-tagged audience partners and six USA-tagged foot-traffic partners, so buyers can bring an existing location vendor to DOOH. (vendor documentation) [189] The parties described 1.1 million screens from about 370 media owners at the January 2025 announcement, a large DOOH supply base. (vendor claim) [336] Its taxonomy is activatable from The Trade Desk through auction packages and PMPs, and Vistar offers device ID passback free of charge. (vendor documentation) [337]
- **Weaknesses.** Vistar publishes no visit definition, control design, minimum sample or uncertainty for attribution; one public lift case asserts an undescribed control, the other states no comparator. (vendor documentation) [337, 8, 9] No Vistar metric or segment appears on MRC's digital accredited or under-review lists, and no independent validation was found. (record) [171, 177] No production volume is documented: no panel, no campaigns-measured count; partner figures such as 200M+ users carry no definition or period. (vendor documentation) [23, 334] Its guide says targeting never relies on personal information, while its privacy notice describes receiving device location histories and selling personal information. (vendor documentation) [338, 339]
- **Opportunities.** T-Mobile told AdExchanger DOOH measurement is on its road map 'to look into for next year', which could give Vistar a carrier-based measurement product. (inference) [335] Vistar's results were immaterial to T-Mobile in 2025 while advertising revenue grew from the Vistar and Blis deals, leaving room to scale inside T-Ads. (inference) [36]
- **Threats.** The D.C. Circuit denied review of T-Mobile's US$80.08 million FCC location-data forfeiture in August 2025; certiorari reaches conference on 6 November 2026. (record) [340, 341] Vistar declares precise-geolocation collection to California's 2026 data broker registry, after answering 'No' in 2025; DROP deletion requests apply from 1 August 2026. (record) [163] Partner InMarket has been barred since May 2024 from selling or licensing precise location data, as the order defines it, a constraint on one live US measurement route. (inference) [132, 23] T-Mobile itself ties the Vistar-only exclusivity to tightening location and device-ID regulation, so further rules could narrow the carrier source. (inference) [335]

- **What is sold.** A programmatic out-of-home platform: a DSP for buyers (managed or self-serve), an SSP, ad server, player and device-management software for media owners, and traditional OOH planning software (5 October 2026 release boilerplate, S04). For this paper two services matter. (1) Audience targeting for DOOH: buyers select screens using partner audience segments (18 USA-tagged audience partners, among them GroundTruth, Foursquare, PlaceIQ, InMarket and Adsquare) T-Mobile carrier-derived segments (app ownership and engagement, available through Vistar's self-service DSP since 6 May 2025; an enhanced set adding 'location insights', announced on 5 October 2026, is exclusive to the Vistar DSP), which T-Mobile turns into area-level audience scores that Vistar matches to screens; POI proximity targeting, weather and day-part triggers. (2) Attribution: foot-traffic, sales-lift, brand and online-conversion studies run by partners (six USA-tagged foot-traffic partners, including Foursquare, InMarket and Cuebiq), each subject to a feasibility test, plus free device ID passback of exposed devices.
- **Who buys it.** Advertisers and agencies buying DOOH (the parties cited 'more than 3,000 brand partner advertisers' in January 2025, global) and media owners selling screens ('nearly 370'); buyers can also reach Vistar inventory through The Trade Desk. No buyer is named in this profile under the paper's naming rule.
- **How it earns.** Not disclosed. Vistar's results are not reported separately and were 'not material' to T-Mobile in 2025 (S01). The TTD guide says measurement studies 'have varying minimums and costs' and device ID passback is free (S15); no fee, take rate or data price was found.
- **Supply and data dependencies.** Location observations: none collected by Vistar directly in the documentation found; the Services Privacy Notice says Vistar receives device location histories, device identifiers and household-location demographics from clients, data suppliers and measurement partners, collected through cookies, SDKs and other trackers, with location from GPS, Wi-Fi or other location features (S16). Audience sources named on Vistar pages: GroundTruth (first-party, third-party and bid stream data), Adsquare (mobile location data; 'SDK-derived' in 2023), Foursquare (visitation-based segments), PlaceIQ (location and purchase data), InMarket (SDK-based location data from 200M+ users, undefined), T-Mobile (carrier signals: app ownership and engagement plus 'location insights'; AdExchanger reports mobile location data used to estimate inventory impressions, app installation and usage read from network requests and, for users who opted into location tracking, proximity to OOH ads, which T-Mobile can use to retarget those users on other channels). Consent route: asserted as opt-in or 'consented' by the vendor and partners, never documented at product level. Precision class: not stated for any source; T-Mobile segments are delivered as area scores; screen selection uses 2- to 3-mile radii in the cases. Bidstream use: named for one audience partner (GroundTruth). Exposure: the 2022 retail case cites ad log timestamps, device location, venue type and screen location; the Audience Impressions metric reported in The Trade Desk for Vistar DOOH comes from impression multipliers supplied by SSPs and media owners (commonly sourced from Route and Geopath), while plain impressions count ad plays.
- **Differentiation.** T-Mobile carrier-derived audience segments for DOOH, available through the Vistar DSP since May 2025, with an enhanced set adding location insights announced on 5 October 2026 as exclusive to the Vistar DSP (LOC-VISTAR-S04). Breadth of named partners: 18 USA-tagged audience partners and six USA-tagged foot-traffic partners in one directory (LOC-VISTAR-S13). Large DOOH supply footprint as stated by the parties: 1.1 million screens from about 370 media owners (global, January 2025) (LOC-VISTAR-S03). Vistar's taxonomy is also activatable from The Trade Desk through auction packages and PMPs (LOC-VISTAR-S15).
- **Risks.** Parent regulatory record: FCC Forfeiture Order FCC 24-43 against T-Mobile (adopted 17 April 2024, released 29 April 2024; US$80,080,000) over customer location information; D.C. Circuit denied review 15 August 2025 and rehearing 23 January 2026; certiorari petition No. 25-1422 set for the 6 November 2026 conference (LOC-VISTAR-S26, S27). Partner regulatory record: InMarket under a final FTC order since 1 May 2024 barring the sale or licensing of precise location data as the order defines it and, separately, any product that categorises or targets on sensitive locations (LOC-VISTAR-S28). State-law exposure: Vistar declares to California that it collects precise geolocation (2026 registry, last updated 29 July 2026; its 2025 registration had declared 'No') and says it sells and shares personal information; California's DROP deletion requests must be processed from 1 August 2026 (LOC-VISTAR-S16, S29). Sensitive-place exposure: a health-place example segment ('general clinics' visitors) appears on Vistar's GroundTruth page with no suppression rule documented by Vistar (LOC-VISTAR-S09, S16). Concentration: the T-Mobile segments' exclusivity is attributed by T-Mobile to 'tightening regulations' on location data and device IDs, so the source's regulatory status drives the product (LOC-VISTAR-S05).
- **Best fit.** DOOH buyers who want to choose screens by partner or carrier audience scores inside one DSP and accept vendor-described segment sources (LOC-VISTAR-S04, S13). Buyers who already use a foot-traffic vendor (Foursquare, InMarket, Cuebiq and others) and want that vendor's method extended to DOOH exposure via Vistar device ID passback (LOC-VISTAR-S15).
- **Poor fit.** Buyers who need a causal visit-lift estimate with a documented randomised or holdout control from the DOOH platform itself (LOC-VISTAR-S15, S19). Buyers who need accredited or independently audited audience or visit metrics (LOC-VISTAR-S22, S23). Buyers who must exclude health or other sensitive-place segments by documented rule (LOC-VISTAR-S09, S16).
- **Limits.** Vistar documents no visit definition, projection, control design, minimum sample or uncertainty of its own for attribution; every study rests on a partner's undisclosed-in-context method (LOC-VISTAR-S15, S19). The only published Vistar lift numbers are two vendor cases, one from 2021-22 and one undated; the first asserts a control without describing it and the second states a lift with no comparator (LOC-VISTAR-S19, S20). No documented production volume for either row: no panel, no campaigns-measured count; partner figures lack period, geography and definition (LOC-VISTAR-S03, S06). Audience targeting is DOOH-only and delivered as screen selection; device-level segment definitions are the partners', and the screen-matching step is described in one paragraph for T-Mobile only (LOC-VISTAR-S04, S15). No MRC accreditation or independent audit of any Vistar metric or segment was found (LOC-VISTAR-S22, S23). Audience impressions reported in The Trade Desk for Vistar DOOH rest on impression multipliers supplied by SSPs and media owners, a modelled input; plain impressions count ad plays (LOC-VISTAR-S15).

| Dimension | Availability | Evidence | Observed |
|---|---|---|---|
| Source transparency | live | C | Vistar documents no location source of its own for attribution; studies run on the measuring partner's data. Vistar-hosted partner pages: InMarket LCI is 'Powered by SDK-based location data from nearly 200M+ permission-based users' (no period, geography or definition) (S06); Foursquare cites '16B+ human-verified check-ins', a cumulative count (S07); Cuebiq's page names no source (S08). The 2022 retail case says studies 'involve consumers who have opted in to have their locations passively tracked' (S19). The Services Privacy Notice says clients and data suppliers collect via 'cookies, software development kits, and other digital trackers' and may collect device location 'via GPS, WiFi, or other location-based features' (S16). Exposure side: 'ad log timestamps, device location, venue type and screen location' (S19). Source types are therefore named only per partner, with no proportions or per-source precision class, and Vistar does not say which partner measures a given study. [23, 416, 417, 339, 8] |
| Location fix and visit definition | live | D | No visit definition published by Vistar on any page opened. The retail case counts visits 'during a four day look back window' (S19); the QSR case states no window (S20). The 3-mile and 2-mile radii in the cases are targeting radii for screen selection around stores, not visit geofences (S19, S20). No polygon or radius type, dwell rule, accuracy threshold or de-duplication rule is stated; the measuring partners' definitions (Foursquare, InMarket) are not inherited by this row. [8, 9] |
| Ground-truth validation | n/e | E | No comparison of Vistar-delivered visit or lift numbers with door counts, transactions or official visitor counts was found in the sources screened (Vistar news, partner pages, TTD guide, OAAA cases, privacy notice; searches listed). Record: none found.  |
| Causal design | live | D | Control asserted without method. Retail case: 'Our solutions compare control versus exposed groups' and a '6.51% lift in store visitation, compared with non-exposed consumers' (S19); minimum sample stated only as 'enough scale' and 'we do not conduct measurement with minimal participants in both exposed and control groups' (S19). InMarket page: 'LCI quantifies the incremental lift in both visits and sales' with no control described (S06). TTD guide: 'All studies require a feasibility test and have varying minimums and costs' (S15). No randomised or holdout default, significance rule or uncertainty is documented by Vistar; partner defaults are not inherited. Code: not documented (matched control asserted, no method). [23, 337, 8] |
| Coverage | n/e | D | No Vistar place base documented; the places measured are the partner's. Vistar's documented reach is screens, not places: 'more than 1.1 million digital screens provided by nearly 370 OOH media owners' (global, 13 January 2025) (S03). Record: n/e for US places. [336] |
| Data access and integrations | live | C | Routes documented by Vistar: (1) partner studies requested through a Vistar representative, each with a feasibility test (S15); (2) 'Vistar also offers device ID passback for free' (S15), with Spotzi tagged device_id_passback in the directory (S13); (3) in The Trade Desk, 'TTD can also support DOOH measurement in platform with integrated measurement partners' and buyers request study log-level data through an 'SSP Log-Level Reports Process' (S15). The directory tags six US foot-traffic partners (Accretive, Arrivalist, Cuebiq, Foursquare, InMarket, Spotzi) (S13). No Vistar measurement API was found; the developer portal lists the Marketplace DSP API (rate-limited; its OpenAPI specification lists audience, POI and geography targeting types without option schemas and contains no measurement endpoint) and the Media Owner API (S32). [189, 337, 418] |
| Documented production volume | n/e | D | Documented production volume: none. No count of studies or campaigns measured per year and no panel. Partner figures on Vistar pages carry no period or definition (InMarket 'nearly 200M+ permission-based users', S06). Supply attributes only: 1.1 million screens, about 370 media owners and 'more than 3,000 brand partner advertisers' (global, January 2025, S03). [336, 23] |
| US availability and coverage | live | C | US availability documented by vendor pages: InMarket and Cuebiq foot-traffic offers are tagged 'Live in: USA' (S06, S08); Foursquare Attribution 'is available in: United States, Canada, & Australia' (S07). US production use is documented in two OAAA-hosted vendor cases: 22 US states, flight 2021-08-15 to 2022-01-29 (S19), and twelve US markets, undated (S20). Both rest on partner measurement. [23, 416, 417, 8, 9] |
| Independent assurance | n/e | A | No Vistar entry on MRC's Accredited Services: Digital or Services Under Review: Digital, read 2026-10-24 (S22, S23); MRC's Out-of-Home accredited and under-review pages returned no listed services (S24, S25). The retail case's 'trusted research partners, rather than grading our own homework' (S19) is a vendor statement, and the partner is a commercial counterparty, not an independent assurer. [8, 171, 177, 173, 178] |
| Method disclosure | live | D | Of the seven elements: (1) sources named only per partner, no proportions (S06, S07); (2) visit definition: a four-day look-back in one case only (S19); (3) consent route asserted ('opted in to have their locations passively tracked', S19), no panel size with period; (4) projection: none; (5) validation statistics: none; (6) control design asserted ('control versus exposed') without method (S19); (7) cadence and uncertainty: none ('varying minimums', S15). At most two or three elements partly published, none with data. [23, 416, 337, 8] |
| Privacy and consent posture | live | C | Vendor documentation: the Services Privacy Notice (1 July 2026) says Vistar receives 'device location histories associated with device identifiers' from clients, data suppliers and measurement partners, uses them for 'Learning whether devices that are exposed to ads are more likely to visit certain locations', acts as a CCPA 'business' for device location histories while providing advertisement measurement as a 'service provider', does 'sell' and 'share' personal information, discloses device location histories to service providers, and does not share them with clients or with advertising intermediaries and market research companies (S16). Consent route: the case PDF asserts opt-in (S19); no product-level consent audit. Sensitive places: no suppression statement found in the notice by text search (sensitive, health) (S16). Record: Vistar Media Inc. is on California's 2026 Data Broker Registry declaring it collects consumers' precise geolocation; in the 2025 registry file, as Vistar Media, Inc., it had declared 'No' to collecting precise geolocation (S29). No regulator order, court finding or breach naming Vistar or its products was found in the sources screened; the FCC forfeiture order of 29 April 2024 is against the parent T-Mobile over its carrier location-based services program, before Vistar was acquired (S26, S27); InMarket, a US measurement partner, has been under a final FTC order since 1 May 2024 barring it from selling or licensing precise location data as the order defines it (S28). [339, 8, 340, 341, 132, 163] |

Performance evidence: 6.51% lift in store visitation 'compared with non-exposed consumers'; 533,000 store visits in a four-day look-back; 72M+ impressions (retail case, 2021-22, 22 US states) (vendor assertion, sponsor Vistar Media (vendor) and the advertiser; measured by Foursquare (Vistar partner; payment terms not stated)) [8]; 8.8% lift in foot traffic; 1.3M+ store visits; 49M+ impressions (QSR case, twelve US markets) (vendor assertion, sponsor Vistar Media and the advertiser; measured by Foursquare) [9]; InMarket LCI 'quantifies the incremental lift in both visits and sales' for DOOH audiences exposed to Vistar campaigns (vendor assertion, sponsor Vistar Media and InMarket) [23]; Foursquare (Placed) attribution lets brands 'analyze lift in foot traffic that was driven by DOOH media' (vendor assertion, sponsor Vistar Media and Foursquare (2020 release)) [187]; Store locators in DOOH creative raise stated visit intent (9% vs control, bottom placement, retail and QSR) and purchase intent (20% vs control, top placement, CPG) (vendor assertion, sponsor Vistar Media Creative Studio with Omnicom Media and Annalect) [419]



## 12. What to ask before you buy


*Location data is sold as ground truth, and the public record does not support that description. Visit counts are documented projections or have no stated basis, visit lift mostly rests on matched comparisons, no independent multi-site validation of projected retail visits was found in the sources screened, and three companies under FTC location orders still document precise or device-level offers. A buyer can still use location well, by asking eight questions before the money moves.*

> **In short.**
> Of the seven registered hypotheses, three are contradicted and four are unresolved on the public record; none is supported. Lift is not measured against a holdout by default (H3 contradicted), FTC orders did not end the sale of precise location (H5 contradicted), and no visit metric holds MRC accreditation (H6 contradicted, as of 24 October 2026). Whether panels shrank and were replaced (H1), whether visit counts match the ground (H2), what location audiences are built from (H4) and whether agents can buy by place (H7) cannot be settled from public sources. The questions below turn those findings into a test a buyer can run on any vendor.

### What the evidence says

The registered hypotheses read as follows at the evidence date.

- **Visit lift is mostly correlation.** Of fourteen cohort A products, eight document a general lift method and six of those eight use a matched or synthetic control; a ninth reports exposed-only visits with no lift claim [5, 19, 119, 17, 20, 21, 22, 25, 6, 446]. Under this paper's rules a matched comparison is correlation, and the published randomised-trial literature shows observational ad-effect estimates missing the experimental result by a wide margin [111] ([chapter 4](#validation)).
- **Visit counts are unchecked.** No independent multi-site comparison of any vendor's projected retail visits with door counts or transactions was found; the independent evidence covers parks, trails, census counts and polling places, mostly before 2021 [98, 99, 102].
- **The orders did not end the trade.** Four of the five FTC location orders bar sensitive location data and leave the sale of precise location lawful under conditions, and three of the five ordered companies document current precise or device-level offers [128, 137, 142, 145, 134, 130]. Five states ban the sale of precise geolocation outright, four of them from 2026 [152] [154, 153] [156] [155, 159] [162, 157] ([chapter 7](#regulation)).
- **Nobody accredits the number.** No visit or location metric sits on MRC's lists [171, 172, 173, 174, 175, 176, 177, 178, 179, 180].
- **Supply is undocumented.** No vendor discloses a comparable panel series across the platform changes [275, 63, 289, 427, 428, 505, 504, 88], and too few audience products name their primary source to say whether bidstream or SDK data dominates [265, 420, 421, 422, 458, 457].
- **Agents can describe a place; production buying by place is not established.** In the protocols and registry screened, the furthest any location signal reaches in agent-driven buying is one implemented connector [198, 200, 201, 205].

### The eight questions

The decision tool below gives each question the test that fits it, what the test assumes, where it fails and what it cannot tell you. It is written for a US buyer reading vendor documentation and contracts, not for a regulator.

| Buyer question | Test that fits | What the inference needs | Where it fails | What it cannot tell you |
|---|---|---|---|---|
| Is this visit number observed, projected, declared or inferred? | Ask the vendor which kind each number is and, for a projection, the panel count it was scaled from, the weighting variables and the uncertainty per estimate. A dashboard label alone does not establish whether visits are observed or projected; ask for the basis of each number. | The vendor can separate panel-observed visits from scaled estimates and will show both. | When the documentation states no basis, which was the case for most visit numbers coded in this review. | Tell you whether the projection is accurate; that needs a comparison with ground truth. [16, 17, 18, 5] |
| Does the lift rest on a holdout or randomised control, or on a matched comparison? | Ask for the default control design, the share of impressions held out, the minimum sample and the significance rule. Ask for the same campaign against the holdout and against unexposed devices: the comparator can change the lift ten-fold. | A holdout or randomised control was set before the campaign and kept clean of exposure. | When the vendor reports a matched or modelled control by default, as most products with a documented method do. A matched comparison is correlation, not causal lift. | Tell you that the visits translated into sales, or that the result generalises to the next campaign. [5, 19, 119, 17] |
| Has anyone checked these counts against door counters or transactions? | Ask for an independent comparison against door counts, transactions or official visitor statistics for the place types you buy, with error statistics, places and period. | The comparison uses data from after the 2019 to 2021 permission changes and covers your place types. | In the sources screened, no independent multi-site retail comparison exists for any named vendor; the independent checks cover parks, trails, census counts and polling places, mostly before 2021. | Tell you the error for your stores; place-level error varies with panel coverage. [98, 99, 102] |
| Where do the location observations come from, and how were people asked? | Ask for each source type (own apps, SDK partners, bidstream, telco, IP) with its share and precision class, and for the consent wording shown to users. A list of sources without proportions does not say which one carries the product. | The vendor discloses its supply and the consent route at the product level. | When the documentation says only "opt-in" or "proprietary", which is most of cohort B. | Prove the consent was valid; that is a legal question regulators have answered differently. [265, 420, 421, 422] |
| Is the vendor under an FTC order or a state law ban, and what does it still sell? | Read the order itself (what it bans: all precise location, or only sensitive locations) and the vendor's current product pages and state privacy notices. Check the data-broker registries in California, Texas, Oregon and Vermont. | The order's definitions and the vendor's public product set are both current. | When a buyer assumes an FTC order ended the sale of precise location: four of the five orders bar sensitive location data only, and three ordered companies still document precise or device-level offers. | Tell you whether the vendor complies with its order; this review does not assess compliance. [128, 137, 142, 145] |
| Is the visit metric accredited? | Read MRC's accredited and under-review lists, not the vendor's page. A vendor page can outlive its accreditation. | MRC's lists are current on the day you read them. | Today: on 24 October 2026 no visit, visitation or location metric was on any of MRC's eleven lists, and both visit data sets MRC once accredited have lost accreditation. | Tell you whether lift is causal; MRC's location guidelines say nothing about control groups. [171, 172, 173, 174] |
| For a DOOH buy, which mobility data sits under the audience number? | Ask the screen network or SSP to name the mobility provider, the panel size and period, and whether the audience is Geopath's currency or the seller's own model. | The seller knows and will name its upstream data providers. | When the seller describes "mobile data" without a provider: half the DOOH products coded here name none. | Turn a modelled audience into an observed one; every DOOH audience figure is an estimate. [182, 23, 192, 195] |
| Can my buying agent carry this location targeting end to end? | Ask which protocol version and fields the seller's agent supports, and for a test result on those fields. Schema support in a specification is not a working connector. | Both agents implement the same protocol version. | Today: the only implemented connector found maps countries, regions and DMAs and refuses postal areas, and no public interoperability test of a location field exists. | Tell you that reach or frequency by place will reconcile across sellers. [198, 200, 201, 205] |

### What would change these readings

A few public documents would move the paper's conclusions more than any amount of vendor marketing:

- An independent comparison of a vendor's projected visits with door counts or transactions across many retail places and months, with error statistics and post-2021 data, would test H2 directly.
- A current MRC accreditation of a visit or visit-lift metric would reverse H6.
- A vendor disclosing a US panel series on one definition from 2019 to 2026 would let H1 be tested instead of left open.
- A product documenting a randomised or holdout control as its default, with minimum samples and uncertainty, would add to the two found; enough of them would change H3.
- A location signal in documented production use in an agent-driven buy, with the protocol version and field named, would support H7.

### The author's positions, tested

Three positions in the author's own writing were registered as rival hypotheses, not evidence ([chapter 1](#defining)). His glossary line that footfall attribution is "correlated and modeled by default" matches what H3 found for the products that document a method. His essay on DOOH measurement calls outcome studies "a model on a model"; none of the DOOH attribution products screened documents a randomised or holdout control by default [196, 195, 194, 23, 187, 257], which is consistent with it, and half the US DOOH products coded name no mobility source for their audience numbers [182, 23, 192, 195, 608, 194] ([chapter 9](#dooh)). His case for governing reach and frequency by place on a geographic cell grid is a proposal in a working group he co-leads; it reaches no protocol state and this paper does not test it ([chapter 10](#agents)).



## 13. Method, limits and corrections


*How the paper was built, what the checks caught and what could not be reached. It is published so a reader can check the paper rather than trust it.*

> **In short.**
> AI research agents worked from one brief and one set of rules. They produced 11 research workstreams (ten from the brief and one of enforcement records added after the second outside review), 1 vendor-universe sweep and 8 company profiles. A separate AI verifier, told to refute, checked each file. The research registered and logged 793 source records and put 502 claims in the ledger: 493 research claims, every one checked by a verifier or a review, and nine more. Eight are author syntheses that state the hypothesis readings and the effect of the privacy cap; the ninth is a legal finding added in the third review, New Jersey's 2026 sale ban, which a separate check read in the enacted statute but which no verifier file covers. Verifiers made 911 checks of claims, capability records, profile fields and table rows: 619 found the item supported as stated, 266 supported it with a correction and 26 contradicted it. Another AI model reviewed the evidence and the design twice before scoring, then the full paper, then the live edition, and every factual point it raised was checked at the primary source or against the paper's own data before anything changed. The corrections ledger holds 221 entries. 47 products were scored on 11 dimensions, and every counted cell was re-checked by a separate verifier. This is AI-assisted research with a documented process. It is not an external audit, not peer review and not independently validated.

### What was fixed before the research

Three registers were written on 24 October 2026, before any research agent ran. The hypothesis register records 11 entries: seven hypotheses (H1 to H7) and four coded questions (Q1 to Q4). Each hypothesis states its population, comparator and observable outcome, and the rules for supported, contradicted and unresolved. Absence of accessible evidence yields unresolved, never contradicted. The product register lists every product scored, listed or historical, with the entity that sells it and the reason; it now holds 86 records, of which 29 are listed and 10 historical. The scoring rules fixed the cohorts, their applicability masks and anchors for every dimension, with numeric thresholds where a scale is measured.

Three things changed after the research and before any cell was scored, all logged and all dated. The product register was rebuilt from the vendor-universe sweep and the profiles: renamed products, new owners, products moved to the listed table and rows added (Foursquare's Visits feeds, two raw-data feeds). Broadsign's audiences were registered on condition that US availability be established before scoring; it was not, and the row, scored in error in the first live edition, was removed in the final live sweep. The scoring rules were amended once, on 24 October 2026, after the second outside review. Projection method moved out of the quotient into a separate module; cohort C split into foot-traffic analytics (C1) and raw movement data (C2); causal design was reworded as causal capability; documented production volume replaced scale, with a lower-bound rule for daily figures and no generic annualisation; the assurance ladders were made monotone, with one independence test for every cohort; and privacy now reports current controls with the author's risk cap beside them. Four hypotheses received dated clarifications (H3, H4, H5 and H7). The original rules are kept in data/hypotheses.json next to each clarification, with its rationale and its effect on the verdict. No clarification changed a verdict: H3 moved from unresolved to contradicted because better sources coded more products, and every other reading is the same under the original and the clarified wording.

### How the evidence was gathered

Each workstream agent read the shared rules, the brief and the registers, and wrote one structured file: sources with their publication, event and access dates, geography, period, funding and access limits; findings with the supporting passage and locator, evidence class, confidence, limitations and counterevidence; and data tables in which every row names its source, period and claim status. The rules require every cited source to be opened; a search snippet is a lead, not support. Where a page refused an automated fetch, agents retried with a browser user agent or the built-in browser and recorded the route. 67 registered source records were access-limited: partly readable, abstract only, metadata only, or not opened at all (one is a search-result lead kept for the trail). Claims resting on them say so.

Evidence classes follow the shared rules: direct record, independent measurement, commissioned measurement, vendor assertion, synthesis, inference, forecast and illustrative scenario. A vendor's own documentation is a vendor assertion, however technical. A study whose funding is unknown is not treated as independent, and a study with platform-employed coauthors is labelled as such. The author's own work is labelled and never counts as corroboration: 4 registered source records are the author's. Every absence claim is bounded to what was searched, and the searches are logged in each file.

### How it was checked

Every workstream, profile and the review-added enforcement stream went to a separate verifier that had not seen the research process and was told to refute it. Verifiers opened the cited sources themselves and returned one of four verdicts per item. A claim's final wording is its verifier's or a review's correction where one exists; contradicted claims appear only in corrected form. Every research claim was checked. Of all 502 claims in the ledger, 314 stand as stated (the syntheses among them), 174 with a qualification and 14 contradicted.

![Every stage was built to remove things, and the counts are computed, not typed.](figures/fig-pipeline.svg)

*Figure fig-pipeline. **Every stage was built to remove things, and the counts are computed, not typed.** The five stages of this project with the actual outputs of each. Every research claim in the ledger was checked by a verifier or a review; the nine other entries are eight author syntheses, which state the hypothesis readings and the privacy-cap effect, and one legal finding added in the third review (New Jersey's 2026 sale ban), read in the statute by a separate check but not covered by a verifier file. Every counted score cell was re-checked by a separate verifier, and the rows nearest a band line were scored a second time, blind. Corrections include all four outside review rounds. Units: counts. Scope: this research project, as of the cutoff. Caveat: Counts describe process volume, not evidence quality; a claim that survived verification is supported by its source, not proven true. Data: data/stats.json*

The research ran on two models. A usage limit stopped the first fan-out after nine workstreams and one verifier had finished; the remaining two workstreams and ten verifiers were rerun on a second model, and the finished files were kept.

There were two outside review rounds before scoring, both by another AI model run by the author. The first spot-checked the evidence and reviewed the design; it returned 119 items. The second, a follow-up in the same conversation, tested the proposed fixes, recounted H3 product by product and withdrew or narrowed ten of its own earlier items; it returned 37. Every factual item in both rounds was checked at its primary source by eight separate checks before any change. The reviewer was right on substance far more often than not, but several of its factual assertions were wrong when checked: among them the list of signatories to the NAI's precise-location standards, its statements that a filing, a state act, a registry and a breach notice to a regulator could not be opened, and its claim that per-agent compliance data in the AgenticAdvertising.org registry is private. Each item is logged in the corrections ledger with the check that settled it. The largest changes were to the scoring rules (above), to the H3 count, to the state-law record (Connecticut and Virginia now ban the sale of precise geolocation) and to the enforcement record, which gained the FTC's order against General Motors and OnStar, further state actions and the data-broker registries. A third round, on the full draft, returned 37 findings, and a final sweep of the published edition returned 28; each was checked at its primary source or against the paper's own data and is logged in the corrections ledger. The final sweep produced a corrected edition on 24 October 2026: it removed one product scored in error, corrected eight score cells and fixed two defects in the web page's evidence buttons.

### How the products were scored

The scoring unit is a named product that serves a stated buying task, mapped to one record in the product register. There are five cohorts: visit measurement and attribution (A), location audiences and targeting (B), foot-traffic and visit analytics (C1), raw movement data (C2), and geofencing and visit-detection SDKs and APIs (D). Each has an applicability mask; causal design applies only to cohort A. n/a comes only from the mask, never product by product.

Each cell was scored from 0 to 5 against the dimension's anchors for that cohort, then capped by the evidence: vendor documentation caps a cell at 4 and vendor marketing at 3. A 0 means a source documents absence or failure; missing evidence is n/e, never 0. Seven AI scoring agents scored the rows in batches from the fixed rules and the verified evidence. A separate verifier for each batch re-opened every counted cell's sources and confirmed, corrected or rejected it: 397 counted cells, none left unverified. A product enters the matrix with at least four counted cells; none fell short. The numeric thresholds in the anchors, such as US places or monthly devices, are the author's assumptions, fixed before scoring.

The rows within two points of a band line were scored a second time, blind, by separate AI scorers given the rules and each row's cited sources and verified claims, but not the first scores, grades or rationales: sixteen rows and 166 cells in cohorts A, B, C1 and C2. The two passes matched exactly on 81.9% of the cells and within one point on 96.9% of the cells both scored (quadratic-weighted kappa 0.90). With every second-pass score substituted at once, one row changes band: Simpli.fi Addressable Geo-Fencing, from Partly to Well documented, which the matrix marks †. One material disagreement on a row whose band it would have moved was adjudicated by re-reading the rule: Kochava's causal design stays n/e, because its documentation offers a synthetic control as one of three report types rather than as the default. Both passes are published in data/second-pass.csv. The published scores started from the first pass and carry the later corrections, and the † marks that historical result only; it is not recomputed on the corrected matrix. [Chapter 11](#gq) sets out the Ground Quotient, its field mean, bands and sensitivity tests. Its field mean is the mean of a cohort's Raw scores, which differs from the pooled construction of the author's Contextual and Native Quotients, so the numbers are not comparable across papers. The agreement figures compare the first scores as they stood when the second pass ran, before the third outside review and the final live sweep corrected some cells; data/second-pass.json keeps that snapshot, labelled as such, and the published matrix carries the later corrections.

### Limits

The paper scores what is documented in public by the cutoff, not what products do in private. Pricing, contract terms, panel economics, adoption and market shares are almost never public, and the paper lists them as unanswered rather than estimating them. A vendor with a thin public record scores lower than its product may deserve.

Status-sensitive facts (accreditation, ownership, products on sale, laws, protocol versions and the agent registry) were read on 24 October 2026 and are dated in the text. The registered evidence freeze was 30 October 2026. The privacy dimension's three-year window moves with the cutoff, to 24 October 2023 through 24 October 2026; no qualifying event lies in the six days either side. The paper was published before that date, so, as with the attention paper in this series, the author set the cutoff to 24 October 2026, the date the status-sensitive facts were read; the hypotheses, product register and scoring rules were not changed by the move. Anything that changes after the cutoff will be logged as a dated correction, not folded in silently.

The United States is the primary market. The US availability dimension counts documented US availability only; international breadth is recorded but earns nothing.

Several documents could not be read in full: InMarket's lift-methodology glossary (behind a sign-in), Oregon's data-broker register (behind search forms), and pages that refused automated access. Claims resting on summaries of them say so. IAB's DOOH Measurement Guide, gated on IAB's site during the research, was later found as a public PDF and read in full in the final live sweep of 24 October 2026.

### Interests

The author has worked at Samba (TV measurement and advertising data) since May 2025. No cohort A documentation opened names Samba as a source of TV exposure data for TV-to-visit measurement; Foursquare and Cuebiq measure linear TV without naming their viewing-data supplier [22, 119, 274] ([chapter 5](#measurement)). He co-leads the Signals & Measurement working group of AgenticAdvertising.org, which maintains AdCP, one of the protocols assessed in [chapter 10](#agents); a geographic cell-grid proposal he authored is open there and earns no protocol state in this paper. No Fluff Advisory, the author's firm, operates agents listed in that organisation's registry; in the registry data read for chapter 10, none declares a DOOH specialism or has a geo storyboard result [200, 597]. Earlier employers include Verve Group, whose lineage includes Verve Wireless, a location-advertising company whose assets Media and Games Invest agreed to acquire in January 2020 [495, 496]. Verve is credited as a contributor to IAB's 2017 location-marketing playbook cited in [chapter 8](#standards) and shows no location product in the sources screened. The others are Hearts & Science and Lotame, which also shows no location product in the sources screened [524, 525, 526]. No vendor funded, reviewed or saw the paper before publication. Every vendor is scored under the same published rules. The author's own essays and glossary argue positions on location and DOOH measurement that this paper tests as hypotheses, not as evidence.

### The claims

Every claim marked in the text is listed here with its evidence class, review outcome, confidence and supporting passages.

**Claim ledger**

| ID | Claim (final wording) | Ch. | Evidence class | Review | Confidence | Sources |
|---|---|---|---|---|---|---|
| LOC-C-F22 | Arrivalist's 2021 Lee County analysis scaled 492,897 panel trips to 9,861,477 projected US adult trips (a factor of 20.0) and 932,706 panel point-of-interest trips to 22,730,788, with only 60,609 'panel attributed arrivals' tied to media exposure, from a panel described as GPS data 'accurate to 30 feet' balanced monthly to the US Census and excluding international visitors, under-18s, trips under 50 miles or under 4 hours, one-way trips and commuters. | 1 | vendor assertion | supported as stated | high | [4] |
| LOC-E-F06 | LCI's reported store visits are projected: the console defines Overall Projected Store Visits as 'Exposed Device Visits * ND Device / DMA Size Ratio' (a rate-up of panel visits to DMA population) and Incremental Projected Store Visits as that total multiplied by the LCI Lift; the General Population is defined as US only. The console text is undated and NinthDecimal-era. | 1 | vendor assertion | supported with qualification | medium | [17] |
| LOC-E-F09 | LCI is marketed as live in 2026 (InMarket's LCI page, Vistar's US partner page and TikTok's 2024 partner post), with a Retail Sales Lift Attribution add-on that uses card transactions from a subset of the same exposed and control users; Vistar's page states 'Live in: USA' and 'nearly 200M+ permission-based users' from SDK-based location data. | 1 | vendor assertion | supported as stated | high | [423, 11, 23] |
| LOC-E-F10 | GroundTruth's Observed Visits metric is an observed, exposed-only count with no control group: a visit is a user visiting the brand or drive-to location within 14 days of exposure, credited last-touch, verified after 7 days, and the help centre states that observed visits may be lower than actual visits because not all users are trackable. | 1 | vendor assertion | supported as stated | high | [6, 18] |
| LOC-E-F11 | GroundTruth's Projected Visits is a panel-based probabilistic projection: a panel of impressed devices with frequent location signals is built per campaign and its visitation applied to all impressed devices, anchored in Observed Visits including first-party WeatherBug data, shown only once a campaign reaches about 40 Observed Visits and 350,000 impressions, cumulative, with no uncertainty reported. | 1 | vendor assertion | supported as stated | high | [18] |
| LOC-E-F14 | Blis announced the global release of Smart Holdout Groups on 9 October 2024. It is a geographic holdout design: machine learning splits an audience into statistically balanced target and control zip codes on demographics and income, allows for commuting, and inserts buffer zones where ads are neither served nor measured. Blis does not describe the assignment as randomised and does not say the design is the default for all its visit-lift reporting. Footfall lift is 'Only available in the US', and the page documents no data source, visit definition, minimum size or significance rule. | 1 | vendor assertion | contradicted | high | [19, 424] |
| LOC-E-F19 | Unacast's current homepage and product navigation name no attribution, lift or campaign-measurement product and no Gravy-branded product; 'measure offline marketing attribution' appears only as a use of its Location Data product, so this review found no named 'Unacast footfall attribution' product in the sources screened by the cutoff. | 1 | synthesis | supported as stated | medium | [125] |
| LOC-U-F04 | Foursquare's 2026 product menu lists Attribution, Audience and Proximity for marketers, the four-tool Geospatial Intelligence Platform, Dev Console, Places API and Movement SDK for developers, Places POI Data, Open Source places and Business Listings, plus the Swarm and Superlocal apps; no product named &quot;Visits&quot; appears. The count of &quot;seven location products&quot; is the researcher's grouping, not the vendor's. | 1 | vendor assertion | supported with qualification | high | [347] |
| LOC-U-F15 | Of the location suppliers screened, only Quadrant documents both a first-party server-to-server collection route and explicit (global) counts (1+ billion opted-in devices, 50+ billion daily location events); Outlogic describes &quot;first-party geolocation data through exclusive partnerships&quot; and &quot;No Bidstream Data&quot; but gives no count; Complementics (a discovery, not a seed vendor) claims &quot;over 1 billion global devices&quot; from &quot;multiple supply sources&quot; without naming the route. None gives a US-only count or a period. | 1 | vendor assertion | contradicted | medium | [425, 426, 128] |
| LOC-U-F22 | Placer.ai (Placer Labs, Inc.) sells an Analytics Platform, Enrichment and a Data Marketplace at the cutoff, and its advertising page offers to 'measure lift or incrementally' from near real-time visitation data without naming a separate measurement product, so the register keeps one cohort C row and notes the measurement use. | 1 | vendor assertion | supported as stated | medium | [293, 294] |
| LOC-BLIS-F04 | Blis's 2024 strategic report says the group has 'key partnerships with telco providers across North America, the UK, and Europe' and names Smart Holdout Groups ('Smartholdout') as patent-pending measurement technology developed in the year; it reports 95% customer retention (undefined). | 1 | direct record | supported as stated | medium | [226] |
| LOC-FOURSQUARE-F17 | Foursquare defines a Visits-feed visit as a stop of at least two minutes snapped to a venue by likelihood, using place shapes in dense locations, filters home and work visits from the Research feed, and acknowledges 'inaccurate snaps' such as visits recorded at closed stores, without publishing an accuracy threshold or error rate; its Movement SDK FAQ instead says there is 'no standard minimum dwell time'. | 1 | vendor assertion | supported as stated | high | [241, 249, 354] |
| LOC-GROUNDTRUTH-F11 | GroundTruth's place base is 'Blueprints', polygons traced from satellite imagery in three layers (in-store, on-lot, retail block), counted at 4.8 million across the United States and Canada, with no US-only count; other GroundTruth pages give 'over 5 million', '5 million ... globally' and '6 million ... globally'. POIs that cannot be Blueprinted are measured with a standard 12-meter radial fence, so not every measured place is a polygon. | 1 | vendor assertion | supported with qualification | medium | [252, 253, 258, 358, 361] |
| LOC-PLACER-F01 | Placer.ai describes its population visit estimates as extrapolations from a device panel: its documentation calls the method &quot;a proprietary extrapolation algorithm&quot; and illustrates it with 74.8 thousand panel device visits to one Michigan grocery store scaled to an estimated 2.4 million visits. Under this paper's definitions those population totals are projected numbers; Placer also exposes raw, unextrapolated panel counts (panelVisits, panelVisitors), which are observed panel figures, not population estimates. | 1 | vendor assertion | supported with qualification | high | [3, 283, 386] |
| LOC-PLACER-F04 | The September 2024 deck's two panel figures imply different denominators: 25 million at about 9% implies a base near 278 million, while Placer's help page puts the panel at about 7% of the US population; the help-page example (74.8 thousand panel visits to 2.4 million estimated) implies a scale factor of about 32, or panel visits near 3.1% of estimated visits. | 1 | inference | supported with qualification | medium | [289, 3] |
| LOC-PLACER-F05 | Placer documents a visit as a cluster of device signals matched to a geofenced POI (visits are also captured outside the geofence by undisclosed venue-detection logic), counted at most once per person per POI per day in the visits metric (breakdown endpoints' estimatedFoottraffic can count the same device more than once a day, and traffic-pin impressions count every pass), captured from 1 minute of dwell since Data Version 2.0 (31 March 2024), with a per-category default duration segment; no horizontal accuracy threshold or detection-error statistic is published. | 1 | vendor assertion | supported with qualification | high | [284, 283, 285, 287, 382] |
| LOC-PLACER-F06 | Placer's published API separates raw panel counts from projections: the glossary defines panelVisits and panelVisitors as &quot;raw, unextrapolated data from our mobile device panel&quot; used as model inputs, alongside extrapolated visits and visitors. | 1 | vendor assertion | supported as stated | high | [283, 3] |
| LOC-RADAR-F05 | Radar documents its geofence detection rules: events are generated server-side from SDK or API location updates; geofences may be circles of 10 to 10,000 metres, polygons of 4 to 2,000 coordinates or isochrones; each event carries a confidence level from 1 to 3 computed from the device-reported accuracy and the geofence geometry; optional stop detection, accuracy-based entry and exit buffers and minute-based dwell thresholds are configurable. | 1 | vendor assertion | supported as stated | high | [298] |
| LOC-UNACAST-F13 | Unacast's machine-learning visit numbers are projections: documentation defines visits as an estimate of &quot;unique people with a stay at a location in a day&quot;, produced by a model with &quot;more than 150 features&quot; whose GPS input is aggregated person counts per H3 hexagon from &quot;between 800k-1.5M daily active users&quot;. | 1 | vendor assertion | supported as stated | medium | [325, 106] |
| LOC-UNACAST-F14 | Unacast's documentation gives a numeric dwell rule for its ML products, multiple pings from one identifier within a 50 m radius lasting longer than 60 seconds, applied identically to every polygon, and says signals are never treated as finer than 10-20 m. | 1 | vendor assertion | supported as stated | medium | [106] |
| LOC-C-F21 | Panel-size figures published by vendors in this stream are quoted in incompatible units and mostly undated: Advan (11 June 2019) 'approximately 200 million cellphone devices, including 50% of the cellphones and trucks in the US', from 1,500 apps; InMarket's LCI glossary (undated) '1B devices or 110M uniques' over 13 months 'across 110+ locations' in the NinthDecimal Location Graph, US only; Placer (undated) 'tens of millions' of US devices, about 7% of the US population; Mobilewalla (2025 footer) 2.2 billion unique devices across 40 countries 'collected from the ad tech ecosystem, third-party data providers, app publishers, and other sources'; Cuebiq (2017) a database of 61 million anonymous users. | 1, 2 | vendor assertion | supported with qualification | high | [117, 17, 3, 15, 118] |
| LOC-FOURSQUARE-F20 | Foursquare retired its City Guide app on 15 December 2024 and its web version on 28 April 2025, AdExchanger reports it bought the Superlocal app in 2025, and Infillion says it will 'evaluate the future' of Swarm and Superlocal; the consumer apps are the source of the check-ins Foursquare uses as visit ground truth and of its first-party panel. | 1, 2 | synthesis | supported as stated | medium | [64, 51, 52, 236] |
| LOC-D-F41 | In the filings, official statements and vendor pages this review opened, no US location vendor discloses a US panel series on one definition spanning a pre-2021 and a post-2023 figure. The dated figures use incompatible units: SafeGraph's July 2020 carrier breakdown, InMarket's 'about 100 million unique devices' a year, Gravy's daily signals from about a billion devices, and Placer's undated '30M Active Devices' and '25M+ devices' in a September 2024 deck. H1 is therefore unresolved on this evidence. The absence holds only for the sources screened, and none of these figures shows a market-wide contraction or substitution. | 1, 2, 3 | inference | supported with qualification | medium | [275, 63, 289, 427, 428] |
| LOC-E-F17 | Azira's measurement panel is documented as a blend of SDK integrations and bidstream location signals from the bid requests of over 100,000 apps, filtered by device activity thresholds, with places as manually drawn polygons from satellite imagery; no panel size and no proportion of SDK versus bidstream supply is stated. | 1, 2, 5 | vendor assertion | supported as stated | high | [5] |
| LOC-FOURSQUARE-F10 | Foursquare documents the sources behind Attribution as GPS pings from its own consumer apps and vetted third-party partners (each third-party provider calibrated against overlapping first-party users), exposure by pixel, SFTP or S3, TV exposure inferred by joining viewership data with ad run-logs, OOH exposure inferred by joining its movement data with billboard locations, and over 2 billion annual card transactions mapped to places across 450+ US chains; no proportions or precision class by source are published. | 1, 2, 5 | vendor assertion | supported as stated | medium | [236] |
| LOC-E-F32 | No opened cohort A documentation names Samba as a source of TV exposure data: Foursquare and Cuebiq document linear-TV-to-visit measurement, Cuebiq from 'impression logs from major linear broadcast partners' and Foursquare across 'TV (both linear and connected/advanced)', without naming the viewing-data supplier. | 1, 2, 5, 13 | synthesis | supported as stated | medium | [22, 119, 274] |
| LOC-G-F01 | Apple announced a one-time location option and background-use information for iOS 13 on 3 June 2019 and shipped them in iOS 13, released 19 September 2019, as 'Allow Once' location permission, background tracking alerts and Wi-Fi and Bluetooth protections against location inference. | 1, 3 | vendor assertion | supported as stated | high | [65, 66, 67] |
| LOC-G-F02 | iOS 14 (announced 22 June 2020, released 16 September 2020) let users grant apps approximate instead of precise location; Apple's WWDC20 session states apps cannot opt out, that reduced-accuracy fixes are quantised to a region with a typical radius of about five kilometres (a couple of kilometres in dense cities, ten kilometres or more in rural areas), recomputed about four times per hour, and that region monitoring requires full accuracy. | 1, 3 | vendor assertion | supported as stated | high | [68, 69, 70] |
| LOC-G-F06 | iCloud Private Relay, released as a public beta with iOS 15 on 20 September 2021 for iCloud+ subscribers who enable it, replaces the client IP with a relay IP that maps to a roughly city-level area by default ('Maintain general location'; the location hint used to choose that IP is a four-character geohash of roughly 800 km2) or only to country and time zone, and covers Safari browsing, DNS queries and insecure HTTP app traffic, leaving HTTPS in-app traffic and its IP geolocation untouched. | 1, 3 | vendor assertion | supported with qualification | high | [71, 72, 429, 430] |
| LOC-G-F07 | Android 10, publicly released on 3 September 2019, introduced the separate ACCESS_BACKGROUND_LOCATION permission and the option to allow location only while an app is in use; apps targeting Android 9 or lower had background access added automatically, and Wi-Fi and Bluetooth scanning APIs began to require fine location for apps targeting Android 10, with MAC addresses randomised by default. | 1, 3 | vendor assertion | supported as stated | high | [82, 83] |
| LOC-G-F09 | Android 12 (source pushed to AOSP on 4 October 2021) lets users grant only approximate location even when an app requests precise location, offers Precise and Approximate options in the permission dialog, lets users set the accuracy for any app in Settings regardless of its target SDK, and Google documents approximate location as accurate to within about 3 square kilometres. | 1, 3 | vendor assertion | supported as stated | high | [86, 87, 431] |
| LOC-G-F13 | Google Play's Data safety section was pre-announced on 6 May 2021, became visible to users on 26 April 2022 and mandatory for all apps after a grace period ending 20 July 2022; declarations must cover data collected through third-party SDKs, define approximate location as an area of at least 3 square kilometres and precise location as less than 3 square kilometres, and require disclosure of location inferred from IP address. | 1, 3 | vendor assertion | supported as stated | high | [432, 89, 90] |
| LOC-G-F14 | Google's Authorized Buyers Program Guidelines (last updated 15 May 2026) state that Google does not share precise location data through the program; that any precise location a buyer otherwise receives (GPS, Wi-Fi or cell-tower data) may be used only to bid on that impression and kept no longer than that purpose needs; that buyers, advertisers and their vendors may not sell any Program Data; that Program Data may not be used for ads personalisation or profile building where the user or seller has not allowed it; and that non-aggregated Program Data may be kept at most 18 months unless law requires otherwise. | 1, 3 | vendor assertion | supported with qualification | high | [433] |
| LOC-G-F16 | Chrome's IP Protection was designed for Incognito mode only, for third-party requests to domains on a Masked Domain List, through a two-hop proxy assigning IPs that represent coarse location including country, with a public city-level geofeed (proposed threshold of one million unique cookies per area over two weeks, about 700 US areas). Google said it would reach Chrome Stable no sooner than July 2025, listed it for retirement on 17 October 2025, archived the explainer repository on 3 November 2025, and by 14 August 2026 classed it as 'Discontinue: scheduled for phaseout' with no date; whether it reached Stable is not established. Its effect on IP geolocation for advertising is confined to Incognito third-party contexts and is being withdrawn. | 1, 3 | vendor assertion | supported with qualification | medium | [434, 435, 436, 91] |
| LOC-RADAR-F06 | Radar’s SDK tracking presets set the stop rule and requested OS accuracy: on iOS the Responsive and Continuous presets treat a device as stopped after 140 seconds within 70 metres, Continuous requests kCLLocationAccuracyBest while Efficient and Responsive request kCLLocationAccuracyHundredMeters (the default &quot;medium&quot;), and on Android all three presets use 140 seconds and 70 metres. | 1, 3 | vendor assertion | supported as stated | high | [300] |
| LOC-A-F18 | Prebid's documented coordinate coarsening happens only when the transmitPreciseGeo activity is denied, either by a publisher rule or by a configured privacy module (such as the GPP usnat control module) translating a consent string; then Prebid Server truncates lat and lon to two decimals and masks IPv4's rightmost 8 bits (PBS-Java 3.0 also removes metro, city, zip, accuracy, ipservice and ext) and Prebid.js allows only two-digit precision; the default is allow, and the Prebid Mobile Android SDK's setLocationDecimalPrecision defaults to no limit. | 1, 3, 8 | direct record | supported with qualification | medium | [437, 438, 439] |
| LOC-B-F20 | Placed's 2015 vendor analysis of three weeks of exchange bid requests (about 4 billion impressions a day across about 50 million devices), against its own first-party panel (about one in 450 US adults), found that 27% of daily bid requests carried any location; the column gives the share of located requests precise enough to identify a business inconsistently (it says both that 12% were too imprecise and that only 12% could place a device at a specific address; its filtered count of 30 million reliable locations a day fits the latter); 91% of observed business visits generated no ad request, less than 1% generated an ad-request location within 100 m, and about 73% of bid requests occurred in transit. | 1, 4 | vendor assertion | supported with qualification | medium | [440] |
| LOC-B-F29 | Placer.ai's help-centre documentation (undated, read 2026-10-24) describes a panel of tens of millions of US devices from hundreds of partner apps, about 7% of the US population (55% iOS, 45% Android); its September 2024 Salt Lake deck gives two different figures, 30 million active devices from 500 SDK apps on one slide and 25 million-plus devices, about 9% of the population, on another, neither defined by period; its visit numbers are projections from a proprietary extrapolation algorithm (an example scales 74.8 thousand panel visits to 2.4 million), and the opened accuracy pages name validation sources (first-party data, transactions, store revenues, counters, MLB attendance) without publishing any result or error statistic. | 1, 4 | vendor assertion | supported with qualification | high | [3, 102, 289] |
| LOC-E-F12 | GroundTruth documents its places as Blueprints, polygons traced from satellite and street-view imagery in three layers (in-store, on-lot, retail block), 'over 5 million' in number, verified against 'trusted location sources' it does not name; it asserts without a cited study that radial geofences measure visits 'with up to 75% inaccuracy'. | 1, 4, 5 | vendor assertion | supported as stated | medium | [254, 361] |
| LOC-B-F18 | Thinknear's vendor-run Location Score Index, the earliest public series on bidstream location accuracy found in the sources screened, reported in 2014 that 67% of mobile ad inventory carried latitude and longitude and that 34% of location-bearing impressions were within 100 metres of the user's location (bands as published sum to 93%; the base is not stated); the Q1 2015 edition (more than one billion impressions sampled; 62% of ad requests carried location) put 37% within 100 m, 9% at 100-1,000 m, 26% at 1-10 km, 18% at 10-100 km and 10% beyond 100 km, with an industry score of 50 against 51 in Q4 2014. | 1, 4, 6 | vendor assertion | supported with qualification | medium | [441, 442, 443] |
| LOC-A-F17 | Google's Authorized Buyers documentation (last updated 17 September 2026) states that bid-request location is obtained only from IP geolocation, never from GPS or other sources, that Google limits lat and lon precision to 0.01 degrees, that it provides only a coarse geolocation shared by a sufficiently large number of users, and that for city-level location 'accuracy' is the radius of a circle with the same area as the city. | 1, 4, 6, 8 | vendor assertion | supported as stated | medium | [444] |
| LOC-A-F05 | The guidelines state that two decimal places of latitude and longitude place a device within about one kilometre, four within about ten metres and five within about one metre, warn that decimal count must not be equated with accuracy, encourage first-party 'ground truth' validation, and set a general expectation of less than 5% error on primary reported characteristics with disclosure above that. | 1, 4, 8 | direct record | supported as stated | high | [445] |
| LOC-J-F05 | Hierarchical geographic cells are not part of AdCP at the cutoff versions: issue #5969 (geo_cells / geo_cells_exclude with an h3, s2, geohash or custom cell-system enum), opened by the paper's author on 21 July 2026, is OPEN with milestone 3.4.0 and labels rfc and needs-wg-review, its two companions (#6013 conversion, #6014 cell-grain delivery reporting) are also OPEN on milestone 3.4.0, no geo_cells pull request was found, and neither v3.2.6 nor v3.3.0-beta.0 contains a geo_cells field; under the brief's rules the proposal earns no protocol state. | 1, 4, 8, 9, 10 | direct record | supported as stated | high | [205, 206, 207] |
| LOC-C-F01 | Foursquare Attribution's documented default control is a modelled 'expected visit rate' built from '10+ core attributes' (a 'modeled twin cohort' per its 30 September 2024 blog), which codes as matched or synthetic control, not a randomised or held-out group, as of the documentation's 12 February 2026 update; the product page calls the output 'incremental lift', while the blog and the metrics page call it 'behavioral lift', and the Attribution FAQ in the same documentation set gives a worked example with a 14-day conversion window plus 7 days of reconciliation. | 1, 5 | vendor assertion | supported with qualification | high | [119, 236] |
| LOC-C-F09 | Cuebiq's current marketing (undated page, 2026 footer) describes its incrementality method as 'causal machine learning to simulate a randomized control trial' with 'behaviorally matched control groups' matched on location patterns, device type, prior visitation behaviour and media exposure history, which under this paper's codes is a matched or synthetic control, not a randomised design, however it is named. | 1, 5 | vendor assertion | supported with qualification | high | [446] |
| LOC-C-F12 | GroundTruth's Observed Visits metric (help article dated 31 March 2025, whose FAQ describes a verification-period rollout 'as of September 17 2024') counts a user visiting the brand or 'drive-to' location within 14 days of exposure, credited by last-touch, with a 7-day verification lag before 'verified visits' appear; the help centre states that observed visits 'may be lower than actual visits', its Blueprints article (25 February 2026) documents the visit place as a satellite-traced polygon with in-store, on-lot and retail-block layers across '4.8M Blueprints' and calls the metric a 'raw' observed count; the metric carries no control group, and the Cost Per Visit model bills on these verified visits (Toyota case, 15 November 2024: 'over 1,200 visits' in a month across DE, PA and NJ). | 1, 5 | vendor assertion | supported with qualification | high | [6, 367] |
| LOC-C-F25 | Attribution windows disclosed by vendors range from 4 days (Vistar/Foursquare DOOH, Target) through 7 days for QSR and 14-30 days for auto (Azira), 14 days plus a 7-day verification lag (GroundTruth), a 14-day example conversion window plus 7 days of reconciliation (Foursquare's Attribution FAQ), two weeks after versus two weeks before exposure (InMarket) to a 30-day look-back (Simpli.fi); Cuebiq's 2017 sports-retailer case shows 37% of exposed visitors arriving within 2.6 days and 14% between 10.5 and 13 days, so window choice alone moves a visit count materially. | 1, 5 | synthesis | supported with qualification | high | [8, 5, 6, 17, 10, 118] |
| LOC-E-F02 | Foursquare's attribution visit volumes are projected, not observed: the 2022 method post documents census weighting of the panel to the national population and projection of visit volumes on the basis of match rates by platform and partner. | 1, 5 | vendor assertion | supported as stated | medium | [16] |
| LOC-E-F18 | Azira's total visits are projected from observed visits by a Poisson detection model: each observed device represents 1/P(Detection) devices, extrapolation runs only when observed visits exceed about 1,000, Bayesian priors apply and the multiplier is capped at 20x; the product page lists mobile/desktop, audio, video, CTV and DOOH as channels, a conversion pixel and API pushes to DSPs as routes. | 1, 5 | vendor assertion | supported as stated | high | [5, 447] |
| LOC-E-F20 | Adsquare's Attribution Dashboard, launched 30 June 2025 inside OnePlatform as part of its 'legacy Attribution solution', documents a simulated control ('Control Condition' or 'Synthetic Control Condition') adjusted for weather, seasonality, local footfall trends and public holidays in place of control groups, consented SDK movement data, visit matching by lat/long and 2D POI polygons with 'no fuzzy radius logic', and 'net visitation uplift' as the headline metric; the US is not named as an attribution market and no minimum sample or significance rule is stated. | 1, 5 | vendor assertion | supported as stated | high | [448, 21] |
| LOC-E-F26 | Arrivalist's MediaInsights documents a matched control of cardholders not exposed to the campaign, card transaction data from Visa, Mastercard and Discover, and exposure across digital display, CTV, social, native and DOOH in every US media market; the partner page says arrivals are verified with 'no panels, no proxies' while the TripInsights page says 'our panel holds steady', and the live pages render their headline statistics as zero placeholders. | 1, 5 | vendor assertion | supported as stated | medium | [25, 449, 450, 451] |
| LOC-GROUNDTRUTH-F24 | GroundTruth's help centre (2021, not since updated) says CTV-to-store visits rely on a partnership with TapAd converting IP addresses to mobile ad IDs, so CTV visit attribution adds an IP-to-device match before the location step. | 1, 5 | vendor assertion | supported as stated | medium | [452, 453] |
| LOC-INMARKET-F02 | InMarket’s 29 July 2026 description of Visit Lift says its counterfactual is a synthetic control, a statistical model of expected visitation built from historical patterns, and stops short of the mechanics; the vendor calls this causal, but under this paper’s rules a synthetic or modelled control is coded matched_or_synthetic, not randomised. | 1, 5 | vendor assertion | supported as stated | high | [121] |
| LOC-A-F07 | The 2017 Location-Based Advertising Measurement Guidelines contain no provision on visit lift, incrementality or control groups (no occurrence of 'lift', 'control group' or 'incremental' in the text); MRC addresses foot traffic as an outcome and incrementality (including RCTs) in its separate Outcomes and Data Quality Standards (September 2022), which require foot-traffic outcomes to follow the 2017 location guidelines. | 1, 5, 8 | direct record | supported with qualification | high | [445, 454] |
| LOC-E-F28 | Of fourteen cohort A seed products, nine document a general visit-measurement method that governs their ordinary reporting, and eight of them claim lift. Of the eight, six use a matched or synthetic control (Foursquare, InMarket, Reveal, Adsquare, Cuebiq, Arrivalist) and two a randomised or holdout control (Azira's randomised PSA control, documented as the default; Blis Smart Holdout Groups, a machine-balanced geographic holdout whose assignment is not documented as random). The ninth reports exposed-only visits with no control and no lift claim (GroundTruth Observed and Projected Visits). Five document no default (Vistar, a partner wrapper; Kochava, where MediaLift synthetic control is one selectable option; Precisely; Unacast; Lifesight, whose location module is not on its current site). Seven of the nine documented visit-measurement methods are non-holdout. | 1, 5, 8, 9 | synthesis | supported with qualification | medium | [5, 19, 119, 17, 20, 21, 22, 25, 6] |
| LOC-F-F08 | Adsquare sells two targeting products in 2026 under names that replace the register's 'Audience Management Platform': ID-Precise Audiences (MAID-based segments from 'verified visits, offline purchases', with home and work inferred at postal-code level and look-back windows of 14, 30, 90 and up to 180 days) and Geo-Contextual Audiences (no device ID; a 'postcode, H3 cell, polygon' scored by the hour from household, census, point-of-sale and weather data), both activated in The Trade Desk, DV360, Adform and 'all other connected partners'; the location signals are 'sourced from vetted partners with explicit, auditable consent' with no route named. | 1, 6 | vendor assertion | supported as stated | medium | [420, 455] |
| LOC-F-F11 | Simpli.fi's Addressable Geo-Fencing (live, US, 2026) converts up to one million street addresses per campaign into parcel-level geofences drawn from plat lines in property-tax and public land-survey records, targets 'any device seen within those plat lines' with mobile, video and CTV ads for up to 30 days via a cross-device graph with 'variable recency settings', and measures with Conversion Zones and 'Geo Conversion Lift'; it says it uses 'GPS data paired with plat lines' and rejects IP because of truncation, but does not say where the GPS fixes come from. | 1, 6 | vendor assertion | supported as stated | medium | [456, 457] |
| LOC-F-F12 | Vistar Media's own guide documents DOOH audience targeting from 'aggregated and anonymized location data' plus third-party partners (Epsilon, Eyeota, InMarket, Affinity Answers). T-Mobile's app-based segments have been available through Vistar since 6 May 2025; on 5 October 2026 Vistar announced enhanced, location-enriched T-Mobile segments available only through the Vistar DSP. Per AdExchanger, T-Mobile's data combines mobile location data with app installation and usage inferred from network requests, with OOH-proximity exposure only for users who opted into location tracking; T-Mobile's Jess Zhu attributes the exclusivity to tightening regulation of location data and mobile device IDs. | 1, 6 | synthesis | supported with qualification | medium | [338, 335] |
| LOC-F-F20 | H4 remains unresolved, and the cohort B source counts must be rebuilt. Foursquare (SDK/API partners plus data suppliers, brokers and ad networks) and InMarket (own app, partner SDK network and 'trusted &amp; certified third-party sources') disclose multiple sources without naming a primary one, so code both 'multiple disclosed sources; primary source not established'. Azira (SDK, bid-stream and transactions) and GroundTruth disclose blends without proportions. No product documents a primary source, so neither the earlier 0/0 count nor the verifier's two SDK votes stands. | 1, 6 | synthesis | contradicted | medium | [265, 420, 421, 422, 458, 457] |
| LOC-F-F21 | No cohort B documentation opened states a fix-accuracy threshold with a method or a detection error statistic. The only numeric precision figure found is Azira Audiences' unexplained &quot;within 3 meters of precision&quot;. Elsewhere the precision class is asserted by geometry: GroundTruth On-Premise &quot;strictly inside the physical building walls&quot;, Simpli.fi parcel plat lines, Adsquare &quot;precise polygons or custom radii&quot; alongside postcode and H3-cell units, Foursquare &quot;snap-to-place&quot; visits, and coarse ZIP, state or DMA options at GroundTruth. GroundTruth's &quot;Blueprint-level precision&quot; is a measurement-product claim, not a targeting precision class. | 1, 6 | synthesis | supported with qualification | medium | [263, 122, 420, 457, 459] |
| LOC-GROUNDTRUTH-F10 | GroundTruth documents the tests it applies to each location signal and visit (GPS-sourced, stale, outlier, fraudulent device, inside a Blueprint, during open-store hours, typical employee behaviour, drive-by speed; centroids and carrier-IP locations rejected) but publishes no numeric threshold for dwell, horizontal accuracy or speed. | 1, 6 | vendor assertion | supported as stated | high | [357, 259] |
| LOC-INMARKET-F06 | InMarket documents its current location sources as its own first-party app, a ‘Comscore-verified partner SDK app network’ and unnamed ‘trusted &amp; certified third-party sources’, with precise coordinates collected via apps operated by InMarket and Third Party Partners; it gives no proportions by source, and its own apps at 30 June 2026 are CheckPoints, ListEase, Key Ring, Subscription Stopper and Out of Milk. | 1, 6 | vendor assertion | supported as stated | high | [267, 134] |
| LOC-INMARKET-F15 | GeoLink, InMarket’s self-serve activation platform from the Thinknear acquisition, combines first-party location data with a machine-learning ‘Location Score’ that evaluates the location carried in ad requests before bidding, and sells block-level geo-audiences, real-time point-of-interest geofences and retargeting on past store visits; so InMarket’s targeting uses bidstream-derived location as a filtered input alongside SDK data. | 1, 6 | vendor assertion | supported as stated | medium | [276, 40] |
| LOC-F-F10 | The FTC's Mobilewalla Decision and Order (Docket C-4811) was issued 13 January 2025 and its finalisation announced 14 January 2025 on a 4-1 vote; it bars Mobilewalla from collecting or retaining information it accesses in online advertising auctions for any purpose other than participating in them, and from selling, sharing, disclosing or otherwise using Sensitive Location Data for the sensitive locations it must identify, except to convert that data into non-sensitive or non-location data. It is the only regulator record in this stream that documents bidstream as a cohort B location source; Azira's own data page also names bid-stream among its sources. | 1, 6, 7 | direct record | supported with qualification | high | [140] |
| LOC-H-F24 | Across the US rules read, statutes define precise geolocation by a radius of 1,750 feet (Maryland, Oregon's sale ban, Washington's precise location information, Connecticut, Virginia) or 1,850 feet (California, Civil Code 1798.140(w)); the five FTC orders instead exclude coarse data, giving as an example zip code or census block location with a radius of at least 1,850 feet; Washington's geofence definition uses 2,000 feet. | 1, 6, 7 | direct record | supported with qualification | high | [152, 154, 151, 155, 162, 161, 143] |
| LOC-A-F16 | OpenRTB 2.6's Geo object carries lat and lon as floats, a 'type' for the source of location (AdCOM Location Types: 1 GPS/Location Services, 2 IP Address, 3 User Provided), 'accuracy' in metres as reported by the device (recommended for type 1), 'lastfix' in seconds since the fix, and 'ipservice' (AdCOM IP Location Services, beginning ip2location, Neustar (Quova), MaxMind), and it states that lat/lon should be passed only if they conform to the accuracy of the stated type and that a postal-code centroid should not be passed; the 2.6 text contains no rule on truncating or rounding coordinates. | 1, 6, 8 | direct record | supported as stated | high | [219, 460] |
| LOC-H-F02 | The X-Mode/Outlogic order (C-4802) treats data revealing only coarse location (its example: zip code or census block with a radius of at least 1,850 feet) as not Location Data, lists seven categories of Sensitive Locations, requires deletion of Historic Location Data within 60 days unless consent is obtained or the data is deidentified or rendered non-sensitive within 90 days, deletion of Data Products within 90 days, a Supplier Assessment Program and a consumer deletion-request mechanism, bars collecting Location Data from devices that opted out of tracking without a consent record, and, from 90 days after the effective date, bars collecting, using or disclosing any Location Data without a record of the consumer's prior consent. | 1, 7 | direct record | supported with qualification | high | [126] |
| LOC-H-F12 | Washington's My Health My Data Act (Chapter 191, Laws of 2023, ESHB 1155) made it unlawful from 23 July 2023 to implement a geofence (a virtual boundary of 2,000 feet or less from the perimeter) around an entity providing in-person health care services to identify or track consumers, collect consumer health data or send them messages; its consumer-health-data duties apply from 31 March 2024 (30 June 2024 for small businesses), define precise location information by a 1,750-foot radius and treat precise location that could indicate an attempt to obtain health services as consumer health data. | 1, 7 | direct record | supported as stated | high | [151] |
| LOC-H-F14 | California defines 'precise geolocation' in Civil Code 1798.140(w) as data used to locate a consumer within an area equal to or less than a circle with a radius of 1,850 feet and lists it as sensitive personal information in 1798.140(ae); its Delete Act registry requires data brokers to register each January 1-31 and to begin processing deletion requests through the DROP mechanism on 1 August 2026. | 1, 7 | direct record | supported as stated | medium | [161, 163] |
| LOC-H-F10 | Maryland's Online Data Privacy Act (2024 ch. 455, SB 541, approved 9 May 2024, effective 1 October 2025) bars a controller from selling sensitive data, which includes precise geolocation defined by a 1,750-foot radius, with no consent exception, subject to the act's coverage thresholds and exemptions; it also bars any person from using a geofence within 1,750 feet of a mental health or reproductive or sexual health facility to identify, track, collect data from or notify consumers about their consumer health data. Section 2 of the chapter makes section 14-4612 as enacted apply only prospectively, not to processing before 1 April 2026; that clause does not postpone the act or the sale ban. | 1, 7, 12 | direct record | supported with qualification | high | [152] |
| LOC-H-F11 | Oregon's HB 2008 (Oregon Laws 2025, chapter 251; passed the House 17 April 2025, the Senate 21 May 2025, repassed 27 May 2025, approved by the Governor 3 June 2025, effective 1 January 2026) amends ORS 646A.578 so a controller may not sell personal data that accurately identifies within a radius of 1,750 feet a consumer's or linked device's present or past location, or personal data of a consumer known or wilfully disregarded to be under 16. | 1, 7, 12 | direct record | supported as stated | high | [154, 153] |
| LOC-H-F13 | Connecticut bans the sale of precise geolocation data: Public Act 26-64 (SB 4, approved 27 May 2026, effective 1 October 2026) provides that no controller and no third party shall sell any consumer's precise geolocation data (1,750-foot radius; content of communications and utility metering data excepted), replacing the consent-to-sell rule PA 25-113 had put in place from 1 July 2026. Data brokers selling or licensing brokered personal data in Connecticut must be registered from 1 January 2027. | 1, 7, 12 | direct record | contradicted | high | [155, 159] |
| LOC-A-F03 | The guidelines separate accuracy (closeness to true location) from precision (number of decimal places) and require both as disclosures; define geo-fencing as either point-radius or boundary (polygon); define physical visitation as the intersection of device location and place with speed or longitudinal analysis to exclude drive-bys; and state that geo-fencing generally cannot represent multi-storey structures. | 1, 8 | direct record | supported as stated | high | [445] |
| LOC-A-F04 | The 2017 guidelines list GPS, Wi-Fi, IP, cell trilateration, registration, beacon, panel and ad/bid-request location as sources and characterise their relative accuracy (beacons within metres; GPS 10-100 m; Wi-Fi 30-100 m; cell at zip or neighbourhood level; mobile IP generally not accurate or precise); they say bid-request location should pass validation edit rules, that zip, DMA or state centroids should not represent precise location unless reported at market level, and that vendors using ad or bid-request data should disclose the number of active devices observed daily, weekly or monthly and the number of measured visits. | 1, 8 | direct record | supported with qualification | high | [445] |
| LOC-A-F14 | Digital Envoy's Digital Element NetAcuity IP-geolocation service, the only other MRC-accredited location-determination service found in the communications screened, was formally moved from hiatus to non-accredited in January 2020 after Digital Envoy chose not to proceed with a new audit; no IP-geolocation service is on MRC's lists at 24 October 2026. | 1, 8 | direct record | supported with qualification | high | [165, 171] |
| LOC-A-F19 | Across the documents opened, two decimal places is the recurring boundary between precise and coarse coordinates, but the bodies attach different distances to it: NAI treats two or fewer decimals, or any area at least a 500 m-radius circle, as imprecise; MRC's 2017 guidelines and IAB's 2016 guide say two decimals is about one kilometre; Google caps at 0.01 degrees; Prebid truncates to two decimals when the activity is denied. NAI's 500 m radius is close to, but about 6% smaller than, the 1,750-foot (533 m) radius used in several state laws (per workstream H). | 1, 8 | synthesis | supported with qualification | medium | [461, 445, 462, 444, 437] |
| LOC-A-F21 | NAI's 2020 Code defines Precise Location Information as data derived through technology able to determine with reasonable specificity a device's actual physical location (for example GPS-level coordinates or RF triangulation) and requires opt-in consent for its use in tailored advertising or ad delivery and reporting; NAI's February 2020 guidance (superseding 2015) describes three methods of rendering location imprecise: two or fewer decimal places, an area at least as large as a 500 m-radius circle, or a non-unique general descriptor of the place. | 1, 8 | direct record | supported with qualification | high | [461] |
| LOC-C-F14 | Vistar Media's two US DOOH cases measured by Foursquare report, for Target (flight 2021-08-15 to 2022-01-29; 22 states; $1.2M+ budget), 72 million+ impressions, 533,000 store visits inside a four-day look-back and a 6.51% lift in visitation 'compared with non-exposed consumers', and for Jack in the Box (twelve markets; undated), 49M+ impressions, 1.3M+ store visits and an 8.8% lift; exposure is inferred from ad-log timestamps, device location, venue type and screen location, and the OAAA document says Vistar's studies 'compare control versus exposed groups' without describing the control. | 1, 9 | vendor assertion | supported as stated | medium | [8, 9, 463] |
| LOC-E-F21 | Adsquare infers DOOH exposure by proximity of mobile advertising IDs to screen coordinates while an ad is live, then cross-references exposed IDs against later in-store visits in its 2D polygons; no exposure radius, lookback window or threshold is stated in the interview. | 1, 9 | vendor assertion | supported as stated | high | [464] |
| LOC-J-F01 | At AdCP v3.2.6 (latest stable release, 23 October 2026) the media-buy targeting overlay represents administrative geography (geo_countries, geo_regions as ISO 3166-2, geo_metros with systems nielsen_dma, uk_itl1, uk_itl2, eurostat_nuts2 or custom, geo_postal_areas) with matching exclusion fields, point-based geo_proximity (radius, travel-time isochrone or buyer-supplied GeoJSON polygon), store_catchments from a synced store catalog, and identifier-based geo_places (first published in the pre-release 3.2.0-beta.0 on 17 August 2026, first stable in 3.2.1 on 1 October 2026); this is schema representation and nothing more. | 1, 10 | direct record | supported with qualification | high | [197, 221] |
| LOC-D-F01 | T-Mobile US agreed on 20 December 2024 to acquire 100% of Vistar Media Inc. and completed the acquisition on 3 February 2025, transferring cash of US$621 million, per its FY2025 Form 10-K. Note 2 of that filing puts the fair value of consideration transferred at US$617 million, because part of the payment settled preexisting relationships with Vistar. | 2 | direct record | supported with qualification | high | [36] |
| LOC-D-F03 | Lamar Advertising received US$115.1 million in cash at the 3 February 2025 closing for its 20% equity interest in Vistar Media, with up to a further US$15.1 million held in escrow, per its 8-K earnings exhibit of 20 February 2025. | 2 | direct record | supported as stated | high | [37] |
| LOC-D-F04 | Near Intelligence, Inc. completed its business combination with the SPAC KludeIn I Acquisition Corp on 23 March 2023 (EDGAR records the former name through 2023-03-23) and began trading on the Nasdaq Global Market under the ticker NIR on 24 March 2023. | 2 | direct record | supported with qualification | high | [465, 466] |
| LOC-D-F05 | Near Intelligence reported revenue of US$33,217,126 for the six months ended 30 June 2023 (US$28,897,952 a year earlier, a 14.9% increase) and a net loss of US$36,899,928, and its 10-Q stated that its liquidity condition raised substantial doubt about its ability to continue as a going concern. | 2 | direct record | supported as stated | high | [465] |
| LOC-D-F06 | Near Intelligence, Inc. and certain of its subsidiaries filed Chapter 11 petitions on 8 December 2023 in the US Bankruptcy Court for the District of Delaware, with up to US$16,000,000 of DIP financing from lenders agented by Blue Torch Finance LLC and a stalking-horse credit bid of not less than US$50,000,000 from BTC Near Holdco LLC; Nasdaq notified the company that trading would be suspended at the opening of business on 19 December 2023; the event is an insolvency. | 2 | direct record | supported with qualification | high | [54] |
| LOC-D-F07 | ... a Form 15-12G deregistration, signed 27 March 2024, was filed with the SEC on 28 March 2024 (EDGAR filing date; accepted the evening of 27 March). | 2 | direct record | supported with qualification | high | [55, 467, 466] |
| LOC-D-F08 | Azira operates as Azira LLC with headquarters in Pasadena and offices in Paris, Bangalore and Sydney, and claims a data infrastructure of 13 trillion location signals, 43 million points of interest and solutions in 44 countries; this review infers, but did not open a settling source showing, that Azira is the business sold out of Near Intelligence's Chapter 11 to Blue Torch's BTC Near Holdco LLC. | 2 | inference | supported as stated | low | [458, 468, 427] |
| LOC-D-F09 | At listing in March 2023 Near described itself as processing data from an estimated 1.6 billion unique user IDs and 70 million points of interest in more than 44 countries; the definition is 'unique user IDs', not devices or a monthly panel, and no geography split was given. | 2 | vendor assertion | supported as stated | medium | [427] |
| LOC-D-F10 | Foursquare announced its acquisition of Placed from Snap on 30 May 2019 with terms undisclosed; Snap's 10-Q records that it divested its membership interest in Placed, LLC to Foursquare Labs, Inc. in June 2019 for total cash of US$77.8 million, of which US$66.9 million was purchase consideration, recognising a US$39.9 million gain; the event is an acquisition (completed June 2019). | 2 | direct record | supported as stated | high | [30, 29] |
| LOC-D-F12 | Foursquare and Factual announced a merger on 6 April 2020 (terms undisclosed, Foursquare name retained, David Shim remaining CEO), claiming combined datasets of more than 500 million devices worldwide, a panel of 25 million opted-in users, more than 105 million points of interest and more than US$150 million of revenue; the event is a merger, announced; no opened source gives the completion date. | 2 | vendor assertion | supported as stated | medium | [39, 38] |
| LOC-D-F13 | Foursquare retired its City Guide consumer app on 15 December 2024 at 11:59 PM ET and the web version on 28 April 2025, stating it did so to focus on Swarm; this is a product retirement. The page does not say whether City Guide contributed location data to Foursquare's panel. | 2 | vendor assertion | supported with qualification | high | [64] |
| LOC-D-F14 | Infillion announced on 18 September 2026 (ACCESS Newswire) that it is acquiring Foursquare, terms undisclosed; contemporaneous reports describe the transaction as pending close, and the release claims Foursquare covers more than 100 million points of interest, 16 billion human-verified check-ins and 'aggregated coverage of 250 million U.S. devices' (AdExchanger paraphrases this as data 'drawn from 250 million US devices'). | 2 | vendor assertion | supported with qualification | medium | [50, 51] |
| LOC-D-F16 | xAd rebranded as GroundTruth on 15 June 2017 (a rebrand, not an exit), claiming at the time 95 million active monthly users and 17 million active daily users across 100 million places, with no geography attached to the user figures. | 2 | vendor assertion | supported as stated | high | [56] |
| LOC-D-F17 | Telenav completed the sale of its advertising business (Thinknear) to inMarket Media, LLC on 16 August 2019 in exchange for inMarket LLC units representing a 14.5% member interest valued at US$15.6 million, recording a US$4.9 million loss on sale and presenting the business as discontinued operations; Thinknear's status is therefore 'sold to inMarket', an acquisition by asset sale. | 2 | direct record | supported as stated | high | [31, 32] |
| LOC-D-F18 | InMarket acquired the assets of NinthDecimal in a transaction reported on 10 September 2020 for an undisclosed sum, stating that it then had 200 employees, US$100 million of annual revenue and 'over 100M SDK-based users', and continued NinthDecimal's LCI, which InMarket's page expands as 'Location Conversion Index'; the event is an acquisition (asset purchase, 2020, not 2019 as the register seeds). | 2 | vendor assertion | supported as stated | medium | [41, 40] |
| LOC-D-F19 | The FTC's January 2024 complaint describes InMarket Media, LLC as a Delaware LLC based in Austin, Texas, whose SDK was incorporated into more than 300 third-party apps downloaded onto over 390 million unique devices since 2017 (its own apps onto over 30 million), and alleges that from 2016 'about 100 million unique devices sent Respondent location data each year'. | 2 | direct record | supported as stated | high | [275] |
| LOC-D-F21 | Digital Envoy acquired X-Mode Social in a transaction reported on 5 August 2021 with terms undisclosed, rebranding it Outlogic and naming X-Mode's CEO Josh Anton as Digital Envoy's chief strategy officer; the acquirer said it shut off US location data flows to defense contractors; the event is an acquisition followed by a rebrand. | 2 | vendor assertion | supported as stated | medium | [42] |
| LOC-D-F22 | The FTC's case record names Outlogic, LLC as the successor to X-Mode Social, Inc.; the proposed order is dated 9 January 2024 and the final order 11 April 2024 (terms are workstream H's). | 2 | direct record | supported as stated | high | [469] |
| LOC-D-F25 | Spectus, Cuebiq's former Workbench clean room, was publicised as a stand-alone data clean room in a release dated 23 February 2022, and an undated Cuebiq page describes it as a separate business unit to which former Workbench clients are being migrated, while Cuebiq 'will remain as an ad tech business solely focused on Media Measurement and Audience Targeting'; the event is a restructuring (split), not an insolvency or exit. | 2 | vendor assertion | supported with qualification | medium | [57, 58] |
| LOC-D-F26 | Unacast announced on 29 November 2023 that it and Gravy Analytics were merging; no opened source states the structure, surviving entity or completion date, so the event is recorded as a merger, announced. | 2 | vendor assertion | supported as stated | medium | [44] |
| LOC-D-F28 | Precisely, backed by Clearlake and TA Associates, signed a definitive agreement to acquire New York-based PlaceIQ on 19 January 2022 (terms undisclosed) and states that the acquisition completed in February 2022; the event is an acquisition (completed). | 2 | vendor assertion | supported as stated | high | [33, 34] |
| LOC-D-F29 | Advan Research announced on 3 November 2022 that it had acquired SafeGraph's Patterns (Weekly, Monthly and Neighborhood Patterns) business, and SafeGraph's documentation marks Patterns a legacy product unavailable 'at the start of 2023'; for SafeGraph this is a product retirement and for Advan an acquisition. | 2 | vendor assertion | supported as stated | high | [43, 63] |
| LOC-D-F30 | SafeGraph's documentation disclosed, as of its July 2020 release, a US opted-in device panel broken down by carrier whose six listed counts sum to 28,914,920 devices (Verizon 10,303,871; AT&amp;T 7,267,146; T-Mobile 7,129,894; Sprint 3,685,988; Altice 323,221; C-Spire 204,800); the definition is 'a panel of opted-in mobile devices' at one release, not a monthly series. | 2 | vendor assertion | supported as stated | medium | [63] |
| LOC-D-F39 | Foursquare documented two data-marketplace routes for its datasets: AWS Data Exchange from 13 November 2019 (Places plus 'foot traffic-powered datasets', delivered to S3 and billed to the buyer's AWS account) and Snowflake Data Marketplace from 8 February 2021 (Places and Visits via secure data sharing); a live AWS Marketplace listing also carries a SafeGraph Places free sample delivered through AWS Data Exchange. | 2 | vendor assertion | supported as stated | medium | [470, 471, 472] |
| LOC-D-F40 | Foursquare's public device figures move from 'more than 100 million monthly devices' (US, May 2019) and 'the movements of 100 million U.S. devices' (November 2019) to 'more than 500 million devices worldwide' plus 'a panel of 25 million opted-in users' (April 2020, post-Factual) to 'aggregated coverage of 250 million U.S. devices' (September 2026); the four figures use four definitions and do not form a series on a constant definition. | 2 | synthesis | supported as stated | medium | [30, 470, 38, 50] |
| LOC-D-F43 | Infillion, the pending acquirer of Foursquare, already owns the location SDK company Gimbal and acquired Catalina (closed February 2026) and MediaMath (bought out of bankruptcy in 2023 for US$22 million), so Gimbal (cohort D) and Foursquare (cohorts A to D) would sit under one owner if the Foursquare deal closes. | 2 | synthesis | supported as stated | medium | [51, 50] |
| LOC-U-F03 | If the Foursquare deal closes, Infillion will own two location product families that the register treats separately: the Gimbal Location Platform (an SDK with radial or polygon geofences 'down to 50 meters', iBeacon micro-location, visit-based measurement and visit-based audience segmentation for app owners' first-party data; gimbal.com now redirects to infillion.com) and Foursquare's attribution, audience, places and SDK products, alongside MediaMath, TrueX, Drawbridge and Catalina. | 2 | synthesis | supported as stated | medium | [473, 52] |
| LOC-U-F07 | Bluedot, the seed developer-geofencing vendor, is no longer independent: Rezolve AI plc agreed on 4 February 2025 to buy the share capital of Bluedot Industries, Inc. (Delaware) and Bluedot Industries Pty. Ltd. and closed on 20 February 2025 for 819,737 ordinary shares, a related-party deal (the seller DBLP is wholly owned by a Rezolve director) accounted for as a common-control transfer; bluedot.io now redirects to rezolve.com. | 2 | direct record | supported as stated | high | [474, 475] |
| LOC-U-F08 | Skyhook, an early Wi-Fi positioning supplier relevant to the developer cohort, was acquired by Qualcomm Technologies in a transaction Qualcomm confirmed by email in May 2022 without a formal announcement, and skyhook.com redirects to Qualcomm's Terrestrial Positioning Service at the cutoff; it is classified as historical (acquisition), outside the advertising cohorts. | 2 | synthesis | supported as stated | medium | [476] |
| LOC-U-F10 | Outlogic (formerly X-Mode, under an FTC order banning sale of sensitive location data) still licenses an 'Observation Panel', attribution data and audiences at the cutoff through Snowflake, daily and hourly AWS S3 exports, a REST API, Eyeota and Adsquare, and its homepage notice under Texas law says it 'may receive and sell the precise latitude and longitude coordinates of your mobile device'. | 2 | vendor assertion | supported as stated | high | [128] |
| LOC-U-F13 | Advan Research sells five named products at the cutoff (FIT, REI, REVEAL, SPENDVIEW and Patterns+) and claims '100+ daily observations from 45M U.S. &amp; 65M global devices', geolocation 'from GPS, Wi-Fi, and beacon signals' with 'No ad exchanges', and '10M+ hand drawn geofences'; this is one of the few dated-at-access US device counts with a stated source type. | 2 | vendor assertion | supported as stated | medium | [477] |
| LOC-U-F14 | At least three location-data sellers screened (Unacast, Veraset and Complementics) sell identity linkage of mobile advertising IDs to hashed emails and/or IPs alongside location data; a fourth, Quadrant, sells hashed-email datasets for identity resolution, with no stated link to its location records. | 2 | synthesis | supported with qualification | medium | [125, 478, 426, 425] |
| LOC-U-F16 | Cuebiq Group, LLC (1460 Broadway, New York) markets Footfall Measurement, Cuebiq Audiences, Footfall Benchmarking and a Data Platform at the cutoff, and spectus.ai redirected to cuebiq.com on 24 October 2026; whether Spectus (split out in February 2022 per workstream D) has been folded back into Cuebiq is an inference from the redirect, not an established corporate event. | 2 | inference | supported with qualification | low | [479, 480] |
| LOC-U-F19 | Lifesight has pivoted out of the location cohorts: its 2026 homepage sells 'causal MMM, incrementality testing and incrementality-adjusted attribution' with AI agents and names no footfall, store-visit or location-data product, so its seed row moves to listed. | 2 | inference | supported as stated | medium | [481] |
| LOC-U-F24 | Three names could not be confirmed active from an opened source by the cutoff: onemata.com returned a Wix &quot;ConnectYourDomain Error&quot; (404), datastreamgroup.com served only a parking redirect, and plotprojects.com returned a Cloudflare 403 to two routes; their status is unconfirmed, not closed. Precisely's PlaceIQ Audiences URL, refused to curl, is readable through a second fetch route and shows the product live as &quot;Precisely Audiences&quot;. | 2 | inference | supported with qualification | medium | [482, 483, 484, 421] |
| LOC-U-F25 | Data-marketplace discovery on Datarade's foot-traffic category (headed '40 Foot Traffic Data Providers' on 24 October 2026) adds location sellers not in the seed list, among them Irys, Intuizi ('400mm+ Devices and growing'), DRAKO, GapMaps and BIGDBM, alongside seed vendors Quadrant and Unacast; the providers' figures are self-reported. | 2 | vendor assertion | supported as stated | medium | [485] |
| LOC-BLIS-F01 | T-Mobile US entered a share purchase agreement for 100% of Blis Holdco Limited on 18 February 2025 and completed the acquisition on 3 March 2025, transferring cash of US$180 million, of which US$174 million was the fair value of consideration transferred and the remainder settled preexisting relationships with Blis; provisionally assigned fair values were total assets acquired of US$264 million (including goodwill) and liabilities assumed of US$90 million, with provisional goodwill of US$105 million assigned to the Wireless segment; Blis's results to 31 December 2025 were 'not material'. | 2 | direct record | supported with qualification | high | [36] |
| LOC-BLIS-F02 | Companies House records T-Mobile US, Inc. as the person with significant control of Blis Holdco Limited (75% or more of shares and votes) from 3 March 2025; LDC GP LLP held more than 50% but less than 75% of shares from 16 June 2022 to 3 March 2025; the holding company was incorporated on 14 March 2022 as Project Balloon Bidco Limited and renamed on 8 January 2024; Blis Global Ltd (incorporated 18 December 2007, Blis Media Limited 2012-2019) is controlled by Blis Group Limited. | 2 | direct record | supported as stated | high | [225, 486] |
| LOC-BLIS-F19 | Blis's About page, last modified 8 October 2026, is titled 'Independent DSP Established in 2004' and does not mention T-Mobile, while its homepage title reads 'Blis, part of T-Ads \| Omnichannel DSP'. | 2 | vendor assertion | supported as stated | high | [487, 228] |
| LOC-FOURSQUARE-F01 | Infillion announced on 18 September 2026 that it was acquiring Foursquare (title 'to Acquire', body 'has announced the acquisition'), with undisclosed terms; its release says leadership will join 'after the transaction closes'. Foursquare's homepage read 'Foursquare has joined Infillion' on 24 October 2026, and Business Insider wrote on 20 October 2026, in its own words, that 'Foursquare was acquired by Infillion', reporting Crowley's Substack post that equity holders received $0. No opened source states a closing date. | 2 | synthesis | supported with qualification | medium | [52, 51, 347, 488] |
| LOC-FOURSQUARE-F19 | Foursquare's public device figures use five different definitions and do not form a series: 'more than 100 million monthly devices' measured in the US (May 2019), 'more than 500 million devices worldwide' plus a 25 million opted-in panel (April 2020), 'an audience of 100M+ unique users' (Visits docs, undated), '9B+ monthly visits from 500M+ devices worldwide' (homepage, October 2026) and 'aggregated coverage of 250 million U.S. devices' (Infillion, September 2026). | 2 | synthesis | supported as stated | medium | [30, 38, 241, 347, 52] |
| LOC-GROUNDTRUTH-F01 | ZeroToOne.AI announced on 25 March 2026 that it had acquired GroundTruth, with no seller, price or closing record disclosed; GroundTruth's site now calls itself 'a ZeroToOne.AI company', and the acquirer's counsel confirmed its role on 24 April 2026. | 2 | direct record | supported as stated | high | [47, 489, 49] |
| LOC-GROUNDTRUTH-F25 | No current US panel size is published for GroundTruth; the dated scale figures found use incompatible definitions: a 50 million opted-in 'mobile panel' target (November 2016), a 'first-party database' of 95 million 'active monthly users' and 17 million 'active daily users' plus a claim to see 'two out of three people who own a smartphone in the U.S.' and 'over one billion physical visits ... each month' globally (June 2017), 'over 30 Billion visits annually' (2021) and 'more than 2,000 enterprise customers globally' (March 2026). | 2 | vendor assertion | supported with qualification | medium | [262, 56, 356, 47] |
| LOC-INMARKET-F08 | InMarket’s scale figures are incompatible and mostly undefined: ‘200 million+ Location-based consumers’ (current pages, no period), ‘200M+ opted-in consumers’ (Maestrø release, June 2026), ‘nearly 200M+ permission-based users’ (Vistar page), 110 million monthly active unique devices (undated NinthDecimal sell sheet), ‘over 100M SDK-based users’ (2020 letter) and GeoLink’s 200M+ mobile, 190M+ desktop and 110M+ CTV devices; none is a dated US monthly device panel. | 2 | vendor assertion | supported as stated | high | [268, 273, 23, 274, 40] |
| LOC-PLACER-F02 | Placer's undated help-centre page describes its panel as &quot;tens of millions&quot; of US devices, about 7% of the US population, 55% iOS and 45% Android, from hundreds of partner apps that collect location &quot;in the background&quot;, with data from 1 January 2017 and US territories excluded. | 2 | vendor assertion | supported as stated | medium | [3] |
| LOC-PLACER-F03 | Placer's stated panel size appears as &quot;over 20 million active devices&quot; (website as quoted May 2022), &quot;30M Active Devices&quot; and &quot;25M+ devices representing ~9%&quot; (Sept 2024 deck), &quot;over 20 million monthly active users&quot; (Santa Cruz fact sheet, Feb 2026) and &quot;roughly 30 million mobile devices (representing about 8% of the U.S. population)&quot; (Sept 2026 commissioned write-up), against &quot;approximately 7%&quot; on the help page; only the city sheet gives a monthly basis. Drop &quot;internally inconsistent&quot; for the app count: the deck counts dozens of SDK partners and hundreds (500) of apps, different units. | 2 | synthesis | supported with qualification | medium | [385, 289, 290, 291] |
| LOC-PLACER-F12 | Placer documents its supply as background location from partner mobile apps that integrate its SDK, plus auxiliary Bluetooth, Wi-Fi beacon, battery and network data; TechCrunch reported in August 2024 that it also gets data &quot;through other third-party sources&quot;; no source opened gives proportions by source, and none documents use of bidstream location. | 2 | synthesis | supported as stated | medium | [289, 379, 380, 385] |
| LOC-UNACAST-F02 | At the cutoff Unacast's own Foot Traffic Data page sells four Location Intelligence APIs (Area Visitors returning &quot;unique device counts&quot; for a submitted polygon, coordinate-level Observations, Trade Areas, Personas) and scheduled DaaS delivery, while its pricing page still lists machine-learning Foot Traffic Datasets and the Location Insights platform as &quot;US only&quot; packages. | 2 | vendor assertion | supported as stated | medium | [331, 332] |
| LOC-UNACAST-F03 | Gravy Analytics, Inc., a Delaware corporation in Ashburn, Virginia whose wholly owned subsidiary is Venntel, Inc., trades as Unacast: its September 2026 privacy policy opens &quot;Gravy Analytics, Inc. dba 'Unacast&quot;, its January 2025 breach notice calls it &quot;a subsidiary of Unacast, Inc&quot;, and gravyanalytics.com redirects to unacast.com; the combination was announced as a merger on 29 November 2023 with no structure or completion date stated. | 2 | synthesis | supported as stated | high | [137, 333, 135, 44, 490] |
| LOC-UNACAST-F06 | The FTC complaint against Gravy and Venntel alleges that they collected no data directly from consumers, obtained it from suppliers who &quot;may themselves obtain the location data from other data suppliers, the mobile advertising marketplace, or mobile applications&quot;, claimed over 17 billion signals from about a billion devices daily, sold raw timestamped coordinates tied to MAIDs and IP addresses in batch deliveries, and failed to take reasonable steps to confirm consent. | 2 | direct record | supported as stated | medium | [135] |
| LOC-UNACAST-F09 | Unacast documents its supply as licensed, not collected: it says it chose &quot;to exclusively license data from other entities in the ecosystem instead of collecting location data through our own SDK&quot;, licenses signals &quot;from aggregators representing SDKs across tens of thousands of mobile apps&quot;, and processes &quot;an almost equal mix of U.S. and international data&quot;, with no source proportions. | 2 | vendor assertion | supported as stated | medium | [329, 331, 137] |
| LOC-UNACAST-F19 | Unacast and Gravy volume figures use incompatible global units and none gives a US monthly device count: &quot;more than 1 billion monthly devices across 180+ countries&quot; (product page), &quot;70 billion global signals per day&quot; and &quot;600 million MAUs globally&quot; (Datarade, self-reported), and the companies' claim of &quot;more than 17 billion signals from around a billion mobile devices daily&quot; (FTC release, December 2024). | 2 | vendor assertion | supported as stated | high | [328, 330, 428] |
| LOC-D-F23 | In December 2020 Apple and Google barred apps in their app stores from using X-Mode's SDK, a platform SDK removal; this review found the event only in trade reporting (AdExchanger, February 2021) and did not open a statement by either platform. | 2, 3 | synthesis | supported as stated | medium | [60] |
| LOC-D-F24 | Cuebiq announced on 23 February 2021 that it was sunsetting its proprietary location SDK in favour of a clean-room product, Workbench (then in private beta for six months), while keeping an open-source SDK version that developers could still use to collect consented data; this is a documented partial shift of a US location vendor away from its own SDK supply, with no panel figure disclosed. The CEO cited privacy and security risk; the article, not Cuebiq, sets the move against Apple's cross-app tracking crackdown and ATT. | 2, 3 | vendor assertion | supported with qualification | medium | [60] |
| LOC-FOURSQUARE-F18 | Foursquare projects 'normalized visits' in the Research feed by weighting each observed visit with a 'SAG Score' (state, age and gender cohort population divided by active panelists), adjusts weights to offset changes in panel size and technical changes 'such as changes to how we calculate a visit in our SDK due to an OS update', and names iOS 13, partner-app MAU changes and new contributing apps as causes of step changes in raw visits. | 2, 3 | vendor assertion | supported as stated | high | [240] |
| LOC-PLACER-F08 | Placer restates history when it changes its model: Data Version 2.0 (31 March 2024) updated &quot;all historical data from January 1, 2017&quot;, and its accuracy page says de-biasing adjustments happen &quot;usually no more than once a year&quot; and &quot;can sometimes impact previously run numbers&quot;, so a Placer series read at two dates can differ for the same past period. | 2, 4 | vendor assertion | supported as stated | high | [285, 102] |
| LOC-BLIS-F11 | Since 10 September 2025 the Blis DSP carries T-Mobile Advertising Solutions' US app ownership and app engagement data, which Blis says it combines with 'real-world movement patterns and verified transactions'; the homepage calls Blis 'part of T-Ads' and 'the only omnichannel DSP to combine T-Mobile app data with real-world movement &amp; spending habits'. The T-Mobile data is described as app data, not location data. | 2, 6 | vendor assertion | supported as stated | high | [227, 228] |
| LOC-BLIS-F12 | No current US panel size was found for Blis; the most recent panel figure is the August 2021 press kit's '370m opted-in global planning and measurement panel' (global, undefined), alongside '18 billion' ad requests a day and a 'Smart Pin' filter that discards about 80% of raw location data; a 2019 release gave reach of over a billion mobile devices a year. | 2, 6 | vendor assertion | supported as stated | medium | [230, 343] |
| LOC-D-F34 | The FTC finalized its order against Mobilewalla on 14 January 2025; the order bans Mobilewalla from collecting consumer data from real-time bidding exchanges for purposes other than participating in those auctions, and the FTC alleged (this was the first time it alleged such a practice was unfair) that Mobilewalla had collected bidstream data; this documents an FTC allegation that bidstream collection was among Mobilewalla's sources, not an admission or a finding. | 2, 6, 7 | direct record | supported with qualification | high | [491] |
| LOC-VISTAR-F01 | T-Mobile US agreed on 20 December 2024 to acquire 100% of Vistar Media Inc. and completed the acquisition on 3 February 2025, transferring cash of US$621 million, of which a portion settled preexisting relationships; the fair value of consideration transferred was US$617 million (FY2025 Form 10-K). | 2, 6, 9 | direct record | supported as stated | high | [36] |
| LOC-VISTAR-F06 | AdExchanger reported on 5 October 2026 that T-Mobile's data for Vistar includes mobile location data used to estimate inventory impressions and app installation and usage read from network requests; for users who opted into location tracking, T-Mobile can also detect proximity to OOH ads and use that exposure to retarget the user on other channels. T-Mobile tied the Vistar-only exclusivity to 'tightening regulations around location data and the use of mobile device IDs for ad targeting' and said DOOH measurement is 'on our road map to look into for next year'. | 2, 6, 9 | vendor assertion | supported with qualification | medium | [335] |
| LOC-VISTAR-F25 | Vistar's T-Mobile partner page describes the segments as 'built from app ownership data, from devices within T-Mobile’s network' with use cases such as targeting people with competitor shopping apps 'near your store locations', while the 5 October 2026 release adds 'location insights' to the signal list. | 2, 6, 9 | synthesis | supported as stated | medium | [492, 334] |
| LOC-U-F09 | Arity, the driving-data company whose SDK data collection is the subject of a Texas action (workstream H), sells a named advertising audience product (Arity Audiences) alongside Arity SDK, Geosight and Routely, and announced on 16 October 2026 an agreement to acquire substantially all the assets of INRIX, expected to close in 2026. | 2, 7 | vendor assertion | supported as stated | medium | [493, 494] |
| LOC-UNACAST-F01 | On 1 September 2026 Pine59 announced that it is &quot;the new home of Unacast's former Aggregated Location Insights business&quot; after &quot;the completion of a management buyout&quot; led by two Unacast co-founders and its former CTO; docs.unacast.com, which documents the machine-learning foot-traffic datasets, now says it &quot;has moved to Pine59&quot;. | 2, 7 | direct record | supported as stated | medium | [326, 413, 325] |
| LOC-D-F20 | Media and Games Invest plc announced on 23 January 2020 that its subsidiary gamigo AG had agreed to acquire substantially all assets of Verve Wireless Inc. (technology, IP and team) for a price 'expected to be in the low double-digit million range'; MGI renamed itself Verve Group SE with immediate effect on 13 June 2024; the 2020 event is an acquisition by asset purchase and the 2024 event a rebrand. | 2, 8, 13 | vendor assertion | supported as stated | high | [495, 496] |
| LOC-D-F31 | Perion Network completed its acquisition of Hivestack Inc. (Montreal, founded 2017, 32 countries) on 12 December 2023 for US$100 million in cash at closing plus a three-year retention and performance plan of up to US$25 million; the event is an acquisition (completed). | 2, 9 | vendor assertion | supported as stated | high | [35] |
| LOC-D-F32 | Broadsign announced on 25 November 2025 (per AV Network; the cited Inavate copy is dated 2 December 2025) that it had acquired the DOOH SSP Place Exchange, with a minority investment from Crestline Investors, terms undisclosed, a combined programmatically transactable network of 1.8 million screens and Place Exchange's PerView measurement product; the event is an acquisition (announced as completed, though the release also uses 'is acquiring'). | 2, 9 | vendor assertion | supported with qualification | medium | [45] |
| LOC-PLACER-F19 | Placer documents live delivery routes beyond its dashboard: a published REST API reference (PAPI v1.2.1), scheduled bulk feeds in CSV or Parquet, Snowflake private sharing into the buyer's account, distribution inside the Bloomberg Terminal's {ALTD} alternative-data platform (since September 2023; Data Entitlements from February 2026, per Placer), and, as a per-account enterprise entitlement, a remote MCP server (&quot;Pi for Placer MCP&quot;) whose documentation page was last updated 22 October 2026 and which is listed in Claude's connector directory and the ChatGPT store; its launch date is not established. | 2, 10 | vendor assertion | supported with qualification | high | [280, 281, 282, 224] |
| LOC-G-F03 | App Tracking Transparency was announced on 22 June 2020 as an iOS 14 requirement, deferred on 3 September 2020 to 'early next year', and became mandatory for all apps on 26 April 2021 with iOS 14.5; Apple's definition of tracking includes sharing user or device data with data brokers. | 3 | vendor assertion | supported as stated | high | [68, 76, 77, 78, 69] |
| LOC-G-F04 | Apple introduced privacy manifests, SDK signatures and required-reason APIs at WWDC23; on 7 December 2023 it set the requirement for 'spring 2024', and on 29 February 2024 it dated it: warning emails from 13 March 2024 and, from 1 May 2024, App Store Connect rejects uploads that use a required-reason API without an approved reason; a manifest's declared tracking domains are blocked unless the user has granted ATT permission. | 3 | vendor assertion | supported with qualification | high | [497, 79, 80, 498] |
| LOC-G-F05 | Apple's WWDC24 session introduced CLServiceSession, under which Always authorization is effective only while the app holds a session started in the foreground and approximate location is delivered only every 15 to 20 minutes; iOS 18 shipped on 16 September 2024, iOS 26 on 15 September 2025 and iOS 27 on 14 September 2026. Apple's Core Location change log lists no change after June 2024, and no change to the user-facing location permission in iOS 26 or iOS 27 was found in the Apple pages screened (WWDC25 and WWDC26 session listings were not opened). | 3 | vendor assertion | supported with qualification | medium | [73, 74, 499] |
| LOC-G-F08 | Android 11, which began rolling out on 8 September 2020, added a one-time ('Only this time') location grant, removed the background-location button from the runtime dialog so that 'Allow all the time' can be set only in Settings, required apps targeting Android 11 to request background location separately from foreground location, and auto-resets permissions of unused apps. | 3 | vendor assertion | supported as stated | high | [84, 85] |
| LOC-G-F10 | Google Play announced on 19 February 2020 that background location access would need approval, with a projected timeline (subject to change) of 3 August 2020 for new apps and 2 November 2020 for existing apps, and said on 8 September 2020 it would not enforce the policy for existing apps until 2021. The current Play policy limits all device location data, foreground and background, to essential functions central to the app's core purpose, says it is never permitted solely for advertising or analytics, and reserves background access for critical features. | 3 | vendor assertion | supported with qualification | high | [88, 84, 94, 500] |
| LOC-G-F11 | The only platform-attributed effect figure found: Google stated on 19 February 2020 that on Android 10 'over half of users select &quot;While app is in use&quot;' when offered the new location options, and that many apps requesting background location did not need it (coded 'documented' under Q4 as an official statement with a figure; the effect is on permission grants, not on any vendor's panel). | 3 | vendor assertion | supported as stated | medium | [88] |
| LOC-G-F12 | Google Play's User Data policy, as read on 24 October 2026, lists device location among personal and sensitive user data, prohibits selling it (defining sale as transfer to a third party for monetary consideration) and requires developers to ensure their SDK providers do not sell such data from the app; the Permissions policy adds 'Don't sell device location' and 'Don't use device location solely for advertising or analytics purposes'. | 3 | vendor assertion | supported as stated | medium | [501, 94] |
| LOC-G-F15 | On 17 October 2025 Google announced it would retire the Attribution Reporting API, Protected Audience and Topics on both Chrome and Android, together with Protected App Signals, SDK Runtime, IP Protection, On-Device Personalization, Private Aggregation, Related Website Sets and SelectURL, with phase-outs to follow Chrome and Android processes and no retirement dates given; CHIPS, FedCM and Private State Tokens continue. | 3 | vendor assertion | supported as stated | high | [91] |
| LOC-G-F17 | The FTC's January 2024 complaint against X-Mode Social and Outlogic records that the X-Mode SDK had been integrated into more than 300 apps, that X-Mode paid app developers per device whose location the SDK collected, that it also bought location data from data brokers and aggregators and collected it from its own apps, and that it sold precise location data revealing visits to sensitive locations; the complaint does not mention Apple's or Google's December 2020 removal of the SDK, which this stream could not open from a primary source. | 3 | direct record | supported as stated | high | [502, 503] |
| LOC-G-F18 | Near Intelligence's February 2023 registration statement describes Apple's 2020 approximate-location option under 'key trends around the lack of precise location data' and cites Apple's iOS 14 and 15 opt-in measures and Android 12 as privacy measures, without attributing any figure or change in its own data supply to them (coded 'asserted' under Q4). | 3 | direct record | supported as stated | medium | [504] |
| LOC-G-F19 | Telenav sold its advertising business, Thinknear, Inc., to inMarket Media, LLC under an Asset Purchase Agreement dated 8 August 2019, receiving a 14.5% member interest in inMarket at closing and a perpetual licence back, with the business presented as discontinued operations (event class: acquisition by asset sale). | 3 | direct record | supported as stated | high | [505] |
| LOC-RADAR-F13 | Radar’s iOS SDK privacy manifest (file last changed 1 June 2024) declares precise location, coarse location, device ID, user ID, performance and diagnostic data as collected and linked to the user, each with tracking set to false and purposes limited to app functionality and analytics, while the privacy notice lists mobile advertising identifiers among SDK data. | 3 | vendor assertion | supported as stated | medium | [406, 304] |
| LOC-B-F07 | SafeGraph's residing-device sampling rate fell after the COVID-19 outbreak and declined until October 2021, recovered to a peak of 13.4% in May 2022 and dropped to 4.9% in July 2022, a fall the authors say 'may relate' to a data-provider disruption; annual means were 7.2% (2018), 8.1% (2019), 7.2% (2020), 6.6% (2021) and 8.5% (2022). That the series does not contract monotonically across the 2020-2021 permission changes is this review's inference: the authors attribute the decline to the pandemic and do not discuss the permission changes. | 3, 4 | independent measurement | supported with qualification | medium | [95] |
| LOC-B-F17 | An academic working-paper abstract (April 2022) using device-level location data from more than 7 million iOS and Android devices reports that, for a location data provider collecting through multiple apps, the frequency of data collection from iOS devices fell after Apple's App Tracking Transparency policy, driven both by devices becoming unobserved and by observed devices emitting fewer pings. | 3, 4 | synthesis | supported as stated | low | [110] |
| LOC-B-F32 | Every independent SafeGraph ground-truth comparison found uses data from 2018-2020, before or during the iOS 13/14, App Tracking Transparency and Android 12 changes; the only post-2021 independent comparisons are the Advan park series (to January 2023) and the Minnesota Placer.ai/StreetLight case (summer 2021), so published validation does not describe the panels as they existed at the cutoff. | 3, 4 | synthesis | supported as stated | high | [98, 99, 100, 101] |
| LOC-B-F01 | An independent audit of SafeGraph's point-of-interest data (about 47 million US devices) against North Carolina's 2018 general-election voter file found that polling places with older and with more non-white voters had lower coverage: rank correlations of coverage with the share of voters over 65 of -0.14 (p&lt;0.001) and with the share of non-white voters of -0.11 (p=0.0067). | 4 | independent measurement | supported as stated | high | [99] |
| LOC-B-F02 | In the same audit, SafeGraph's election-day marginal traffic correlated with actual turnout across polling places (rank correlation 0.383, p&lt;0.001) but place-level estimates were noisy: the root-mean-squared error was 1,375 voters, and polling places that registered 20 marginal SafeGraph devices had between roughly 300 and 2,300 actual voters. | 4 | independent measurement | supported as stated | high | [99] |
| LOC-B-F03 | Allocating a resource in proportion to SafeGraph traffic rather than to actual voter counts would have under-allocated by 37% to the oldest and most non-white group of polling places and over-allocated by 33% to the youngest and whitest group (both p&lt;0.05), and in a rank regression each percentage-point rise in voters over 65 was associated with a five-place drop in SafeGraph's ranking relative to the true ranking. | 4 | independent measurement | supported as stated | high | [99] |
| LOC-B-F04 | SafeGraph's own 2019 check (vendor blog and notebook) imputed device demographics from each device's home census block group and compared them with census totals by demographic category nationally (r=0.99 for education and income), plus a geographic check of device counts against county population (r=0.97; a block-group correlation of 0.176 per Li et al. 2024); it concluded the panel was 'well-sampled across demographic categories' and did not test bias for place-specific visit inferences. | 4 | vendor assertion | supported with qualification | medium | [99] |
| LOC-B-F05 | Across 2018-2022, SafeGraph's monthly sampling rate (residing devices divided by total population) averaged 7.5%. Monthly values ranged from 4.5% to 14.5% across the national, urban and rural series, and the overall trend peaked at 13.4% in May 2022 before falling to 4.9% in July 2022. In 2019, device counts tracked census population closely at county level (Pearson r 0.986 urban, 0.934 rural) and moderately at tract level (r above 0.683 urban, above 0.830 rural). The urban-tract fit weakened to r-squared 0.375 by 2022. These results compare the panel with population counts and do not validate visit counts. | 4 | independent measurement | supported with qualification | high | [95] |
| LOC-B-F06 | The 2018-2022 census comparison found Hispanic residents slightly over-represented in 2018 and early 2019 and under-represented from March 2020 to July 2021 (state-median bias -0.058 in 2020 and -0.059 in 2021, easing to -0.022 in 2022); low-income and low-education groups were slightly under-represented, more so during the pandemic, but at state level in 2022 the pattern reversed (lower-education and lower-income groups over-represented); most group biases were within about +/-0.05; rural units were sampled at higher rates than urban ones at tract and block-group level from April 2019. | 4 | independent measurement | supported with qualification | high | [95] |
| LOC-B-F08 | SafeGraph-derived origin-destination flows for March-May 2020 (panel described as about 10% of the population; weekly active devices fell from over 22 million to 13 million in the week of 13-19 April 2020) correlated with pre-pandemic ACS commuting flows at county level at 0.974-0.985 for inferred population flows and 0.915-0.961 for raw visitor flows, and with Descartes Labs mobility indices in five metros at 0.92 or above; the authors state that no ground truth exists for these flows. | 4 | independent measurement | supported with qualification | high | [96] |
| LOC-B-F09 | A peer-reviewed study of six western US national parks compared SafeGraph Social Distancing Metric records with National Park Service visitation statistics for 2019-2020 and concluded that device mobility records capture park visitation patterns at finer spatiotemporal granularity; the abstract read reports no correlation statistic. | 4 | synthesis | supported as stated | medium | [97] |
| LOC-B-F10 | In Yellowstone, SafeGraph Patterns visits (January 2018-November 2020) correlated with NPS monthly recreation visits at r=0.88 and r=0.85 for the two most-visited POIs, with a trail counter's daily counts at r=0.76 (2018) and 0.77 (2019), and with park-wide gate vehicle counts at r=0.79 in 2019 but only 0.47 in 2020, when the compared facility imposed indoor capacity limits. | 4 | independent measurement | supported as stated | high | [98] |
| LOC-B-F11 | In the same Yellowstone study only 40 of the 80 SafeGraph POIs located inside the park were valid (the rest had mismatched coordinates or names, mostly attractions), the ACS-imputed visitor demographics differed significantly from the visitor survey on education and income, and short-term international visitors (23% of the park's visitors per NPS surveys) cannot be assigned a home block group. | 4 | independent measurement | supported as stated | high | [98] |
| LOC-B-F13 | A 2025 preprint using Advan Weekly Visit Patterns (about 35 million US devices from about 1,000 apps, described as roughly a 10% sample) for January 2019-January 2023 reports a Pearson correlation of 0.875 between its visit series and National Park Service visitation for the ten most-visited parks, a comparison the authors call qualitative and based on visual inspection. | 4 | synthesis | supported as stated | medium | [100] |
| LOC-B-F14 | The only Veraset-specific coverage study found, a June 2026 preprint benchmarking 2,478 Mexican municipalities against the 2020 census, finds that the multi-app Veraset data concentrate users in larger, wealthier and more digitally connected places, that coverage bias is spatially clustered, and that it differs by source from Facebook's single-app data. | 4 | independent measurement | supported as stated | medium | [105] |
| LOC-B-F15 | Two 2025 preprints (funding not stated) find mobility-data bias in all ten US cities studied but specific to each city, with tract wealth, ethnicity and education predicting how much data residents produce, and, for four UK datasets against the 2021 census, that multi-app datasets do not necessarily reduce bias relative to single-app sources; the US authors say this raises serious questions about general debiasing approaches. | 4 | synthesis | supported with qualification | medium | [108, 109] |
| LOC-B-F16 | For Cuebiq, a 2017 Federal Highway Administration case study comparing its data with a smartphone GPS household travel survey found demographic biases by traveller age and income and a bias toward trips longer than 9 miles, and a 2022 University of Washington analysis of device-to-population ratios in 11 metros found the direction of racial bias varying by metro (other-race residents under-represented by 9.60% against 1.04% for white residents in Los Angeles; the pattern reversed in Boston and Seattle). | 4 | independent measurement | supported as stated | medium | [104, 103] |
| LOC-B-F21 | Location Sciences, which sold the Verify verification product, told the market in September 2019 that of 550 million impressions it verified in the first half of 2019 only 14% carried high-quality GPS signals, that 36% of GPS-enabled applications displayed location fraud and that 65% of advertising budgets were being wasted on mistargeted and poor-quality location data; Verify's revenue in that half was 113,536 pounds. | 4 | vendor assertion | supported as stated | medium | [506, 507] |
| LOC-B-F22 | This review found no peer-reviewed measurement of bidstream location accuracy with published statistics in the sources screened by the cutoff (web searches only; ACM, IEEE, USENIX and PETS databases were not searched directly): the closest academic items are a July 2024 IMDEA Networks talk on bid requests from a mid-size DSP reporting insufficient numeric precision and abnormal spatial concentrations suggesting manipulation (no numbers; no paper found), and the 2017 ADINT study, in which ads bought with a budget of about US$1,000 could localise a device to about 8 metres once it stayed about 5 minutes. | 4 | synthesis | supported with qualification | medium | [508, 509] |
| LOC-B-F28 | Randomised field experiments of location-targeted mobile promotions exist in the academic literature with store-level outcomes: a 2015 JMR geo-conquesting experiment across competitive, focal and benchmark shopping areas found increasing returns to discount depth when targeting consumers near a competitor and decreasing returns at the focal location, and a 2019 Management Science experiment in one shopping mall (83,370 user responses over 14 days in June 2014) found trajectory-based targeting raised redemption probability, redemption speed and transaction amounts. | 4 | synthesis | supported as stated | medium | [510, 511] |
| LOC-B-F30 | The Federal Reserve Bank of Chicago's weekly retail trade index (CARTS) uses retail foot traffic as one of eight weekly indicators and lists SafeGraph and Advan Research among its data sources, benchmarking the index to Census monthly retail sales; this is documented production use of panel foot traffic by a central bank, not a published validation of visit counts. | 4 | direct record | supported as stated | medium | [512] |
| LOC-B-F31 | No independent comparison of any named vendor's projected retail visits with door counts or transactions across many places and months was found in the sources screened by the cutoff, and no vendor publishes a held-out validation with error statistics on the pages opened. The third-party evidence that exists is historical and narrow: a 2018 audit of SafeGraph's panel found lower place-level coverage at polling places with older and non-white voters (adverse evidence for that panel and use), park and trail comparisons from 2018 to 2021, and one Placer.ai trail-site check whose funding is not stated, so it is third-party but not independent under the rules. H2 stays unresolved overall; no single verdict is given across products and data vintages. | 4 | synthesis | supported with qualification | medium | [98, 99, 102] |
| LOC-C-F19 | Unacast's own 'ground truth' post (published 25 February 2021, updated 5 August 2024) compares its average workday foot-traffic estimates at five US manufacturing plants with 2019 employee counts taken from online sources and reports that 'the only outlier was the Everett facility' and 'All other data matched within 20%', with the facility-level numbers shown only in a chart image and the reference being headcount rather than visits. | 4 | vendor assertion | supported with qualification | medium | [14] |
| LOC-C-F20 | Huq (2026-08-06) publishes a median MAPE of 14.5% for its footfall estimates 'against ground truth in Germany', tested against 'hundreds of authoritative external sources', without naming the references, the site count or the period; Advan (2019) guarantees '99.5% accuracy of geofences' across 1.2 million locations, a claim about geofence geometry rather than visit counts; neither is a US visit-count validation. | 4 | vendor assertion | supported as stated | medium | [116, 117] |
| LOC-C-F26 | Cuebiq's 2017 benchmark paper asserts, citing an MMA report released in October 2015 ('Demystifying Location Data Accuracy', gated behind MMA member login), that 'less than 33% of location data collected through ad exchanges' is accurate within 100 metres, a vendor-cited figure about bidstream location whose underlying sample and method neither this stream nor its verifier could read. | 4 | vendor assertion | supported with qualification | low | [118] |
| LOC-E-F31 | Of the opened documentation for the fourteen cohort A products, only Azira's undated white paper reports a comparison of its visit numbers with an independent reference: a vendor-asserted median absolute percent error of 10 per cent and a correlation of 0.85 between extrapolated visits and 'ground truth sales data and foot traffic counters', with no study design, sample, period or places disclosed. No product publishes such a comparison as a documented study. The other validation claims found are bare assertions: InMarket's 'Comscore-Verified Visit Methodology' and GroundTruth's Blueprints 'verified using a selection of trusted location sources'. | 4 | synthesis | contradicted | medium | [274, 254, 5] |
| LOC-FOURSQUARE-F11 | Foursquare's only published validation of its attribution panel is a trend check of scaled transactions against 'public financial statements' and of visitation trends against transaction trends, with no places, period or statistic; the 'Third-party verified' Forrester item on the product page is a Forrester Consulting Total Economic Impact ROI study commissioned by Foursquare (June 2022), not a validation; this review found no independent ground-truth comparison of Foursquare Attribution or Visits numbers in the sources screened by 24 October 2026. | 4 | vendor assertion | supported with qualification | medium | [236] |
| LOC-FOURSQUARE-F26 | A 2013 third-party user study (244 users, January to July 2013; the paper carries no funding or acknowledgement statement) found that Foursquare check-ins covered roughly 10% of the locations users visited and that about 75% of check-ins did not match a GPS-detected visit; Foursquare today trains its Snap-to-Place visit model on 'human-verified' check-ins, so the study bears on that ground truth, though its data predate the current products by a decade. | 4 | synthesis | supported with qualification | low | [355, 28] |
| LOC-GROUNDTRUTH-F15 | GroundTruth's ground-truth evidence is a vendor-commissioned receipt-panel comparison: GroundTruth commissioned InfoScout (now Numerator) in 2017, and its 2019 post reports the 'third annual' audit at '94.4%* visit accuracy', footnoted as an analysis of '10,000 matched panelists who shopped at a sample of key retailers' and submitted receipts; help pages later claim '99% accurate as verified by Numerator'. The sample size and reference (receipts) are stated, but no report, period, retailer list, method or error statistic is published, and the measure checks whether recorded visits are confirmed rather than whether visit counts match a reference. | 4 | commissioned measurement | contradicted | medium | [259, 258] |
| LOC-INMARKET-F09 | No published comparison of InMarket’s visit numbers with an independent reference (door counts, transactions or official visitor counts) with error statistics was found; validation statements are vendor claims naming Comscore (‘Comscore-Verified Visit Methodology’; ‘LCI Attribution Comscore validated for visit metrics accuracy and aptness’, 2020) and an unnamed set of ‘truth sets’, with no report, date or scope published. | 4 | vendor assertion | supported as stated | medium | [274, 40, 267, 374] |
| LOC-PLACER-F09 | In the only third-party comparison of Placer.ai found (funding not stated) in the sources screened by the cutoff, a University of Minnesota Extension case study at one Minnesota recreation area (29 May to 6 September 2021), Placer.ai estimated 13,233 visits against 10,138 counter-recorded visits, 30.5% above, and StreetLight 14,544 (43.5% above); the author reports no statistically significant difference, but the platforms showed weekends 7 times busier than weekdays against 2.1 times in the counter averages. | 4 | synthesis | supported with qualification | medium | [101] |
| LOC-PLACER-F10 | The same Minnesota case study found Placer.ai's inferred visitor demographics differed significantly from the intercept survey on race, education and income: for example 30.8% with a bachelor's degree or higher and 21.8% with household income over US$100,000 in Placer.ai data, against 62% and 61% in the survey. | 4 | synthesis | supported as stated | low | [101] |
| LOC-PLACER-F11 | Placer publishes accuracy claims without method: its home page states &quot;+92% accuracy&quot;, a California city's 2026 fact sheet relays that &quot;Placer.ai claims 92–96% accuracy&quot; with correlations to ground truth &quot;consistently exceeding 90%&quot;, and the help page names references (Major League Baseball attendance, transactions, store revenues, conference attendees, vehicle counters) without any result, sample or statistic. | 4 | vendor assertion | supported as stated | high | [293, 290, 102, 289] |
| LOC-PLACER-F26 | An interested site-selection vendor's blog reports that, across 12 retailer-quarter comparisons for six large US retail and restaurant chains (Q1 2025 to Target's Q2 FY26), Placer.ai visit trends and the companies' SEC-reported traffic or transaction figures &quot;agreed on direction 9 times out of 12&quot;, with a median gap of about 2.2 percentage points. | 4 | commissioned measurement | supported with qualification | low | [383] |
| LOC-UNACAST-F16 | The only accuracy statistics Unacast publishes for its foot-traffic estimates are an R-squared &quot;of 91.6% or higher&quot; against the foot traffic of one unnamed nationwide clothing retailer, with no places, period, sample or holdout stated, and a 2021 post (updated 2024) saying workday estimates at four of five US manufacturing plants &quot;matched within 20%&quot; of 2019 online headcounts (Everett the outlier) and that stadium visits matched approximate capacity in October 2018 to February 2019; per-site values appear only in chart images. | 4 | vendor assertion | supported with qualification | high | [106, 14] |
| LOC-B-F24 | Across 663 Facebook experiments with more than 5,000 user-level features, double/debiased machine learning and stratified propensity-score matching could not recover experimental lift: median RCT lifts were 29%, 18% and 5% for upper-, mid- and lower-funnel outcomes, against 83%, 58% and 24% from DML and 173%, 176% and 64% from SPSM, and the authors conclude they were unable to reliably estimate a campaign's causal effect. | 4, 5 | synthesis | supported as stated | high | [112] |
| LOC-B-F25 | Even with randomisation, advertising effects are hard to measure: across 25 large field experiments with US retailers and brokerages (2.8 million dollars of digital spend, most reaching millions of customers) the median confidence interval on return on investment was over 100 percentage points wide, individual sales had a standard deviation about ten times their mean, informative experiments can require more than 10 million person-weeks, and the authors call selection bias from targeting a crippling concern for observational methods. | 4, 5 | synthesis | supported with qualification | high | [113] |
| LOC-B-F26 | A randomised experiment with 1.6 million customers of a major retailer on Yahoo! measured a 5% increase in purchases from online ads, 93% of it in brick-and-mortar stores and 78% from consumers who never clicked; a cross-sectional observational estimator on the same data showed a large negative bias, three times the experimental effect (it understated the effect), while a difference-in-differences estimator served as a specification check. | 4, 5 | synthesis | supported with qualification | high | [114] |
| LOC-B-F27 | Two randomised designs that location attribution can borrow have published methods: ghost ads (Google-affiliated authors, 2016-2017), which identify the control-group counterparts of exposed users so that lift can be measured without PSA controls, which the authors say become invalid under performance-optimised delivery, and geo experiments (Google, 2011), in which non-overlapping geographic regions are randomly assigned to treatment or control and realised through geo-targeted advertising. | 4, 5 | synthesis | supported as stated | high | [115, 513] |
| LOC-B-F23 | Fifteen US Facebook advertising campaigns run as randomised controlled trials in 2015 (500 million user-experiment observations, 1.6 billion impressions) were used to test observational methods. Matching and regression methods using rich individual-level data mostly overestimated lift relative to the experiments, and in half of the studies the purchase lift was off by a factor of three across all methods. The study is an academic-Facebook collaboration on Facebook's own data: two coauthors are affiliated with Facebook, and the two Kellogg academics declare no financial interest or compensation. | 4, 5, 12 | synthesis | supported with qualification | high | [111] |
| LOC-B-F19 | Thinknear editions after 2015 showed no improvement on its own measure: a May 2016 report put 32% of real-time location data from publishers and exchanges within 100 metres (industry score 45, down from 50, the lowest since tracking began in 2014) and 7% of data targeting 100-1,000 m accurate; an undated reseller deck that advertises the Q4 2017 index shows, on an unlabelled slide, 37% of impressions within 100 m, 35% within 100-100,000 m and 28% off by more than 100,000 m (the slide glosses this band as 6 miles, about 10 km, so the threshold is ambiguous). | 4, 6 | vendor assertion | supported with qualification | medium | [514, 515] |
| LOC-BLIS-F13 | Location Sciences' 2019 verification of Blis location data ('near perfect scores (99%)') was a paid engagement, and in its final results for 2020 (RNS, 18 March 2021) Location Sciences stated that Blis Global Ltd, 'one of Verify's key customers', threatened legal action after Verify identified and reported 'data manipulation', that it rejected the claim as baseless, and that Blis had not followed up; Adweek reported on 15 December 2020 that Blis denied the allegations. | 4, 6 | direct record | supported as stated | medium | [343, 231, 232, 233] |
| LOC-I-F04 | Geopath's reach-and-frequency approach introduced with its 2023 data release is 'centered around observed mobile location data from smartphone applications and connected cars', uses machine learning trained on that observed data, and is self-reported to align 'with the observed data with ±5% accuracy'; the page (updated 15 November 2022) names no provider, sample size or validation method. | 4, 8, 9 | vendor assertion | supported as stated | medium | [184] |
| LOC-B-F12 | In the only third-party comparison of Placer.ai found in the sources screened (a University of Minnesota Extension study whose funding is not stated), the case study (April 2024) at one Minnesota recreation area, Placer.ai estimated 13,233 visits and StreetLight 14,544 for 29 May-6 September 2021. Those totals are 30.5% and 43.5% above the 10,138 recorded by trail counters. StreetLight's figure is vehicle trips multiplied by a median of three passengers. The authors found no statistically significant difference, but the comparison is not like for like: polygons differ, counters cover only the cross-country trails, and events are handled differently. The 7x versus 2.1x weekend ratio and the demographic differences should be read with those limits. | 4, 11 | synthesis | supported with qualification | medium | [101] |
| LOC-C-F05 | Azira's methodology page (footer 2025) is the only cohort A vendor document opened in this stream that documents a randomised control as the default: 10-15% of impressions serve public-service creative as a PSA holdout, with a synthetic control from its identity graph as the alternative, polygon-based visits, extrapolation of observed visits by a detection probability capped at 20x, minimums of about 1,000 observed visits and 100,000 exposed devices for breakouts, 99% confidence intervals and a two-tailed t-test. | 5 | vendor assertion | supported as stated | medium | [5] |
| LOC-C-F06 | Azira's Domestos case (Rize, Turkey; undated) reports an exposed footfall rate of 1.672% against a control rate of 1.168%, an 'Attribution Lift Index' of 143.151 presented as a 43% increase in store traffic against a stated 27% market benchmark, from 236,495 exposed users, 3,955 tracked visits and 577,978 impressions, without stating the control design, window or dates. | 5 | vendor assertion | supported as stated | medium | [516] |
| LOC-C-F07 | Cuebiq's 2017 benchmark paper (client reports Q3 2016 to Q1 2017; database of 61 million anonymous users) defines uplift against a control of unexposed users matching the exposed group's criteria, publishes benchmark uplift ranges of 10-30% for retail and big-box, 10-40% for QSR and 20-45% for automotive, and reports average campaign uplift between 18.62% and 37.70% across the three quarters. | 5 | vendor assertion | supported as stated | medium | [118] |
| LOC-C-F08 | Cuebiq's 2022 benchmark report (campaigns run in 2020-21) sets vertical thresholds at the 33rd and 66th percentiles of its client reports: retail uplift 0-1% low, 1.01-7.61% medium, 7.62-52.5% high; QSR uplift 0-3.88%, 3.89-11.86%, 11.87-38.5%; grocery uplift 0-1%, 1.01-11.25%, 11.26-35.7%; it defines uplift against a control group and adds an 'Incrementality Effect' (share of visits that are incremental) and a cost per incremental visit. | 5 | vendor assertion | supported as stated | medium | [517] |
| LOC-C-F10 | adsquare's Attribution dashboard (page dated 2025-05-20) reports uplift from a 'Synthetic Control Condition model' that simulates unexposed behaviour after adjusting for weather, seasonality, local footfall trends and public holidays, which the page says 'removes reliance on traditional control groups', with visits matched to 2D POI polygons from SDK data collected with opt-in; no uncertainty or significance reporting appears on the page. | 5 | vendor assertion | supported as stated | high | [21] |
| LOC-C-F24 | Azira is the only vendor whose public pages opened disclose numeric minimum-sample and significance rules for visit lift (about 1,000 observed visits and 100,000 exposed devices; 99% confidence intervals; two-tailed t-test). InMarket's thresholds (reported as 10 or 25 million impressions and &quot;&gt;95%&quot; confidence) sit in a glossary behind a sign-in and could not be verified from a public route; Choozle documents a different minimum, 30 days and 1 million impressions a month, for its InMarket Visit Stream integration only. Reveal's measurement page discloses confidence intervals and p-values without a numeric minimum; Vistar states only that it requires &quot;enough scale&quot;; Foursquare, Cuebiq, Adsquare, Blis, Simpli.fi and GroundTruth state none on the pages opened. | 5 | synthesis | supported with qualification | medium | [17, 5, 8, 518] |
| LOC-E-F01 | Foursquare Attribution's documented default lift method is a modelled control: the 2022 method post describes regression adjustment that builds a synthetic control from static features, followed by difference-in-differences against pre-exposure visitation; the current developer docs describe a 'modeled control group' weighted on 10+ attributes and a 'model-based approach'. No randomised or holdout option appears in the opened documentation. | 5 | vendor assertion | supported with qualification | high | [119, 16] |
| LOC-E-F03 | Foursquare Attribution documents measurement across digital, social, audio, linear and connected TV and out-of-home, GPS pings from first-party and vetted third-party sources, OOH proof-of-play log ingestion, a Spotify integration and '550+ integrated partners', with daily dashboard updates; it documents no pricing, no numeric minimum sample, no significance rule and no visit-definition thresholds, but it does document a mandatory pre-campaign feasibility gate (a score from estimated impressions) and releases reports only once results are 'statistically viable'. | 5 | vendor assertion | supported with qualification | high | [119, 28] |
| LOC-E-F04 | InMarket's LCI documents a matched control by default: control devices are drawn from visitors to the same locations with employees removed and matched 1:1 to exposed devices on eight audience dimensions, and lift is the difference between exposed and control changes in visits two weeks after versus two weeks before exposure. | 5 | vendor assertion | supported as stated | medium | [17, 11] |
| LOC-E-F05 | InMarket's LCI console gives two different denominators for LCI Lift on the same page: the Glossary divides the exposed-minus-control lift by 'Visits', the Calculations section by '(Exposed Visits + Control Visits)'. | 5 | vendor assertion | supported as stated | high | [17] |
| LOC-E-F07 | InMarket's legacy LCI console glossary, read once during the research and since behind a sign-in, documented minimum samples and uncertainty: 10 million impressions for mobile app with valid device IDs and 25 million for mobile web or desktop per measured dimension, a standard error on the lift and a confidence level where '&gt;95%+ constitutes a valid measure', with waivers subject to a feasibility assessment. These thresholds cannot be reproduced from a public route at the cutoff; the synthetic-control design itself is also described in InMarket's public July 2026 post. | 5 | vendor assertion | supported with qualification | medium | [17] |
| LOC-E-F13 | GroundTruth's documentation states a payment model: Cost Per Visit, which it says it introduced in spring 2017, under which 'marketers only pay for ads that result in a store visit'; visit measurement is the basis of a media buy here. Whether measurement is also sold separately is not stated, and among the other thirteen products the opened pages state no payer or fee. | 5 | vendor assertion | supported with qualification | medium | [266] |
| LOC-E-F15 | Reveal's current product documentation names only an OOH 'Net Lift Study' built on a synthetic control chosen by propensity-score matching from a 'persistent panel of 60M US DAUs', with confidence intervals and p-values 'computed using closed-formed formulas and simulations'; its sample dashboards show 95 per cent confidence, but no confidence level is stated as a rule; VISIT Local is not named on the current site, and the 2021 case-study URL now serves the homepage. | 5 | vendor assertion | supported with qualification | medium | [20, 518] |
| LOC-E-F16 | Azira's white paper defines lift as (exposed rate - control rate) / control rate, but then computes incremental visits as 'Exposed Group Lift * Exposed Group Devices'. Read literally, that multiplies a relative lift by a device count. Under the paper's own definition, incremental visits are (exposed rate - control rate) x exposed devices. Do not reproduce or endorse the published formula without that correction. | 5 | vendor assertion | supported with qualification | high | [5] |
| LOC-E-F23 | Cuebiq's 2026 measurement content documents a behaviourally matched control (unexposed consumers similar in movement and shopping behaviour), opted-in device location data, linear TV exposure from broadcast partners' impression logs at household level, and coverage of linear TV, CTV, mobile, OOH, social and streaming audio, reporting incremental store visits and cost per incremental visit; no named product, visit definition, threshold or significance rule appears. | 5 | vendor assertion | supported as stated | medium | [22] |
| LOC-E-F27 | Precisely's audience product page documents no named PlaceIQ attribution product: measurement appears as a capability of Precisely Audiences ('tying media exposure to verified store visits and purchase activity') built from 'anonymized mobile signals', with activation partners listed and no measurement partner, control design, visit definition or panel size. | 5 | synthesis | supported as stated | medium | [27] |
| LOC-E-F33 | Every cohort A product with a documented control design at the cutoff names at least one non-mobile channel it measures: CTV or TV by Foursquare, InMarket, Azira, Blis, Cuebiq and Arrivalist; DOOH or OOH by Foursquare, InMarket, Azira, Blis, Reveal (OOH only), Adsquare, Cuebiq, Arrivalist and GroundTruth (DOOH views via Blueprints); audio by Foursquare, Azira, Blis, Cuebiq and Arrivalist. Lifesight's help page and Vistar's partner page also name OOH or DOOH without a documented method, and Precisely's CTV is an activation channel, not a measured one. | 5 | synthesis | supported with qualification | high | [28, 19, 5, 25, 27] |
| LOC-BLIS-F05 | Blis documents Smart Holdout Groups as a geographic holdout: areas are scored on seven data points including demographics, household characteristics and income, pairs of similar areas are matched with one designated control and one target, and buffer zones built from Blis movement data are left unserved and unmeasured; the documents do not state that designation within a pair is random. | 5 | vendor assertion | supported as stated | high | [123, 19, 124] |
| LOC-BLIS-F06 | Blis announced the US rollout of Smart Holdout Groups on 11 July 2024 ('now available in the US') and the global release on 9 October 2024; footfall lift is 'Only available in the US', app-usage lift draws on UK mobile-network data and is 'Coming soon to the US', and in October 2024 Blis said clients 'will soon' be able to choose their own control factors. | 5 | vendor assertion | supported as stated | high | [123, 124, 19] |
| LOC-BLIS-F07 | On the product page, both launch releases and the technology page, Blis publishes no visit definition, no source for the footfall outcome, no minimum sample, significance rule or confidence interval and no projection method for Smart Holdout Groups, while describing it as 'statistically robust' and stating that measurement 'can be independently verified'. | 5 | synthesis | supported as stated | medium | [124, 229, 19, 519] |
| LOC-BLIS-F16 | Blis's privacy centre lists location-attribution and audience-measurement companies among its 'Ad Effectiveness &amp; insights' partners, whose role includes verifying 'whether a user visited a store after seeing an ad'; the listed partners include Cuebiq, Foursquare, InMarket, Kochava, Lifesight and Samba TV, and no source says which, if any, supplies Smart Holdout Groups' footfall outcomes. | 5 | vendor assertion | supported as stated | medium | [342] |
| LOC-BLIS-F17 | Blis's US restaurant case study reporting a '55% store visit uplift (control vs exposed)', a 0.76% store-visit rate against a '0.5%' benchmark and 36% same-day visits was published on 12 September 2019, about five years before Smart Holdout Groups, and describes location audiences built from devices 'seen in competitor smoothie bars, restaurants or similar brands in close proximity'. | 5 | vendor assertion | supported as stated | medium | [13] |
| LOC-FOURSQUARE-F07 | Foursquare Attribution's visit numbers are projected, not observed: the 2024 method describes tenure filters on third-party panelists, census weights at census block-group level with demographics predicted from behaviour, place weights, and 'projection weighting' that scales matched panelists to each publisher's delivered impressions. | 5 | vendor assertion | supported as stated | high | [236, 120] |
| LOC-FOURSQUARE-F08 | Foursquare Attribution publishes no numeric minimum sample or significance rule: campaigns pass a feasibility score (Excellent, Good or 'Consult Foursquare') computed from estimated impressions, reports appear 'once the results are statistically viable' at least two weeks after launch and are finalised after the conversion window plus 7 days, impression minimums are given only through the support portal, and the calculator shows store-level feasibility and online or hybrid conversions as 'currently unavailable'. | 5 | vendor assertion | supported as stated | high | [247, 246] |
| LOC-FOURSQUARE-F09 | Foursquare's documentation page describes its output as 'incremental lift' measured against a modelled expected visit rate, while reports and case studies use 'Behavioral Lift' (also 'proprietary visitation behavioral lift', S23), which no opened page defines; separately, Foursquare says it 'cannot offer a direct view into the calculation' of the star ratings that compare partners' metrics, Behavioral Lift among them. | 5 | vendor assertion | supported with qualification | high | [119, 248] |
| LOC-FOURSQUARE-F12 | Foursquare's published Attribution results carry no design detail: a 2026 linear-TV QSR case reports 1.32% behavioral lift, 3.4M+ incremental and 58.9M total attributed store visits, and the product page reports 9.2%, 7.8% and 2.5% lift figures and a '23% lift in in-store visits among exposed users versus the control group', none with exposed and control rates, sample, period or control construction. | 5 | vendor assertion | supported as stated | medium | [351, 28] |
| LOC-GROUNDTRUTH-F09 | GroundTruth's Observed Visits counts a user's visit to the brand or 'drive-to' location within 14 days of exposure to a GroundTruth ad, credited by last touch, and shows only 'verified' visits after a 7-day verification period (rolled out to accounts created after 17 September 2024); the help centre says observed visits 'may be lower than actual visits'. | 5 | vendor assertion | supported as stated | high | [6] |
| LOC-GROUNDTRUTH-F12 | GroundTruth's Projected Visits applies the visitation of a per-campaign panel of impressed devices 'with frequent location signals' to all impressed devices, using a proprietary model that includes WeatherBug first-party data, appears once a campaign usually reaches 40 Observed Visits and 350,000 impressions, is cumulative and rounded, and reports no uncertainty. | 5 | vendor assertion | supported as stated | high | [18] |
| LOC-GROUNDTRUTH-F13 | GroundTruth's visit measurement reports exposed-only visitation with no documented control group: Visitation Rate is Visits / Impressions; its '89% Incremental Visits' case defines 'incremental' as visits from newly acquired customers, not visits above a control; the 'more than half' incremental figure for 42,000 DOOH-exposed visits states no comparator; and its only comparative metric, Market Share, is a directional pre/post trend in share of all visits that the vendor says other factors may drive. | 5 | synthesis | supported with qualification | high | [6, 122, 257] |
| LOC-GROUNDTRUTH-F26 | GroundTruth sells a Cost Per Visit buying model, which it says it introduced in spring 2017, under which 'marketers only pay for ads that result in a store visit'; its help centre labels it a managed offering, and a November 2024 case reports an auto dealers' association paying 'only for verified visits' for 'over 1,200 visits' in a month, with no cost, impressions or control. | 5 | vendor assertion | supported as stated | medium | [266, 258, 367] |
| LOC-INMARKET-F03 | The only detailed public LCI method, the undated NinthDecimal-era glossary on InMarket’s LCI console, documents an audience-matched control (1:1 on eight demographic, travel and technology traits, drawn from devices seen at the same client-specified locations with employees removed), lift as the exposed minus control change in visits two weeks after versus two weeks before exposure, a standard error, a ‘&gt;95%’ confidence rule and minimums of 10 million (mobile app with device IDs) or 25 million (mobile web/desktop) impressions per campaign and per report dimension. | 5 | vendor assertion | supported as stated | medium | [17] |
| LOC-INMARKET-F04 | Across InMarket’s own and partner documentation, every lift design described is matched or modelled: a 1:1 audience-matched device control (console), a synthetic control from historical visitation (2026 Visit Lift), ‘similar control users’ (TikTok, 2024) and a synthetic matched household control for CPG sales (2026); Visit Stream and the Yahoo DSP feed are observed visits for correlation and ‘directional’ reporting with no control. | 5 | synthesis | supported as stated | high | [11, 272, 270, 23] |
| LOC-INMARKET-F05 | LCI’s headline store-visit totals are projected, not observed: the console defines Overall Projected Store Visits as exposed device visits rated up by a device-to-DMA ratio (with a ± range from a regression of population against device data across DMAs) and Incremental Projected Store Visits as that total times the LCI Lift; only ‘Exposed Device Visits’ are panel observations. | 5 | vendor assertion | supported as stated | medium | [17] |
| LOC-INMARKET-F20 | InMarket’s visit-lift measurement reaches buyers through partner platforms documented by the partners or in InMarket releases (TikTok from 2024, Reddit for all US advertisers from 12 November 2024, Vistar Media for US DOOH, Yahoo DSP for a daily observed-visit stream in the US and Canada from 23 April 2026, Basis from 27 May 2026), plus its own console, exports and a daily Visit Stream feed; no developer API reference was found. | 5 | vendor assertion | supported as stated | medium | [11, 269, 23, 270, 271, 272] |
| LOC-INMARKET-F22 | InMarket publishes headline lift and return figures for this row with no design, base or uncertainty beyond its own descriptions: 121 of 126 TikTok campaigns with ‘statistically significant lift’ (2024), a 26% incremental visit lift against a 23% benchmark with 285,000 incremental visits for an unnamed dining chain (2026), a 40% incremental visit lift for an unnamed restaurant chain on CTV, ‘4.6M Incremental Visits’, an ‘11x’ media-effectiveness gain (InMarket and CMO Council, June 2023) and a 25% rise in repeat store visits (Forrester TEI, July 2021). | 5 | vendor assertion | supported as stated | medium | [11, 121, 267, 375] |
| LOC-C-F13 | Simpli.fi's case studies count a visit when a device served the advertiser's creative later enters a Conversion Zone drawn around the store, an exposed-only design: a western-US QSR with 61 locations reported 12,768 in-store visits at a $1.70 cost per visit with a 30-day look-back (case dated April 2025), and a fast-food brand with 353 locations reported more than 215,000 visits at $1.09 against a $5 goal plus a 'Geo-Conversion Lift' of 31.76% defined as the difference between visitors who had received an ad and visitors who had not, with no matching described. | 5, 6 | vendor assertion | supported as stated | medium | [10, 520] |
| LOC-A-F13 | Both MRC location accreditations on record since 2020 covered data sets, not a visit-lift or attribution metric; in the 35 MRC communications from January 2020 to October 2026 screened by name for Foursquare, GroundTruth, Placed, Cuebiq, NinthDecimal, inMarket, PlaceIQ, Blis and Reveal Mobile, and on the current lists, no lift, incrementality or attribution metric of a location product is recorded. | 5, 8 | direct record | supported with qualification | high | [166, 168, 521, 171] |
| LOC-FOURSQUARE-F04 | MRC granted accreditation for 'location information reported in Visits data set' to 'Foursquare (formerly Placed) Visits data set' in its July-September 2020 actions, suspended it in October 2022 and revoked it in December 2022; Foursquare's own 2020 release said its Attribution product 'has not been audited or accredited by the MRC'. | 5, 8 | direct record | supported as stated | high | [166, 167, 245] |
| LOC-VISTAR-F09 | Vistar documents no attribution method of its own at the cutoff: its TTD activation guide says Vistar 'works with trusted industry leaders' for foot-traffic, brand, online-conversion and sales-lift studies, that 'All studies require a feasibility test and have varying minimums and costs', and that it 'offers device ID passback for free'; vistarmedia.com/measurement returned HTTP 404. | 5, 9 | vendor assertion | supported as stated | medium | [337, 522] |
| LOC-VISTAR-F10 | In a 2022 OAAA-hosted Vistar case for a national retailer (flight 15 August 2021 to 29 January 2022; 22 US states; budget $1.2M+), Foursquare-measured results were 72 million+ impressions, 533,000 store visits within a four-day look-back and a 6.51% lift in visitation 'compared with non-exposed consumers'; the document asserts a control-versus-exposed design without describing how the control was built or its size. | 5, 9 | vendor assertion | supported as stated | medium | [8] |
| LOC-INMARKET-F21 | InMarket announced Maestrø in beta on 18 June 2026 as an agentic optimisation layer that feeds InMarket outcomes to AI agents, sold three ways including a fully managed ‘Guaranteed iROAS’ offer and an ‘Open Access’ route ‘Built on MCP’; no public MCP server or AdCP agent for InMarket was found, so this is an announcement, not a documented protocol implementation. | 5, 10 | vendor assertion | supported as stated | medium | [273, 523] |
| LOC-U-F21 | The author's former and current employers checked by this stream show no named location product on the pages opened: Lotame's homepage (Spherical Platform, Addressable Audiences, Data &amp; Identity Services) has no 'location' or 'geo' text, Verve's homepage names audience and ID-less publisher products without a location product, and the Samba TV consumer site refers business visitors to samba.com, which was not opened. | 5, 13 | vendor assertion | supported as stated | low | [524, 525, 526] |
| LOC-D-F42 | Azira's privacy policy (updated 6 August 2025) documents that it collects personal information from third-party data providers and data brokers, from bid requests received when acting as a DSP, and from 'raw human movement data collected by mobile apps consistent with user permissions'; it therefore documents a blend of partner-sourced app data and bidstream data without proportions. | 6 | vendor assertion | supported as stated | medium | [527] |
| LOC-F-F01 | Foursquare Audience, live in 2026, builds segments from past visitation with location-specific high, medium and low frequency buckets, a recommended look-back of 3 to 6 months (custom ranges up to 18 months), a 5,000-device minimum segment size, and is sold off-the-shelf on The Trade Desk, LiveRamp and DV360; its documentation does not state the collection route of the visits. | 6 | vendor assertion | supported as stated | medium | [528, 459] |
| LOC-F-F02 | GroundTruth Ads Manager (live, US, $0 minimum spend) sells four location-targeting layers documented in its help centre as of 23 September 2026: Target Locations (city to precise business address, lat/long or nationwide), Target by Audience (historic visitation, with Deterministic first-party, Predictive and Datonics 'Premium' tiers), Target Neighborhoods (trade-area lookalikes) and Target On-Premise (real-time, mobile only, with In Store, On Lot, Retail Block and Radial boundaries); the platform calls geofencing 'Location Targeting'. | 6 | vendor assertion | supported as stated | medium | [263, 122] |
| LOC-F-F03 | GroundTruth's privacy policy (effective 2 April 2026) documents a blend of location sources without proportions. It names collection by SDKs, scripts and similar techniques, and lists these sources for geolocation information: consumers, customers, ad inventory supply partners (publishers), service providers, consumer data resellers and other third parties. It does not use the words bid request or bidstream, gives opt-out choices only, and states &quot;We may sell your sensitive personal data.&quot; | 6 | vendor assertion | supported with qualification | medium | [265] |
| LOC-F-F04 | Blis's product documentation could not be read by this stream at the cutoff: the homepage title reads 'Blis, part of T-Ads \| Omnichannel DSP' and the technology page title 'Cookieless DSP Technology \| Geospatial AI \| Blis', but both bodies were truncated or bot-blocked, so Blis audiences are coded 'not disclosed' pending a readable source. | 6 | inference | supported as stated | low | [228] |
| LOC-F-F06 | The FTC's final order with InMarket (finalised 1 May 2024, vote 3-0-2) prohibits InMarket from selling, sharing or licensing any precise location data and from any product or service that categorizes or targets consumers based on sensitive location data. The complaint it settled, announced January 2024, described collection through InMarket's own apps and through third-party apps that use its SDK. InMarket's current data page and privacy policy (30 June 2026) describe the same own-app and partner-SDK route. | 6 | direct record | supported with qualification | high | [132] |
| LOC-F-F07 | Azira Audience (live, 2026) builds segments from real-world visitation, purchase behaviour and intent, at a stated &quot;3 meters of precision&quot; with no method given. It activates through Azira's Advance DSP (claimed 80-90% addressability, January 2025 blog) across desktop, mobile, CTV, audio, DOOH and Meta/TikTok. Azira's data page documents a blend of location sources without proportions: &quot;premium SDK, bid-stream, and verified consumer transactions&quot;. Its privacy page describes a sensitive-location data program that removes sensitive locations from its datasets. | 6 | vendor assertion | contradicted | medium | [458, 529] |
| LOC-F-F09 | Mobilewalla Audience Segments is still presented on a page whose copyright line reads 2025, with 200+ segments built by predictive modelling over app usage, demographics and GPS location data. The company homepage now leads with telecom and consumer-lending AI products. Mobilewalla's business-services privacy notice (updated 9 January 2026) says all its data comes from third-party ad-tech information providers. It names movement-data fields (MAID, latitude and longitude, IP, timestamp) but not whether they come from SDKs or bid requests, and it states that Mobilewalla does not knowingly collect visits to named sensitive locations. | 6 | vendor assertion | supported with qualification | low | [422, 141] |
| LOC-F-F13 | Precisely sells the PlaceIQ audience line as &quot;Precisely Audiences&quot; in 2026; the old placeiq-audiences URL now redirects to /product/precisely-audiences/. The page offers 1,500+ pre-built segments &quot;curated from anonymized mobile signals, demographic profiles, shopping patterns, and categorical interests&quot;, exportable to DSP, CTV or social platforms, and states no collection route, panel or accuracy method. | 6 | vendor assertion | supported with qualification | medium | [421] |
| LOC-F-F16 | Reveal (Reveal Mobile, Inc.) builds location audience segments in 2026 through its Geofencing Solutions, which captures Mobile Advertising IDs at &quot;your locations, competitor locations, and audience hotspots&quot; for activation in The Trade Desk, LiveRamp, Pontiac Media, Basis, Choozle and social platforms. Its OOH platform offers retargeting of OOH-exposed devices across display, social and CTV. Neither page names the source of device location. | 6 | vendor assertion | supported with qualification | medium | [530, 531] |
| LOC-F-F18 | Across the cohort B documentation opened by 24 October 2026 (including InMarket's data page and Azira's audience page), this review found no independent on-target or accuracy validation of a location audience with a published design. The accuracy statements are vendor assertions without method: GroundTruth's Predictive Audiences &quot;over 90% predictive accuracy&quot;, Foursquare's &quot;industry's most accurate location-based datasets&quot;, Mobilewalla's age and gender &quot;verified&quot; against Nielsen DAR and US Census, InMarket's 3.7x against Wordstream display benchmarks, Azira's 80-90% addressability and &quot;3 meters of precision&quot;, and Simpli.fi's plat lines &quot;for maximum accuracy&quot;. InMarket also asserts that its location data is &quot;Comscore verified and Location Sciences certified&quot;, naming certifiers but publishing no certificate or scope. | 6 | synthesis | supported with qualification | medium | [263, 459, 422, 268, 529, 456] |
| LOC-F-F19 | None of the cohort B product pages opened describes sensitive-place suppression, but vendor privacy documents do. Foursquare commits to the NAI enhanced standard and lists 13 sensitive-place categories. Mobilewalla says it does not knowingly collect visits to places of worship, shelters, treatment centres or reproductive health clinics. Azira describes a sensitive-location data program that removes such data. These are self-attested policy statements with no verified enforcement, alongside FTC orders imposing programs on InMarket, Mobilewalla, and Kochava with Collective Data Solutions. GroundTruth's notice that it 'may sell your sensitive personal data' is a state-law disclosure and does not show that it sells sensitive-place visits. | 6 | synthesis | contradicted | medium | [265, 263, 132, 140, 144, 459] |
| LOC-F-F22 | Geofencing and geo-conquesting are sold to US local advertisers in 2026 through at least three self-serve or managed routes with different mechanics: GroundTruth Ads Manager (real-time On-Premise and 'Pure Geographic' delivery to anyone inside a boundary, plus retargeting of 'your competitor's' past visitors, no minimum spend), Simpli.fi Addressable Geo-Fencing (parcel fences around up to one million addresses with 30-day cross-device follow-up) and Reveal's Geofencing Solutions (MAID capture at own and competitor locations for DSP and social activation). | 6 | synthesis | supported as stated | medium | [263, 122, 456, 530] |
| LOC-F-F24 | GroundTruth Ads Manager documents six channels (mobile, CTV and OTT, desktop and tablet, audio, DOOH with &quot;1M+ digital out-of-home screens across the US&quot;, and direct mail). Azira's Advance DSP also documents six (desktop, mobile, CTV, audio, DOOH and Meta/TikTok social), so the two are tied for the widest documented channel set among the products opened. GroundTruth names no third-party DSP activation, only reporting integrations with TapClicks, NinjaCat and Agency Analytics. Foursquare, Adsquare, Reveal and Mobilewalla name external DSPs or marketplaces, and Precisely shows partner logos. Azira, Vistar and Simpli.fi activate through their own platforms. | 6 | synthesis | contradicted | medium | [122, 459, 530, 529, 335] |
| LOC-BLIS-F08 | Blis Global Ltd's advertising privacy notice (effective 17 November 2025) states that its location data includes precise GPS-sourced latitude/longitude and IP-derived geolocation, that it may build historic location profiles and link devices on residential IP addresses into households, and that such data reaches it from its own cookie or from data partners, SSPs, exchanges and clients; it deletes or anonymises personal data within 13 months. | 6 | vendor assertion | supported as stated | high | [235] |
| LOC-BLIS-F10 | Blis describes its audience targeting (Dynamic Audience Targeting) as AI trained on seed panels of 'consented personal data' supplied by clients or drawn from Blis's 'first-party real-world consumer behavior data', which then targets combinations of 'approximate geography, content, and context'; it translates census-tract and Vodafone quadkey data to postal districts for activation; how the movement data is collected is not stated. | 6 | vendor assertion | supported as stated | high | [229] |
| LOC-BLIS-F18 | Blis's documented activation routes for its audiences are its own DSP, offered managed or self-service on display, native, audio, online video, CTV and DOOH inventory, and, from 17 June 2025, PMG's Alli Marketplace; Blis says it does not mark up media and caps its data fees. | 6 | vendor assertion | supported as stated | medium | [229, 532, 533] |
| LOC-FOURSQUARE-F13 | Foursquare Audience builds segments from past visitation with location-specific high, medium and low frequency buckets, a look-back of up to 18 months (3 to 6 recommended), optional look-alike expansion and reach estimates from a random device sample; its minimum segment size is stated as 5,000 devices in the FAQ and 10,000 devices on the requirements page, and it is priced at US$1.50 CPM 'in most scenarios' against US$0.75 for Proximity. | 6 | vendor assertion | supported as stated | high | [528, 238] |
| LOC-FOURSQUARE-F14 | Foursquare's eligible-platforms documentation lists 26 activation platforms for Audience across DSPs, social platforms (Meta, LinkedIn, Pinterest, Reddit, Snapchat, TikTok), CTV (Roku/OneView, Samsung Ads), audio (Spotify, SoundCloud (Pandora)) and DOOH (Vistar Media), and the requirements page lists ready-to-use segments in 13 platforms; Vistar's partner page says over 2,300 Foursquare audiences are available in its DOOH DSP. | 6 | vendor assertion | supported as stated | medium | [237, 238, 416] |
| LOC-GROUNDTRUTH-F14 | GroundTruth's help centre names its location sources as owned-and-operated WeatherBug, direct app partnerships, SDK integrations and the bid-stream, with a single proportion ('90% of our location signals' from the first three) in an article whose body was last edited in April 2018; Vistar Media's partner page describes GroundTruth's audience data as including 'bid stream data'. | 6 | vendor assertion | supported as stated | medium | [260, 261, 191] |
| LOC-GROUNDTRUTH-F17 | GroundTruth's location guidelines (updated 28 September 2026) define three protected classes that change targeting, audience building and measurement: Consumer-Protected (1-mile minimum; no device-level collection), Healthcare-Protected (0.4-mile minimum for healthcare ads in WA, NY, CT, NV and CA; visits not tracked) and Prohibited (no location targeting; visits not measured), each with a published category list. | 6 | vendor assertion | supported as stated | high | [256] |
| LOC-GROUNDTRUTH-F19 | GroundTruth Ads Manager documents four targeting layers as of 23 September 2026: Target Locations (city to precise business address, lat/long or nationwide; 'Pure Geographic' delivery to anyone inside the boundary), Target by Audience (Predictive, Deterministic and Datonics 'Premium' tiers), Target Neighborhoods and Target On-Premise (mobile only; In Store, On Lot, Retail Block, Radial). | 6 | vendor assertion | supported as stated | high | [263] |
| LOC-GROUNDTRUTH-F20 | GroundTruth Ads Manager documents six channels (mobile, CTV and OTT, desktop and tablet, audio, DOOH with '1M+' US screens, and direct mail), a $0 minimum spend in its own comparison table, and buys inventory in real-time auctions across a stated 'over 100,000' publishers. | 6 | vendor assertion | supported as stated | medium | [122, 534] |
| LOC-GROUNDTRUTH-F21 | Outside GroundTruth's own self-serve and managed platform, the only named activation route for GroundTruth audiences found is Vistar Media's programmatic DOOH integration, documented on Vistar's partner page; GroundTruth's help centre mentions PMPs and unnamed DSP partners, and its named integrations (TapClicks, NinjaCat, Agency Analytics, a reporting API) carry reporting, not segments. | 6 | vendor assertion | supported as stated | medium | [191, 258, 122, 363] |
| LOC-GROUNDTRUTH-F22 | Since the ZeroToOne.AI acquisition, Ads Manager sells 'Predictive AI Audiences' from ZeroToOne.AI's model (called a 'Location Behavior Model' in the help centre and a 'Large Behavioral Model' in the acquirer's release), built from location visitation, interests, transactions, digital engagement and demographics with 24-48-hour to 14-30-day prediction windows, marketed at 'over 90%' (or '90%') predictive accuracy without a published definition or method; the catalogue grew from an 'initial' 40 segments (June 2026) to '1,500+' (October 2026). | 6 | vendor assertion | supported with qualification | high | [263, 264, 122] |
| LOC-INMARKET-F13 | At 30 June 2026 InMarket’s privacy policy says it does not sell ‘(in the plain language sense of that word)’ or license to customers personal information containing precise geolocation, while its state-law notice says that in the past 12 months it ‘sold (or shared for interest-based advertising)’ categories including coarse and precise geolocation to Third Party Partners, and it uses precise location to create audience segments and for attribution and measurement. | 6 | vendor assertion | supported as stated | high | [134] |
| LOC-INMARKET-F14 | InMarket documents sensitive-place handling only as company-level statements: its Ethics Principles say it does not use sensitive location visit data and suppresses and deletes data from sexual and reproductive healthcare providers, it describes a Power to Decide partnership for filtering those locations and NAI membership (a board seat on its home page), and its privacy policy retains precise location for 37 months from collection. | 6 | vendor assertion | supported as stated | high | [376, 535, 133, 134] |
| LOC-INMARKET-F16 | InMarket Audiences (also ‘Audience Suite’) is sold as 10,000+ syndicated segments, custom audiences, PMPs, GeoType personas built by machine learning on 1,000+ attributes, and, from May 2026 in Basis, RMN audiences and ‘Lift Optimized Audiences’ of known visitors, purchasers and look-alikes; activation is claimed on ‘any’ DSP with Basis the one external DSP route named in the sources opened. | 6 | vendor assertion | supported as stated | high | [268, 271] |
| LOC-VISTAR-F20 | Vistar's GroundTruth partner page offers, among example US segments, 'Visitors who visited general clinics at least 1 time in the past 120 days nationwide', and neither that page nor Vistar's Services Privacy Notice states a sensitive-place suppression rule. | 6 | synthesis | supported as stated | medium | [191, 339] |
| LOC-VISTAR-F21 | Vistar's TTD activation guide says buyers using The Trade Desk can apply audiences 'through Vistar or The Trade Desk': 'Vistar's entire taxonomy is available through both Auction Packages and PMPs' and 'Custom audiences must pass a feasibility check via Vistar's data team (4-5 business days)', while 'Third-party audiences must contain mobile device IDs' when applied in TTD. The Vistar DSP route is documented separately (solutions page; 5 October 2026 release). | 6 | vendor assertion | supported with qualification | medium | [337] |
| LOC-F-F14 | Kochava Collective's status at the cutoff cannot be confirmed from kochava.com: the /collective/ URL serves the App Measurement product page with no Collective content, and the FTC's 4 May 2026 release on the stipulated order says Collective Data Solutions 'has taken over Kochava's data broker business' and will be barred from selling, sharing or disclosing sensitive location data without affirmative express consent, with the consent exception limited to data 'used to provide a service directly requested by the consumer'. | 6, 7 | direct record | supported as stated | medium | [147, 144] |
| LOC-H-F05 | The Mobilewalla order (Docket C-4811, issued 13 January 2025, finalisation announced 14 January 2025 by 4-1 with Commissioner Holyoak dissenting) adds a ban on collecting or retaining data accessed in online advertising auctions for any purpose other than participating in them, bars use or sale of sensitive location data across ten categories with only a conversion exception (no direct-relationship consent exception), requires deletion of Historic Location Data within 90 days and Data Products within 120 days, and runs 20 years. | 6, 7 | direct record | supported as stated | high | [139, 140] |
| LOC-H-F21 | At the cutoff Mobilewalla documents continued licensing of device-level location-derived data: its business privacy notice (updated 9 January 2026) says it processes 'movement data' that includes MAID, latitude and longitude, IP address and timestamp and visits to commercial locations, that 'During the last year, Mobilewalla has sold personal information it has collected to our Clients and others, including ad tech partners', and that it does not knowingly collect visits to sensitive locations; its site sells Market Insights (mobility patterns), Feature Mart, Audience Segments, Data Enrichment and 'Location-Based Marketing', with no stated precision class for what is sold. | 6, 7 | vendor assertion | supported as stated | medium | [142, 141] |
| LOC-H-F22 | At the cutoff Kochava's data-broker business operates as Collective Data Solutions, LLC (Sandpoint, Idaho; a co-defendant under the 2026 order), whose site sells 'Data Feeds: License premium data including consumer characteristics, interest &amp; behavior, precision location (protected by Privacy Block)', Audience Targeting and Audience Enrichment, cites '&gt;500M Unique connected devices' and '16M+ Geofenced retail &amp; commercial locations', offers QSR and retail brands brick-and-mortar visitation and competitor conquesting and hedge funds consumer signals, and markets Privacy Block as blocking health clinics, religious institutions and schools; the former kochava.com/collective URL now serves an app-analytics 'Data Management' page. | 6, 7 | vendor assertion | supported as stated | high | [145, 146, 143, 147] |
| LOC-H-F23 | Under the preregistered rule H5 is contradicted. Three of the five ordered companies document at the cutoff the sale or licensing of precise geolocation to third parties: Outlogic, Collective Data Solutions for Kochava, and Gravy Analytics trading as Unacast, each in its own current policy or product pages. Mobilewalla documents licensing of device-linked lat/long data but not the precision of what it licenses, and InMarket documents no such product. Compliance with the orders is not assessed. | 6, 7 | synthesis | supported with qualification | high | [128, 137, 142, 145, 134, 130] |
| LOC-FOURSQUARE-F15 | Foursquare Proximity is a real-time radius geofence with no look-back, delivered to DSPs where 'your ad platform is able to validate that a user is within the radius boundaries of the geofence by matching the lat/long of the impression', and for national Trade Desk segments Foursquare requires a separate New York, Washington and Connecticut set with a minimum 3,000-foot (914 m) radius 'to follow state privacy laws'. | 6, 7 | vendor assertion | supported as stated | high | [536, 537, 237] |
| LOC-GROUNDTRUTH-F16 | GroundTruth documents that, for locations in Maryland (from 1 October 2025), Oregon (1 January 2026), Virginia (1 July 2026), New Jersey (8 July 2026) and Connecticut (1 October 2026), Visit Tracking, On-Premise and Neighborhood targeting, radius targeting below 1 mile and audience building from new precise location signals are unavailable, attributing the restriction to state laws on precise geolocation. | 6, 7 | vendor assertion | supported as stated | high | [256] |
| LOC-INMARKET-F11 | FTC Decision and Order C-4803, issued 29 April 2024 and finalised by a 3-0-2 vote announced 1 May 2024, bars InMarket from selling or licensing precise Location Data for any valuable consideration, and from products that categorise or target consumers on Sensitive Location Data tied to the sensitive locations its own programme identifies (nine categories, Definition I), with a carve-out for compliance use; it excludes coarse data at a radius of at least 1,850 feet and runs 20 years from issuance, extendable by a federal-court complaint alleging violation. | 6, 7, 8 | direct record | supported with qualification | high | [130, 132] |
| LOC-FOURSQUARE-F22 | Foursquare's consumer privacy policy (updated 20 February 2026, effective 22 March 2026) says it receives location from users who opt in to precise or background location in its apps, that SDK and API partners 'may license data to us', that Visits, Audiences, Proximity and Attribution are built from that data, and that as an NAI Precise Location Enhanced Standards participant it does not use or share activity correlated with 13 listed sensitive-place categories; NAI's signatory page (last updated March 2025) lists Cuebiq, Foursquare, InMarket and Outlogic. | 6, 8 | vendor assertion | supported as stated | high | [244, 243] |
| LOC-INMARKET-F17 | Location Sciences, which signed InMarket as the first (commercial) customer of its Verify Audience platform, audited several InMarket audiences in H1 2020 (department store, big box, QSR, auto, pet, amusement park) and rated them 'Optimal', counting a visit only inside store boundaries; the RNS says the deal followed 'paid for tests across multiple agencies and suppliers' in H1 2020, without saying who paid for InMarket's audits. Scope was announced as up to 1,000 audiences (certifier) or up to 500 segments (InMarket); no later audit, renewal or report was found, although InMarket's 2026 data page still says 'Location Sciences certified'. | 6, 8 | commissioned measurement | supported with qualification | medium | [378, 373, 267] |
| LOC-F-F17 | Broadsign Ads' DOOH audience targeting (found in discovery; not in the register) is the only DOOH audience product opened whose documentation names both the mobility supplier and the collection route: Native Touch data 'pulled from active application SDKs and APIs', collected 'from smartphones once audiences enter the defined location boundaries', including 'device IDs, languages, app IDs, age, gender', translated into 6 targeting groups and over 190 personas per screen; no panel size, consent route or region is stated. | 6, 9 | vendor assertion | supported as stated | medium | [538] |
| LOC-I-F16 | Broadsign's measurement page describes foot-traffic studies that 'track store visits after ad exposure', 'Device ID passback' that 'involves collecting anonymized device data from mobile devices that have been exposed to OOH ads', and case results including Decathlon ('218 digital screens'; 'a 46% total increase in store footfall, delivering an incremental 8 percentage point lift directly attributable to the campaign' against 'non-exposed control groups') and San Diego Tourism via Place Exchange ('a 49.8% net lift in physical visits to San Diego within 120 days'); measurement partners appear as logos only and no control-formation method, panel or provider is documented. | 6, 9 | vendor assertion | supported as stated | medium | [194] |
| LOC-GROUNDTRUTH-F23 | GroundTruth's DOOH product, launched on 9 October 2025 through Place Exchange, attributes visits by capturing 'mobile device IDs within viewing range of a screen when an ad plays' and reuses those IDs for retargeting; its Blueprints post says DOOH views are counted when a device enters a screen's mapped viewshed 'for the appropriate amount of time', with no time stated. | 6, 9 | vendor assertion | supported as stated | high | [257, 254] |
| LOC-VISTAR-F07 | Vistar-hosted partner pages name the data behind its US audience partners: GroundTruth 'first-party data, third-party partnerships, and bid stream data'; Adsquare 'spatial data, audience data and mobile location data'; Foursquare segments 'based on historical visitation patterns'; PlaceIQ 'location and purchase data'; InMarket '200M+ location-based users'; none gives proportions or a precision class. | 6, 9 | synthesis | supported as stated | medium | [191, 539, 416, 540, 23] |
| LOC-VISTAR-F22 | Vistar's audience guide states that DOOH targeting 'uses aggregated and anonymized location data to identify trends, without ever relying on personal information', while its Services Privacy Notice says the services receive device location histories associated with device identifiers and that Vistar sells and shares personal information. | 6, 9 | synthesis | supported as stated | medium | [338, 339] |
| LOC-D-F33 | Kochava Inc. is an Idaho-based company that the FTC sued in the District of Idaho on 29 August 2022 (describing a data sample covering more than 61 million unique mobile devices in one week and data 'derived from hundreds of millions of mobile devices'); in a release dated 4 May 2026 the FTC announced a proposed stipulated order, filed in the District of Idaho, against Kochava and its subsidiary Collective Data Solutions, 'which has taken over Kochava's data broker business', a move of the data business into a subsidiary. | 7 | direct record | supported with qualification | high | [541, 144] |
| LOC-H-F04 | The Gravy Analytics/Venntel order (Docket C-4810, issued 13 January 2025, finalisation announced 14 January 2025 by 5-0 after a proposed order announced in December 2024) bans selling, disclosing or using sensitive location data across eight categories (adding military installations) with a direct-relationship consent exception, excludes from 'Location Data' data used for Security Purposes, federal National Security purposes or federal law-enforcement emergencies, and runs 20 years; it does not ban sale of precise location data to commercial customers. | 7 | direct record | supported as stated | high | [135, 136] |
| LOC-H-F06 | FTC v. Kochava (D. Idaho, No. 2:22-cv-00377-BLW) was litigated for almost four years: complaint filed 29 August 2022; first motion to dismiss granted with leave to amend 4 May 2023 (Dkt. 24); First Amended Complaint 5 June 2023; motion to dismiss it denied 3 February 2024 (Dkt. 71); Kochava transferred its data-broker business to subsidiary Collective Data Solutions, LLC, so a Second Amended Complaint naming CDS was filed 15 July 2024 (Dkt. 86); motion to dismiss it denied 3 February 2025 (Dkt. 101); a stipulated order was announced 4 May 2026 (Commission vote 2-0) and entered 25 June 2026 (Dkt. 138). | 7 | direct record | supported as stated | high | [542, 543, 544, 148, 143, 144] |
| LOC-H-F07 | The Kochava order (entered 25 June 2026, 10-year term) requires defendants within 90 days to deidentify or render non-sensitive all Historical Location Data, unless they obtain within that period records showing consumers consented to its collection, use and disclosure. Its definition of Precise Location Data excludes coarse data (radius of at least 1,850 feet) and data used for security purposes, federal national-security purposes, or a law enforcement response to an imminent risk of death or serious bodily harm. | 7 | direct record | supported with qualification | high | [143] |
| LOC-H-F09 | The widely cited scale of the Gravy breach (&quot;several terabytes&quot;, &quot;historical location data of millions of smartphones&quot;) is the hacker's claim on a closed-access Russian-language forum, first reported by 404 Media; in the sources read, the only other figure is security researcher Baptiste Robert's statement that the copy of the leaked dataset he obtained held more than 30 million location data points (method unpublished); 404 Media found much of the location data was inferred from IP addresses, and TechCrunch wrote that it is unclear how Gravy derived its data. | 7 | synthesis | supported with qualification | medium | [545] |
| LOC-H-F15 | In the California Data Broker Registry file (603 registrants that operated as data brokers in 2025; file dated 29 July 2026, downloaded 24 October 2026), Collective Data Solutions, Outlogic, InMarket, Mobilewalla, Unacast, Inc. and Venntel each self-report collecting precise geolocation (115 of the 603 registrants do), and Unacast and Venntel report past-year sale or sharing to the federal government and other state governments. Kochava Inc. is not listed. These are the brokers' own declarations about collection in 2025, not evidence of what they sold at the cutoff. | 7 | direct record | supported with qualification | high | [251, 163] |
| LOC-H-F17 | On 13 January 2025 (date of the Attorney General's release; the posted petition is unstamped) the Texas Attorney General announced a suit in the District Court of Montgomery County, Texas against The Allstate Corporation, two Allstate insurers, Arity, LLC, Arity 875, LLC and Arity Services, LLC under the Texas Data Privacy and Security Act, the Texas Data Broker Law and the Insurance Code, alleging that Arity paid app developers to embed its SDK (the release names Life360) and collected location and driving-behaviour data from over 45 million consumers nationwide without notice or consent; the state calls it the first enforcement action by a state attorney general under a comprehensive privacy law. These are allegations; the case status at the cutoff was not checked. | 7 | direct record | supported with qualification | high | [546, 547] |
| LOC-H-F19 | On its home page and privacy policy (updated 30 June 2026), the only InMarket pages read, InMarket documents no product that sells or licenses precise location data to customers: the policy says it does &quot;not sell (in the plain language sense of that word) or license to customers any personal information ... that contains precise geolocation information&quot;, while its state-law disclosure says that in the past 12 months it &quot;sold (or shared for interest-based advertising)&quot; categories including &quot;coarse and precise geolocation information&quot; to Third Party Partners. | 7 | vendor assertion | supported with qualification | medium | [133, 134] |
| LOC-H-F20 | At the cutoff Gravy Analytics trades as Unacast (its privacy policy opens 'Gravy Analytics, Inc. dba Unacast'; gravyanalytics.com redirects to unacast.com) and documents continued sale of precise location data to third parties: it sells 'Location Data', 'Foot Traffic Data' and 'Data Linkages: connecting MAIDs to hashed emails (HEMs) and IPs', shares customers' 'device's location information, mobile advertising identifier, venues and events you attended', lists 'Precise Geolocation Data' among categories sold in the preceding twelve months, states customers may resell and names 'law enforcement and national security' among customer uses; its subsidiary Venntel sells Location Intelligence, Location Stream and an Intelligence Portal to national-security buyers. | 7 | vendor assertion | supported as stated | high | [137, 125, 138] |
| LOC-X-F01 | On 14 January 2026 the FTC announced its final order against General Motors and OnStar, which bans GM for five years from disclosing consumers' geolocation and driver-behaviour data to consumer reporting agencies; the order was proposed on 16 January 2025. It is a connected-vehicle case and lies outside the five location-data companies covered by hypothesis H5. | 7 | direct record | supported as stated | high | [149] |
| LOC-X-F02 | On 13 August 2024 Texas sued General Motors and OnStar under its deceptive trade practices law, alleging that GM collected and sold vehicle driving data, including GPS latitude and longitude, to companies that built driving scores for insurers without drivers' informed consent; GM's quarterly filing of 20 October 2026 says it is defending state enforcement suits, so the claims remain allegations. | 7 | direct record | supported as stated | high | [548, 549, 550] |
| LOC-X-F03 | Arkansas (26 February 2025, Phillips County Circuit Court), Nebraska (8 July 2025, Lancaster County District Court) and Iowa (26 February 2026, Polk County District Court) also sued General Motors and OnStar over the collection and sale of driving data. The Nebraska and Iowa releases name location among the data tracked; for Arkansas, location appears in the complaint (GM collected drivers' location data), not in the release. All three are allegations. | 7 | direct record | supported with qualification | high | [551, 552, 553] |
| LOC-X-F04 | On 8 May 2026 California's Attorney General, with four county district attorneys, announced a US$12.75 million settlement with General Motors over the sale of Californians' location and driving data to two data brokers (Verisk and LexisNexis Risk Solutions); the Napa County Superior Court entered the final judgment on 12 May 2026 (case 26CV001011). | 7 | direct record | supported with qualification | high | [554] |
| LOC-X-F05 | On 10 March 2025 California's Attorney General announced an investigative sweep of the location data industry, sending CCPA notice letters to advertising networks, mobile app providers and data brokers; it is an investigation, not a lawsuit. | 7 | direct record | supported as stated | high | [555] |
| LOC-X-F06 | State attorneys general have settled with Google over deceptive location tracking at least six times, with primary releases opened: Arizona US$85 million (4 October 2022), 40 states US$391.5 million (14 November 2022), Indiana US$20 million (29 December 2022), Washington US$39.9 million (18 May 2023), California US$93 million (announced 14 September 2023, subject to court approval at the time) and Texas US$1.375 billion (agreed in principle 9 May 2025, signed as announced 31 October 2025; the Texas sum also resolves incognito-browsing and biometric claims). A District of Columbia settlement (reported at about US$9.5 million, December 2022) was not opened at source. | 7 | direct record | contradicted | high | [556, 557, 558, 559] |
| LOC-X-F07 | On 4 April 2017 the Massachusetts Attorney General settled with Copley Advertising, a Massachusetts company that had geofenced reproductive-health and methadone clinics in five cities outside Massachusetts to serve ads promoting abortion alternatives; under an Assurance of Discontinuance filed in Suffolk Superior Court, Copley agreed not to use geofencing at or near Massachusetts healthcare facilities to infer anyone's health status. The release mentions no payment. | 7 | direct record | supported with qualification | high | [560] |
| LOC-X-F08 | Texas, Oregon and Vermont keep data broker registers alongside California's; on 24 October 2026 the Texas register's 456 records included Outlogic, Collective Data Solutions, Unacast, Inc. and Venntel but not Mobilewalla, InMarket or Kochava, and Vermont's revised statute, effective 1 January 2027, requires disclosure of precise-geolocation collection. Registration is a filing, not an assurance, and absence from a register does not show that a company is unregistered elsewhere. | 7 | direct record | supported as stated | medium | [561, 562, 563] |
| LOC-X-F09 | In the searches logged on 24 October 2026 (the researcher's 14 and the verifier's 8), one US state attorney-general lawsuit against an alleged location data broker was found for 2023 to 2026: Texas v. Allstate and Arity (filed 13 January 2025, Montgomery County), which pleads a Texas Data Broker Law count against Arity alongside TDPSA claims over location data collected through app SDKs, and which remains in litigation. No state attorney-general lawsuit against an advertising-technology company over location data was found. The other state actions found concern a carmaker (General Motors), a platform (Google, including Washington's May 2023 settlement) and a 2017 advertiser case, plus California's 2025 sweep. | 7 | synthesis | contradicted | medium | [555] |
| LOC-INMARKET-F12 | The order requires InMarket to delete Historic Location Data collected through its own apps within 90 days of the effective date, to delete, deidentify or render non-sensitive Historic Location Data collected from third parties within 120 days, to delete audience segments built from Historic Location Data within 120 days, and to run a Sensitive Location Data Program (within 90 days of issuance) and an SDK Supplier Assessment Program confirming consent for SDK-supplied data. | 7 | direct record | supported as stated | high | [130] |
| LOC-PLACER-F14 | Placer documents sensitive-place suppression at product level: it says it does not knowingly sell or provide data on military facilities, PreK-12 schools, places of worship, rehabilitation centres and women's health facilities, and it names Washington's My Health My Data Act and Nevada SB370 as reasons it withholds property, chain and industry reports for in-person health-care locations in those states. | 7 | vendor assertion | supported as stated | high | [286] |
| LOC-UNACAST-F05 | FTC Decision and Order C-4810 against Gravy Analytics, Inc. and Venntel, Inc., issued 13 January 2025 after a complaint and proposed order announced on 3 December 2024 and a 5-0 final vote announced on 14 January 2025, bans selling, sharing or using Sensitive Location Data tied to the sensitive locations the companies must identify (eight categories, with narrow exceptions), bars collecting or using any Location Data without a consent record from 90 days after the effective date, requires a Supplier Assessment Program, and requires deletion of Historic Location Data within 60 days unless, within 90 days, consent records are obtained or the data are deidentified or rendered non-sensitive; it runs 20 years from issuance (or from any later federal-court complaint alleging a violation); respondents neither admit nor deny the allegations. | 7 | direct record | supported with qualification | high | [135, 428, 136] |
| LOC-UNACAST-F07 | On 24 October 2026 Unacast markets &quot;Privacy-friendly device-level location data for areas of interest globally&quot; from &quot;more than 1 billion monthly devices across 180+ countries from 15+ suppliers&quot;, and its pricing page offers &quot;Access to granular, deduplicated, and cleaned device-level data&quot; for the &quot;US &amp; International&quot;. | 7 | vendor assertion | supported as stated | high | [328, 332] |
| LOC-UNACAST-F10 | Gravy Analytics told Norway's Datatilsynet it was informed on 4 January 2025, by the intruder, of unauthorised access to its AWS storage apparently through a misappropriated access key, that &quot;some files&quot; were obtained and processing was operational again by 9 January 2025; Unacast's Texas registration, submitted 9 January 2026, adds that a small subset of copied data, &quot;covering primarily a few days around New Year's 2025&quot;, was briefly posted on a dark web forum. | 7 | vendor assertion | supported as stated | high | [333, 278] |
| LOC-UNACAST-F11 | In the California Data Broker Registry download of 24 October 2026, Unacast, Inc. and Venntel, Inc. (both Ashburn, Virginia) each answer Yes to collecting precise geolocation and to sharing or selling data to the federal government and to other state governments in the past year, and No to sharing with law enforcement except under subpoena or court order; Unacast reports 561 delete requests, all complied with in whole. | 7 | direct record | supported as stated | high | [251] |
| LOC-UNACAST-F21 | Unacast's sensitive-location controls are documented only as assertions: PrivacyCheck &quot;ensures that data generated by consumer mobile devices while visiting sensitive locations is never used, shared, or resold&quot; and &quot;All products remove data from sensitive locations&quot;, with no category list on the pages read, while the FTC order defines eight categories of Sensitive Locations. | 7 | synthesis | supported as stated | medium | [327, 332, 135] |
| LOC-A-F20 | NAI's Precise Location Enhanced Standards were first issued on 22 June 2022 with Cuebiq, Foursquare and Precisely PlaceIQ as signatories. They were reissued 'Updated and effective October 2024', and the PDF's signatory page, 'Last updated March 2025', shows four logos: Cuebiq, Foursquare, InMarket and Outlogic. Within US activities they bar using or transferring activity correlated to a known Sensitive POI (11 definitive and 2 non-definitive categories) and law-enforcement, national-security or bounty-hunting uses, subject to section IV's exceptions for consented direct-relationship services, blocking directories, technical providers and compliance processing. | 7, 8 | direct record | supported with qualification | high | [564, 565, 243] |
| LOC-H-F01 | The FTC's X-Mode Social/Outlogic order (Docket C-4802, issued 11 April 2024 after a proposed settlement announced in January 2024; final approval 3-0-2 announced 12 April 2024) bars selling, licensing, transferring, sharing, disclosing or using sensitive location data, with a direct-relationship consent exception, and runs 20 years; it does not ban sale of precise location data as such. | 7, 8 | direct record | supported as stated | high | [126, 127] |
| LOC-H-F18 | At the cutoff Outlogic (X-Mode's successor) documents continued sale of precise location data to third parties: its site says the company may 'receive and sell the precise latitude and longitude coordinates of your mobile device', offers data 'to License Via Data Exchanges, Platform Partnerships, or directly to you' with observation and attribution delivery through Snowflake, AWS S3 and a REST API and audiences through Eyeota and Adsquare, claims 'No Bidstream Data', and states in its January 2026 privacy notice that it collects and sells precise geolocation while filtering sensitive health-based locations in specific states. | 7, 8 | vendor assertion | supported as stated | high | [128, 129] |
| LOC-H-F16 | Virginia bans the sale of precise geolocation data: Code of Virginia 59.1-578(A)(6), added by 2026 Acts ch. 820 (SB 338, approved 13 April 2026, effective 1 July 2026), requires a controller to &quot;not sell or offer for sale precise geolocation data concerning a consumer&quot;, with no consent exception; other processing of sensitive data, including precise geolocation, still requires consent under 59.1-578(A)(5). | 7, 12 | direct record | contradicted | medium | [162, 157] |
| LOC-R3-F01 | New Jersey P.L.2026, c.25 (A5328 1R), approved and effective 30 June 2026, amends N.J.S.A. 56:8-166.12(a)(6) so that no controller may sell sensitive data regardless of how many consumers' data it controls or processes, and separately bars any data broker or data collector from selling or licensing sensitive data; sensitive data includes precise geolocation data defined by a 1,750-foot radius, with no consent exception, making it a flat ban on sale of precise geolocation alongside Maryland's. | 7, 12 | direct record | supported as stated | high | [156] |
| LOC-A-F02 | The 2017 guidelines cover the assignment of device and user location for physical visitation and ad exposure on mobile and static devices (desktop, connected TV, OTT), place measurement as the reference for visitation and location targeting, and a benchmark for voluntary third-party audit; they are guidance and a benchmark, not a certification. | 8 | direct record | supported as stated | high | [445] |
| LOC-A-F06 | The guidelines require a fully disclosed universe for projections and, where vendors set minimum reporting thresholds, disclosure of them; they say vendors should disclose the share of each assignment method where methods are mixed, whether assignment is probabilistic or deterministic, editing volumes above about 5% of records, and should retain detailed data for at least one year. | 8 | direct record | supported with qualification | high | [445] |
| LOC-A-F08 | The guidelines recommend, but do not require, annual third-party certification of every digital location-based measurement product used in buying and selling, and the document itself notes that at the time some US location products 'have been accredited by the MRC in some fashion'; conformance to the guidelines ('built to MRC guidelines') and MRC accreditation are therefore distinct states, and only the latter appears on MRC's lists. | 8 | direct record | supported as stated | high | [445, 566] |
| LOC-A-F09 | The guidelines require clear opt-in for location collection, permission prompts on mobile web and in apps (re-requested when the user changes the device setting), and a privacy policy stating why location is collected and how it is shared, and point audited services to the IAB 2016 publisher guide, the DAA principles and the NAI Code. | 8 | direct record | supported as stated | high | [445] |
| LOC-A-F10 | On 24 October 2026, none of MRC's six accredited-services lists and none of its five under-review lists carries a visit, visitation, foot-traffic or location-determination metric. The closest under-review entries are Kevel's 'Conversions and incrementality' and Kroger's attributed and incremental sales. MRC's April–June and July–September 2026 communications record no location or visit action, and the Out-of-Home and Outside-the-U.S. accredited lists are empty. | 8 | direct record | supported with qualification | high | [171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 567, 568] |
| LOC-A-F11 | Foursquare's Visits data set was granted MRC accreditation in July 2020 for 'location information reported in Visits data set', suspended in October 2022 and revoked in December 2022 (MRC communications of 13 October 2020 and 24 January 2023); Foursquare said its Attribution product was not audited. MRC's own September 2018 presentation already reported one accredited location service among six applicants without naming it, so the 2020 grant is not established as the first location accreditation. | 8 | direct record | supported with qualification | high | [166, 167, 521, 569, 164] |
| LOC-A-F12 | GroundTruth's 'Location, Place, and Visit data set' was granted MRC accreditation in July 2021 and placed on a six-month hiatus at GroundTruth's request in September 2023. MRC's communication of 26 July 2024 reports its accreditation revoked following withdrawal from the accreditation process; the row is labelled February inside the April-June report, so the effective month is unconfirmed. GroundTruth's 2021 release claiming the accreditation is still online. | 8 | direct record | supported with qualification | high | [168, 169, 170, 356] |
| LOC-A-F15 | Epsilon holds the only accredited 'outcomes' metric on MRC's Digital list, 'Correlated Outcomes (online and offline)' in desktop, mobile web and mobile in-app, continued by letter of 1 October 2025; the letter cites the Minimum Standards, IVT 2.0, the Data Integration Guidelines and the IAB and MRC impression guidelines, not the Location-Based Advertising Measurement Guidelines, and does not define 'offline' outcomes as store visits, so the entry does not satisfy H6. | 8 | direct record | supported as stated | high | [570] |
| LOC-A-F22 | IAB's location-data guidance found by the cutoff runs from the gated 'Mobile Location Data Buyer's Guide' (July 2014) and the February 2016 'Mobile Location Data Guide for Publishers' to the April 2017 'Location-based Marketing Playbook for Retailers'. The playbook is buyer-facing and asks whether device data come 'through panels or pass through ad exchanges' and how store-visit-lift control groups are selected. It cites 'IAB Marketing ROI and Location Data' (February 2015) and '12 Questions Every Buyer Should Ask about Location Data'. No IAB or IAB Tech Lab location guide later than April 2017 was found in the sources screened. | 8 | direct record | contradicted | medium | [462, 571] |
| LOC-BLIS-F14 | On 24 October 2026 Blis does not appear on MRC's Accredited Services: Digital list, its Services Under Review: Digital list or its accreditation landing page, so no Blis visit, lift or input metric is MRC-accredited or under MRC review in the sources screened. | 8 | direct record | supported as stated | high | [171] |
| LOC-FOURSQUARE-F05 | On 24 October 2026 MRC's Accredited Services: Digital list and its Services Under Review: Digital list contain no entry for Foursquare or Placed, and the Out-of-Home accredited and under-review pages returned no readable list of services; this review found no current MRC accreditation or audit for any Foursquare product in the MRC sources screened. | 8 | direct record | supported as stated | medium | [171, 173] |
| LOC-GROUNDTRUTH-F05 | MRC granted accreditation to 'GroundTruth's Location, Place, and Visit data set' in July 2021, as recorded in its communication of 20 October 2021; the scope is a data set, not a visit-lift or attribution metric. | 8 | direct record | supported as stated | high | [168] |
| LOC-GROUNDTRUTH-F06 | MRC granted GroundTruth a six-month accreditation hiatus at its own request in September 2023, during which the service was not accredited, and later recorded the accreditation as revoked after the hiatus 'due to withdrawal from the MRC accreditation process' in a row labelled 'February' in its April-June 2024 communication dated 26 July 2024; the October-December 2023 and January-March 2024 communications carry no GroundTruth row, so the effective month is unconfirmed. | 8 | direct record | supported as stated | high | [169, 170, 368, 369] |
| LOC-GROUNDTRUTH-F07 | At 24 October 2026 GroundTruth appears on neither MRC's Accredited Services: Digital list nor its Services Under Review: Digital list, so no GroundTruth visit or visit-lift metric holds or is in process for MRC accreditation at the cutoff. | 8 | direct record | supported as stated | high | [171, 177] |
| LOC-GROUNDTRUTH-F08 | GroundTruth's own pages still present the MRC accreditation as current: its 3 August 2021 release remains online (page modified 4 February 2026) with no hiatus or revocation notice, and a comparison page last modified 9 November 2023, after the September 2023 hiatus began, states in present tense that GroundTruth is 'Accredited by the Media Rating Council (MRC) for our Place, Location, and Visitation datasets' and that 'The accreditation validates our accuracy of place-based visits.' | 8 | synthesis | supported with qualification | high | [356, 358, 169] |
| LOC-INMARKET-F10 | On 24 October 2026 neither MRC’s Accredited Services: Digital list nor its Services Under Review: Digital list carries InMarket, NinthDecimal or any InMarket visit or lift metric; InMarket’s assurance claims (Comscore verification, Location Sciences certification, a Forrester TEI) are not accreditations. | 8 | direct record | supported as stated | high | [171, 177, 267] |
| LOC-PLACER-F18 | No MRC accreditation, audit in progress or independent data-quality certification of Placer's visit metrics was found: Placer does not appear on MRC's accredited or under-review digital and out-of-home lists on 24 October 2026, and its documented certifications are information-security ones (ISO/IEC 27001:2022, SOC 2 Type II as stated by Placer, and a CSA STAR Level 1 self-assessment listed since 21 July 2026). | 8 | direct record | supported as stated | high | [171, 177, 386, 292] |
| LOC-UNACAST-F20 | No Unacast or Gravy product appears on MRC's Accredited Services: Digital or Services Under Review: Digital lists on 24 October 2026; Unacast's privacy policy cites NAI membership &quot;in good standing&quot; with &quot;annual reviews&quot;, a self-regulatory status with no public product-scoped result. | 8 | direct record | supported as stated | high | [171, 177, 137, 414] |
| LOC-A-F23 | IAB's 'Digital Out-Of-Home (DOOH) Measurement Guide' was published in July 2025 (8 July per IAB's blog); IAB's page gates it, but the full PDF is publicly retrievable and was read in the final live sweep. It covers OTS and LTS impression types and dwell time, recommends synthetic-control and matched-market or matched-store tests for incrementality, and describes geo holdouts as a way to reduce control-group contamination (pp. 23-24). | 8, 9 | direct record | supported with qualification | medium | [572, 573, 574, 575] |
| LOC-A-F24 | MRC's Out-of-Home Measurement Standards, Phase 1 &amp; 2 Combined, Final December 2025 (developed with DPAA, Geopath and OAAA after public comment) treat location data as a foundational element where persons are not measured directly, require measurement services to ascertain the accuracy of third-party data with disclosure when a source cannot be independently verified, name 'mobile location' among deterministic inputs for adjusting likelihood-to-see to audience, and in their attribution section require a measurement organisation to collect at least a portion of characteristics directly; the 2017 Digital Place-Based Audience Measurement Standards (Version 1.0 Final, 15 March 2017) remain listed alongside them. | 8, 9 | direct record | supported as stated | high | [576, 577] |
| LOC-A-F25 | At 24 October 2026 MRC's Out-of-Home accredited list is empty, the only OOH service under review is JCDecaux's Proof of Plays metric, and Geopath appears on no MRC list opened; the US out-of-home audience currency is therefore not MRC-accredited at the cutoff in the sources screened. | 8, 9 | direct record | supported as stated | high | [173, 178, 171] |
| LOC-A-F27 | Geopath's methodology overview states that it is not designed to provide a comprehensive statistical validation of Geopath data and reports no confidence intervals, referring such requests to its data science team. | 8, 9 | vendor assertion | supported as stated | medium | [578] |
| LOC-I-F07 | Geopath releases its audience estimates as annual planning datasets: the 2026 planning dataset went into preview on 30 September 2026 and, per Geopath's 19 October 2026 post, becomes the default transactional planning dataset on 12 November 2026 (a plan at the cutoff); that later post names Bentley Systems as Geopath's roadside data provider and says the 2026 dataset uses 'updated data sources' without listing them. | 8, 9 | direct record | supported with qualification | high | [579, 580] |
| LOC-I-F08 | Geopath's own 'Our Organization' page states it was 'Established in 1934' as 'the Traffic Audit Bureau for Media Measurement Inc.' and 'is a not-for-profit organization governed by a tripartite board' 'comprised of advertisers, agencies, and media companies'; the 1934 founding year conflicts with the 1933 used in the brief and in the author's essay. | 8, 9 | direct record | supported as stated | medium | [581] |
| LOC-VISTAR-F15 | On 24 October 2026 Vistar had no entry on MRC's Accredited Services: Digital or Services Under Review: Digital lists, and MRC's Out-of-Home accredited and under-review pages returned no listed services, so no Vistar visit, lift, audience or impression metric was found to hold or be under MRC accreditation in the pages screened. | 8, 9 | direct record | supported as stated | medium | [171, 177, 173, 178] |
| LOC-J-F11 | The non-agentic baseline is further along than any agentic protocol: OpenRTB 2.6 defines a Geo object (lat, lon, type, accuracy, lastfix, ipservice, country, region, metro, city, zip, utcoffset) plus DOOH venue types and a qty impression multiplier, and Google Authorized Buyers documents production bid requests in which geo type is always set to IP (2) and accuracy reflects coarse IP-based geolocation, i.e. documented production use of coarse, inferred location rather than device-observed location. | 8, 10 | direct record | supported as stated | high | [219, 220] |
| LOC-A-F26 | Geopath's public methodology documentation consists of Confluence pages: a 'Media Measurement Methodology Overview' (published 31 May 2023) naming Streetlytics and HERE road data and a mobile device panel; an 'OOH Measurement Methodology' index (last updated 11 August 2025) listing a 'Geopath Visibility Model 2021.11' and monthly release schedules for 2025 and 2026; a 'Mobile Device Location &amp; Movement Patterns' page (31 May 2023) stating that source data from multiple location SDK aggregators covers more than 100 million unique US devices a month, that Intermx culls it to a sample of seven to ten million qualified devices a week, that records carry device ID, UTC timestamp, coordinates and horizontal precision in metres, with imprecise sightings removed under no stated numeric threshold; and a 'Movement Patterns' page (31 May 2023) recording AirSage-supplied origin-destination matrices for the 2019-2020 methodology and a 2021 expansion contracted to Standard Data. | 9 | vendor assertion | supported as stated | medium | [578, 582, 185, 583] |
| LOC-I-F01 | Geopath's methodology documentation, published 31 May 2023, names MotionWorks device panels, Teralytics/Streetlytics and HERE Maps for roadways, HERE Maps for vehicle travel, and government agencies, MotionWorks and Claritas for demographics, with no proportions. That list is as documented in 2023: Geopath's 19 October 2026 post names Bentley Systems as its roadside data provider, and its 2026 Planning Dataset, which credits 'updated data sources' for pedestrian-impression changes, becomes the default only on 12 November 2026, after the cutoff. | 9 | direct record | supported with qualification | high | [182, 578] |
| LOC-I-F02 | Geopath's Mobile Device Panel page (31 May 2023) describes a weekly panel of GPS-sighted devices, active about eight hours a day with a consistent Census Block Group home, culled by Standard Data, and illustrates it with a 2020 example (19.5% of devices, 78% of locations). A companion 31 May 2023 page says the feeds cover 'more than 100 million unique devices in the United States every month' and that the sample is 'between seven and ten million qualified devices each week'. That is a documentation-dated disclosure with no stated observation vintage, not a current audited panel count. | 9 | direct record | supported with qualification | high | [183] |
| LOC-I-F03 | Geopath's core outputs are 'likelihood-to-see impressions', audience composition and reach and frequency; opportunity-to-see counts are adjusted by its proprietary 'visual acuity index' (VAI) model for the probability of noticing the unit, audience composition is assigned by treating a device's longest-dwell location as its home and applying census-block demographics, and composition is computed as an annual average broken down by hour of day. | 9 | direct record | supported as stated | high | [584, 182] |
| LOC-I-F06 | Between the 31 March 2026 release and the cutoff, this review found no Geopath or Ipsos document describing the pilot's data sources, panel or method in the sources screened: Ipsos's OOH Media Con session page (12 May 2026) refers to 'the groundbreaking OAAA/Geopath/Ipsos pilot' without method, and Geopath's homepage news list and its 30 September 2026 dataset post do not mention the pilot. | 9 | inference | supported as stated | low | [585, 579] |
| LOC-I-F09 | OpenRTB 2.6's Qty object carries 'multiplier' (float; required: the quantity of billable events deemed to occur, e.g. 'a DOOH opportunity may be considered to be 14.2 impressions'), 'sourcetype' (integer; recommended) and 'vendor' (string, 'required if sourcetype is present and type = 1': 'The top level business domain name of the measurement vendor providing the quantity measurement'); AdCOM's 'DOOH Multiplier Measurement Source Types' list enumerates 0 Unknown, 1 Measurement Vendor Provided, 2 Publisher Provided and 3 Exchange Provided; neither document records the vendor's mobility source, panel or method, and neither enumerates multiplier vendors. | 9 | direct record | supported as stated | high | [586, 587] |
| LOC-I-F10 | WOO's PwC-aggregated study, announced 5 June 2026, puts global programmatic DOOH at 'approximately $1.4 billion in revenue globally in 2025 — equivalent to 7% of total DOOH expenditure' ('$1.339 billion in 2025' in a quoted figure), from data submitted by 11 participating SSPs (the page lists 12 names), with the Americas at $670m and 14.2% penetration, EMEA $521m and 9.4%, APAC $149m and 1.7%; PwC 'has not independently verified the information which it received from respondents', and the announcement says nothing about audience or mobility data, so it does not bear on mobility-based buying. | 9 | commissioned measurement | supported as stated | medium | [588] |
| LOC-I-F11 | The IAB published its DOOH Measurement Guide on 8 July 2025; IAB's page gates it behind a free account, but the 31-page PDF is publicly retrievable and was read in the final live sweep (24 October 2026). It recommends two methods for incrementality, synthetic control and matched-market or matched-store testing (p. 24), says that blending synthetic models with geo-based holdouts can reduce contamination of the control group, and lists control groups and A/B testing with geo holdouts as baselines for foot-traffic lift (p. 23). | 9 | direct record | supported with qualification | medium | [572, 573, 575] |
| LOC-I-F12 | Vistar Media's partner page for InMarket documents DOOH foot-traffic and sales attribution 'Powered by SDK-based location data from nearly 200M+ permission-based users' and 'transaction insights from over $2.5T+ anonymized commerce transactions' (the same page also says '~$122B+ verified transactions'), through a process in which 'The target audience is first exposed to Vistar advertising campaigns', 'InMarket then identifies the exposed audience and qualifies users who visited a retail location and/or completed a purchase' and 'LCI® quantifies the incremental lift in both visits and sales'; no control-group design, panel period or definition of 'users' is given. | 9 | vendor assertion | supported as stated | medium | [23] |
| LOC-I-F13 | On 11 February 2020 Vistar Media announced Foursquare as a DOOH measurement partner, combining Vistar's exposure data from 'direct integrations with more than 70 media owner networks' with 'Foursquare's location data' and the 'Placed powered by Foursquare attribution methodology' so that 'brands can analyze lift in foot traffic that was driven by DOOH media'; the release describes no control design or panel size. | 9 | vendor assertion | supported as stated | medium | [187, 189] |
| LOC-I-F14 | Vistar Media's June 2023 partnership with Adsquare gives Vistar 'SDK-derived location data from Adsquare' to score poster locations, starting in EMEA and extending to the US, Canada and Australia. Vistar's partner directory lists 37 partners with role tags (audience segments, point-of-interest data, foot traffic, sales lift, device-ID passback) and region tags, and Vistar-hosted partner pages state the data behind several US audience partners: GroundTruth (first-party, third-party and bid-stream data), Adsquare (mobile location data), Foursquare (visitation-based segments), PlaceIQ (location-based segments) and T-Mobile (app-ownership data from devices on its network). None of these pages gives a panel size or period. | 9 | vendor assertion | supported with qualification | medium | [188, 189, 190] |
| LOC-I-F15 | Place Exchange's 2026 sales presentation (now 'part of the Broadsign family') says 'PerView uses daily updated mobile data to calculate up-to-date reach, frequency, and impression metrics for any set of OOH assets', that its 'patented process of capturing exposed mobile IDs for consumers in the viewshed of every screen at the time of ad play' supplies exposure data for targeting and measurement, that it has 'preconfigured 15+ industry-leading measurement partners enabling server-to-server data', and that buyers can 'Plan against over 5,000 actionable data segments'; no mobile-data provider, panel size or partner name is given. | 9 | vendor assertion | supported as stated | medium | [192, 193] |
| LOC-I-F17 | GroundTruth's DOOH attribution, announced on 9 October 2025 with Place Exchange, 'captures mobile device IDs within viewing range of a screen when an ad plays' and checks whether those devices later visit a store, using GroundTruth's own 'location and purchase data'; the release cites an omnichannel (DOOH plus mobile) belVita campaign credited with 221,000 Target store visits in one month at a 3.44% visitation rate, and a separate Place Exchange campaign for an unnamed national retailer with over 42,000 visits from exposed consumers in New York, Chicago and Los Angeles, 'more than half' called incremental with no control design described; captured IDs are reused for retargeting. | 9 | vendor assertion | contradicted | medium | [257] |
| LOC-I-F18 | Hivestack's footfall lift study for Zoom Media UK x Gillette (London campaign, 1-30 November 2022; 115 screens; 4,001,001 delivered impressions) defines the exposed group as 'MAIDs observed within a screen geofence when advertiser ad was playing' and the control group as 'MAIDs observed within a screen geofence when advertiser ad was not playing', reports a '2.12x lift' in exposed-versus-control visitation, 8,574 'Modeled incremental walk-ins' and a cost per incremental visit of £2.28, and scales walk-ins so that 'Modeled walk-in is multiplied based on ratio of data partner access to phones vs total GB population'; the data partner is unnamed and the control is a non-randomised time-split comparison, not a holdout. | 9 | vendor assertion | supported as stated | medium | [196] |
| LOC-I-F19 | Perion's DOOH page (2026 copyright; hivestack.com redirects to perion.com) says 'Perion connects campaign playlogs with anonymised device-level location data' and offers footfall lift 'comparing exposed audiences with a matched control group' without naming a data provider or matching method; an undated Perion blog says platforms 'like Perion' have pre-integrated partnerships with third-party footfall providers it names (Foursquare, Cuebiq, PlaceIQ, Adsquare) and advises partners to use matched-market or propensity-score matching, without saying which provider or method Perion's own product uses. | 9 | vendor assertion | supported with qualification | medium | [195, 589, 590] |
| LOC-I-F21 | Route, the Great Britain out-of-home Joint Industry Currency for adults 15+, combines passive GPS tracking of a recruited, consented sample with volumetric counts and Lumen eye-tracking data; its People page discloses c.7,000 participants a year and 25,420 people's journeys collected September 2018 to February 2020 and September 2022 to September 2024, recruited and processed by Ipsos; it does not use a third-party smartphone SDK panel. | 9 | direct record | supported with qualification | medium | [591, 592] |
| LOC-I-F22 | COMMB describes itself as 'the national not-for-profit organization for the Canadian out-of-home (OOH) industry' that is 'developing proprietary audience measurement methodologies for a variety of OOH media formats', with a Research Committee that 'approves all COMMB methodologies for outdoor, place-based, moving media etc.'; its public site is client-rendered and no methodology page with data sources could be read by the cutoff, so COMMB's mobility sources are coded 'not disclosed' in this review. | 9 | direct record | supported as stated | low | [593] |
| LOC-VISTAR-F03 | Lamar Advertising received US$115.1 million in cash at the 3 February 2025 closing for its 20% equity interest in Vistar Media, with up to a further US$15.1 million payable on release from escrow (8-K Exhibit 99.1, 20 February 2025). | 9 | direct record | supported as stated | high | [37] |
| LOC-VISTAR-F04 | At the 13 January 2025 announcement, Vistar was to join T-Mobile Advertising Solutions at about US$600 million in cash, and the parties described a global network of more than 1.1 million digital screens from nearly 370 OOH media owners serving more than 3,000 advertisers. | 9 | vendor assertion | supported as stated | medium | [336] |
| LOC-VISTAR-F08 | Vistar's partner directory (37 cards, read 24 October 2026) tags 18 partners for audience segments in the USA and six USA-tagged foot-traffic measurement partners (Accretive, Arrivalist, Cuebiq, Foursquare, InMarket, Spotzi), while Lifesight is tagged only for Canada and APAC and Locala only for Canada. | 9 | synthesis | supported as stated | medium | [189] |
| LOC-VISTAR-F12 | Vistar's InMarket partner page markets LCI for DOOH as measuring 'incremental lift in both visits and sales', 'Powered by SDK-based location data from nearly 200M+ permission-based users', with no control design, period, geography or definition of users, and the same page gives two different transaction figures (~$122B+ and $2.5T+). | 9 | vendor assertion | supported as stated | medium | [23] |
| LOC-VISTAR-F13 | Vistar has sold DOOH foot-traffic studies through rotating third parties since at least January 2019 (a suite with NinthDecimal, MFour, Cuebiq and PlaceIQ, with Factual also in the subhead) and added Foursquare on 11 February 2020, when it also claimed 'in-house' foot-traffic measurement. | 9 | vendor assertion | supported as stated | medium | [594, 187] |
| LOC-VISTAR-F14 | Audience impressions reported in The Trade Desk for Vistar DOOH are, per Vistar's TTD guide, 'the result of impression multiplier calculations provided by SSPs &amp; media owners (these are commonly sourced from verification partners like Route and Geopath)', while plain impressions count ad plays; the 2022 case says exposure for studies uses 'ad log timestamps, device location, venue type and screen location'. | 9 | vendor assertion | supported with qualification | medium | [337, 8] |
| LOC-VISTAR-F16 | Vistar's Services Privacy Notice (last updated 1 July 2026) says it may receive device location histories tied to device identifiers from clients, data suppliers and measurement partners, processes those histories as a CCPA 'business' while providing advertisement measurement as a 'service provider', does 'sell' and 'share' personal information, uses the data to learn whether exposed devices are more likely to visit locations, discloses device location histories to service providers, and does not share them with clients, advertising intermediaries or market research companies. | 9 | vendor assertion | supported with qualification | high | [339] |
| LOC-VISTAR-F17 | Vistar Media Inc. is on California's 2026 Data Broker Registry (2025 activities; last updated 29 July 2026) and declared 'Yes' to collecting consumers' precise geolocation, 'No' to reproductive health care data and 'No' to sharing or selling data to foreign actors, federal or state governments or law enforcement. In the 2025 registry file it is listed as Vistar Media, Inc. and declared 'No' to collecting precise geolocation; it was not found in the 2024 file. | 9 | direct record | supported with qualification | high | [163, 595, 596] |
| LOC-VISTAR-F18 | The FCC's Forfeiture Order against T-Mobile USA (adopted 17 April 2024, released 29 April 2024) imposed US$80,080,000 for failing to take reasonable steps to protect customers' location information; the D.C. Circuit denied review on 15 August 2025 and rehearing on 23 January 2026, and the carriers' certiorari petition (No. 25-1422) was distributed on 14 October 2026 for the 6 November 2026 conference. | 9 | direct record | supported as stated | high | [340, 341] |
| LOC-VISTAR-F23 | Vistar's 23 September 2026 'store locator effect' study, run by its Creative Studio with Omnicom Media and Annalect, randomised 7,200 consumers aged 18-59 in the Netherlands (February to April 2026) to a control creative or one of three store-locator placements and reports stated visit and purchase intent (for example 9% higher visit intent versus control for bottom placement in retail and QSR); it measures intent, not visits, and is not US. | 9 | vendor assertion | supported as stated | medium | [419] |
| LOC-J-F02 | AdCP v3.2.6 represents DOOH delivery and visit outcomes in its delivery-metrics schema: a top-level plays count, a dooh_metrics block (loop_plays, screens_used, screen_time_seconds, sov_achieved, calculation_notes and a per-venue venue_breakdown with venue_id, venue_type, impressions, loop_plays and screens_used), a measurement_source slug whose examples include geopath and route, an experimental ooh_metrics block whose estimated_impressions carry a currency_measured or seller_modeled tier, and a foot_traffic metric defined as incremental visits over baseline that is qualified by attribution_methodology (panel_based, deterministic_purchase, probabilistic, modeled) but carries no field declaring the control design. | 9, 10 | direct record | supported as stated | high | [211, 212] |
| LOC-J-F03 | AdCP v3.2.6 represents venue identifiers for DOOH: placement identifiers accept the types venue_id, screen_id and openooh_venue_type and authority-prefixed external IDs (the schema's example is geopath:30961), DOOH placements declare slot and loop durations, and the experimental static-OOH panel identifiers accept a geopath Spot ID scheme; the sales_dooh compliance specialism tests a seller exposing a venue_id and reporting plays and dooh_metrics. | 9, 10 | direct record | supported as stated | high | [208, 211, 209, 597] |
| LOC-I-F24 | Of six DOOH footfall-attribution products or cases screened (four US, one UK, one global), none documents a randomised or holdout control: Hivestack's 2022 UK study compares passers-by at the same screens when the ad was and was not playing; Perion asserts 'a matched control group' without method; Broadsign cites 'non-exposed control groups' without method; Vistar's InMarket and Foursquare pages and GroundTruth's Place Exchange release describe no control (GroundTruth calls 'more than half' of a national retailer's 42,000 exposed-consumer visits incremental without a comparator). Coded 0 randomised or holdout, 1 matched asserted, 2 unexposed comparisons, 3 not disclosed. | 9, 12 | synthesis | supported with qualification | medium | [196, 195, 194, 23, 187, 257] |
| LOC-J-F04 | Identifier-based place targeting (geo_places, geo_places_exclude) entered AdCP through PR #6093, merged on 15 August 2026, first published in the pre-release 3.2.0-beta.0 (17 August 2026) and first in a stable release in 3.2.1 (1 October 2026); it keys places by stable IDs in a declared system (geonames, google_ads, microsoft_ads or an owner HTTPS URI) and a place_type drawn from administrative units plus airport, so its registered vocabulary names places such as cities and neighbourhoods rather than store-level points of interest, and the PR explicitly defers place-level forecasting and delivery breakdowns. | 10 | direct record | supported with qualification | high | [203, 202, 597] |
| LOC-J-F06 | An implemented AdCP connector maps administrative geo targeting into an ad server: the Prebid Sales Agent v2.0.0 (1 July 2026), built against AdCP spec 3.1.0-beta.3 (adcp Python SDK 5.7.0) rather than 3.2.6, translates geo_countries, geo_regions and geo_metros (Nielsen DMA only) and their exclusions into Google Ad Manager targetedLocations, but rejects geo_postal_areas with an explicit error ('not implemented in GAM static mapping') and contains no handling for geo_proximity, store_catchments or geo_places in the targeting manager read. | 10 | direct record | supported with qualification | high | [198] |
| LOC-J-F08 | The strongest public claims of location signals in agent-driven AdCP buys are vendor releases that do not record the signal or the protocol version: PubMatic and Butler/Till (15 December 2025) describe an AdCP-enabled CTV campaign for Clubtails &quot;delivering geo-targeted, premium CTV inventory&quot; without naming a field or geographic unit, and Broadsign (27 May 2026) describes a Netherlands OOH campaign in which agents using AdCP informed &quot;audience and venue targeting&quot; without naming identifiers or delivery metrics; neither reaches documented production use of a named location signal under H7's rule. | 10 | vendor assertion | supported as stated | medium | [201, 210] |
| LOC-J-F09 | IAB Tech Lab's AAMP components represent geography only as coarse codes at schema level: OpenProposal 3.0 draft 1 (published for comment 22 September 2026) gives the targeting envelope a settable geo object with an allowed[] list of geo codes, iab-agentic-primitives 0.5.1 (a pre-release) gives packages geo_targets as ISO 3166-2 codes and linear-TV quotes Nielsen DMA codes, and the reference seller-agent 2.4.2 carries geo_targets in its media kit, but a grep of its ad-server clients found no mapping of geo_targets into Google Ad Manager geo targeting even though its GAM model declares a geoTargeting field. | 10 | direct record | supported as stated | medium | [213, 214, 215] |
| LOC-J-F10 | ARTF 1.0 (released 12 November 2025) carries location only inside the embedded OpenRTB 2.x BidRequest that agents receive; its eight mutation intents (activate segments, activate or suppress deals, adjust deal floor or margin, bid shade, add metrics, add content IDs) include none specific to geography, so an ARTF agent can read device.geo and could activate a location-derived audience segment by segment ID, but has no standard intent to change or add a geographic field. | 10 | direct record | supported with qualification | high | [216] |
| LOC-J-F12 | Among location vendors, this review found agent interfaces only outside agent-driven buying: Foursquare publishes an MCP server (created January 2025, last pushed April 2025) whose tools search places and snap coordinates to Foursquare place IDs; Mapbox publishes an MCP server (v0.16.0, 15 October 2026) for geocoding, POI search and isochrones; Radar runs a public documentation MCP endpoint and, announced 14 October 2026, an authenticated enterprise pilot with read-only tools over a customer's own location records; and Placer.ai documents an enterprise MCP server, Pi for Placer MCP, whose page was updated 22 October 2026. No MCP server or AdCP agent was found for GroundTruth, Blis, InMarket or Vistar Media, and none of these vendors appears among the 34 agents in the AAO registry. | 10 | synthesis | supported with qualification | medium | [217, 218, 222, 200] |
| LOC-J-F13 | Inference for H7 at 24 October 2026: across AdCP (3.2.6 schema; the only connector found is built on 3.1.0-beta.3), AAMP components, ARTF 1.0 and the location-vendor interfaces screened, the highest documented state for any location signal in open-protocol agent-driven buying is an implemented connector (AdCP administrative geo to Google Ad Manager in the Prebid Sales Agent v2.0.0, AdCP 3.1.0-beta.3); geographic cells reach no state, geo_proximity, store catchments, geo_places, venue identifiers, DOOH metrics and foot_traffic reach schema representation only, no interoperability test result is public, the two vendor-claimed agentic AdCP campaigns do not record a location field, and proprietary DOOH seller agents reached over MCP (VIOOH, Displayce) claim geographic and proximity handling without naming fields or reporting a transaction, so H7 is unresolved under its registered rule. | 10 | inference | supported with qualification | medium | [198, 200, 201, 205, 598, 599] |
| LOC-RADAR-F16 | Radar runs two MCP servers: the public docs.radar.com/mcp endpoint serves documentation search and retrieval (plus a docs-feedback tool), and an authenticated server at mcp.radar.com/mcp, announced 14 October 2026 and in pilot for enterprise customers by request, gives AI assistants read-only tools to query a customer’s own location records (query_data, get_query_page, render_locations_map); Radar pitches it for fraud investigation, operations and building first-party audience exports, and neither server is an advertising-transaction interface. | 10 | vendor assertion | supported with qualification | medium | [222, 223, 404] |
| LOC-J-F07 | In the public AAO agent registry on 24 October 2026, none of the 34 registered agents shows a passing geo or DOOH conformance test: geo_region_targeting, geo_region_all_values_support, geo_region_exclusion_support and geo_place_targeting appear only on the two agents graded on AdCP 3.2.6 (goTom Sandbox and Publink) and are untested (0/0 steps) on both, no agent declares the sales-dooh specialism, and the 11 agents holding a verified badge hold it only in spec mode, none in the live-traffic mode. | 10, 13 | direct record | supported with qualification | medium | [200, 597] |
| LOC-GQ-F01 | Seven scored products carry the author's privacy risk cap because a regulator order, entered judgment or confirmed breach involving their location data falls inside the three years before the cutoff: InMarket's measurement and audience products, Mobilewalla's audiences, Unacast's foot-traffic and device-level data, Collective Data Solutions' feeds and Outlogic's Observation Panel. Kochava's DOOH measurement keeps the order as an event flag without the cap, because no link between the order and that product's data is established. With the cap removed and each cohort's field mean recomputed, the seven Ground Quotients rise by 2 to 5 points. Three would move up a band (InMarket visit-lift measurement and Outlogic Observation Panel from Partly to Well documented, Mobilewalla audiences from Least to Partly documented), and two uncapped products would move down as their cohorts' field means rise (Placer.ai from Most to Well documented, Veraset from Well to Partly documented). | 11 | synthesis | supported as stated | high | [130, 131, 132, 143, 128, 137, 142, 145] |
| LOC-VISTAR-F05 | On 5 October 2026 Vistar announced enhanced T-Mobile audience segments for DOOH (thousands of segments on an 'expanded universe of consented data', now including 'location insights'), available exclusively through the Vistar DSP, which T-Mobile turns into area-level audience scores that Vistar matches to screens. T-Mobile app-ownership and engagement segments had been available through Vistar's self-service DSP since 6 May 2025. | abstract, 1, 2, 6, 9 | vendor assertion | supported with qualification | medium | [334] |
| LOC-C-F11 | PlaceIQ's undated A/B case for a national big-box retailer reports a Place Visit Rate of 2.75% for exposed devices against 2.13% for a standing 10% holdout never served a PlaceIQ ad (29% lift). The document says how the holdout is maintained but not how it is selected. Against the open population the rate is 0.67% (310% 'open lift'). The same campaign therefore shows a 10.7-fold larger lift when the comparator is unexposed non-targeted users rather than the holdout. The case does not show whether PlaceIQ (now Precisely) runs this design today. | abstract, 1, 5 | vendor assertion | supported with qualification | medium | [12] |
| LOC-E-F29 | Under the Q1 coding rule, the visit numbers sold by cohort A products are documented as projected for Foursquare Attribution, InMarket LCI (projected store visits), GroundTruth Projected Visits and Azira (extrapolated total visits), and as observed for GroundTruth Observed Visits, Azira observed visits and InMarket's exposed device visits. Arrivalist MediaInsights' spend lift is declared (card transactions), while the basis of its arrival counts is not stated (the page lists location pings as well as cards). The basis is 'not disclosed' for Blis, Reveal, Adsquare, Cuebiq, Vistar, Kochava, Lifesight and Precisely. Declared card-based sales metrics are also sold by Foursquare, InMarket (RSA) and Cuebiq (Affinity Solutions). | abstract, 1, 5, 9 | synthesis | supported with qualification | medium | [16, 17, 18, 5, 449] |
| LOC-A-F01 | MRC's only location-specific standard is the 'MRC Location-Based Advertising Measurement Guidelines', Final Version 1.0, dated March 2017 and sponsored by MRC, IAB and MMA; it was issued in draft for a 30-day public comment period in November 2016, finalised on 29 March 2017, and MRC's standards index at 24 October 2026 shows it as issued March 2017 with no later update. | abstract, 1, 8 | direct record | supported as stated | high | [445, 600, 601] |
| LOC-D-F15 | ZeroToOne.AI announced on 25 March 2026 that it had acquired GroundTruth (terms and seller undisclosed), GroundTruth's site now describes itself as 'a ZeroToOne.AI company', and the acquirer's counsel confirmed the transaction on 24 April 2026; the event is an acquisition, announced as completed. | abstract, 2 | vendor assertion | supported as stated | medium | [47, 48, 49] |
| LOC-U-F02 | Infillion, which owns Gimbal, announced on 18 September 2026 an agreement to acquire Foursquare (terms undisclosed), describing closing as a future event. By 24 October 2026 Foursquare's homepage says it &quot;has joined Infillion&quot;, co-founder Dennis Crowley wrote on 20 October 2026 that Foursquare had been acquired about a month earlier, and a shareholder wrote on 24 September 2026 that it had been sold. No closing release, filing or exact closing date was found, so completion rests on participants' statements. | abstract, 2 | vendor assertion | supported with qualification | medium | [52, 347, 602, 603, 604] |
| LOC-U-F05 | T-Mobile US completed its acquisitions of Vistar Media on 3 February 2025 (agreement 20 December 2024) and of Blis Holdco Limited on 3 March 2025 (agreement 18 February 2025), transferring cash of US$621 million and US$180 million respectively. Excluding payments that settled preexisting relationships, the 10-K puts the fair value of consideration at US$617 million (Vistar) and US$174 million (Blis), both provisional. Blis's homepage is titled 'Blis, part of T-Ads \| Omnichannel DSP'. | abstract, 2, 6, 9 | direct record | supported with qualification | high | [36, 228] |
| LOC-G-F20 | In the sources screened by the cutoff, this review found no filing, official platform statement or independent measurement that attributes a quantified change in US SDK-sourced location supply to a named Apple or Google permission change; the Telenav FY2019 and FY2020 10-Ks do not mention iOS 13 or location permissions, the Near S-4/A attributes no figure, and EDGAR full-text searches for 10-Ks pairing location data with App Tracking Transparency, iOS 13 or background location returned no location vendor. | abstract, 3 | synthesis | supported with qualification | low | [505, 504, 88] |
| LOC-H-F03 | The InMarket order (C-4803, issued 29 April 2024) bans selling or licensing Location Data and bars sensitive-location products; for Historic Location Data (data collected without affirmative express consent) it requires deletion within 90 days of data collected through InMarket's own apps unless the consumer gives affirmative express consent to retention, deletion, deidentification or rendering non-sensitive within 120 days of data obtained from third parties, and deletion of audience segments built from historic data within 120 days. The order runs 20 years. | abstract, 7 | direct record | supported with qualification | high | [130, 131, 132] |
| LOC-H-F08 | In an undated notice to Norway's Datatilsynet (filed by BakerHostetler, published by NRK), Gravy Analytics, a Unacast subsidiary, said it was informed on 4 January 2025, by the intruder, of unauthorised access to its AWS cloud storage, apparently through a misappropriated access key; that the intruder obtained some files whose contents and personal-data status were under investigation; that the incident's duration was under investigation; that it took its data-processing platform offline and had it operational again as of 9 January 2025; that it was still assessing whether the incident was a reportable personal data breach; and that it was also informing the UK ICO. This is a security incident, not an enforcement action. | abstract, 7 | direct record | supported with qualification | medium | [605] |
| LOC-I-F05 | On 31 March 2026 Geopath and the OAAA announced that their measurement modernisation 'began in early 2025', that 'Ipsos has been selected to support an industry pilot launching in the second half of 2026' to 'evaluate a next-generation OOH audience measurement system across multiple formats and environments' 'including updated methods for estimating impressions, reach, and frequency', with 'an orderly transition from the current Geopath measurement system to a next-generation framework beginning in 2027' and 'full adoption in 2028'; the release names no data sources. | abstract, 8, 9 | direct record | supported as stated | high | [606, 607] |
| LOC-I-F23 | Of six US DOOH products and currencies coded (Geopath; Vistar attribution; Vistar audience targeting; Place Exchange PerView; Broadsign; Perion; Talon is excluded as coded only from a UK-hosted global page), three name their location-data sources in their own or the operator's documentation: Geopath, Vistar attribution (InMarket, Foursquare/Placed) and Vistar audience targeting (Vistar-hosted partner pages naming GroundTruth's first-party, third-party and bid-stream data and Adsquare mobile location data). Place Exchange, Broadsign and Perion do not. Only Geopath states a panel size with a period: 7–10M qualified devices weekly from more than 100M US devices monthly, in 31 May 2023 documentation with no observation vintage stated. Counts: 3 of 6 disclose a source and 1 of 6 a panel size. | abstract, 9, 12 | synthesis | supported with qualification | medium | [182, 23, 192, 195, 608, 194] |
| LOC-HYP-H2 | H2 (ground-truth agreement) is unresolved overall: no independent multi-site comparison of any vendor's projected retail visits with door counts or transactions was found. The independent evidence that exists is historical and narrow, and for SafeGraph's 2018-vintage panel it is adverse (lower coverage at polling places with older and non-white voters). | abstract, 11, 12, appendix | synthesis | supported as stated | medium | [98, 99, 102] |
| LOC-HYP-H6 | H6 (a US visit metric holds current MRC accreditation) is contradicted as of 24 October 2026: none of MRC's eleven accredited-services and under-review lists carries a visit, visitation, foot-traffic or location metric, and both visit data sets MRC once accredited have lost accreditation. | abstract, 11, 12, appendix | synthesis | supported as stated | high | [171, 172, 173, 174, 175, 176, 177, 178, 179, 180] |
| LOC-HYP-H1 | H1 (supply contraction and substitution) is unresolved: no vendor discloses a US panel series on one definition from before 2021 to after 2023, and no filing or official statement attributes a quantified change in US SDK location supply to a named platform change, in the sources screened by the cutoff. | abstract, 12, appendix | synthesis | supported as stated | medium | [275, 63, 289, 427, 428, 505, 504, 88] |
| LOC-HYP-H3 | H3 (randomised or holdout control by default) is contradicted: of the eight cohort A seed products that document a general lift method, six use a matched or synthetic control and two a randomised or holdout control (Azira, Blis); a ninth product documents exposed-only visit counts with no lift claim, and counted with it seven of nine documented visit-measurement methods are non-holdout. Five products document no default. | abstract, 12, appendix | synthesis | supported as stated | high | [5, 19, 119, 17, 20, 21, 22, 25, 6, 446] |
| LOC-HYP-H4 | H4 (bidstream or IP as the main source of location audiences) is unresolved: fewer than four cohort B products document a primary source, and the vendors that list several sources (Foursquare, InMarket, Azira) do not say which is primary. | abstract, 12, appendix | synthesis | supported as stated | medium | [265, 420, 421, 422, 458, 457] |
| LOC-HYP-H5 | H5 (the five FTC-ordered companies no longer sell precise or sensitive location data) is contradicted: Outlogic and Collective Data Solutions (Kochava's subsidiary) market precise-location offers at the cutoff, and Gravy Analytics, trading as Unacast, offers device-level data and reports selling precise geolocation in the preceding year; Mobilewalla is unclear on precision and InMarket documents no such product. | abstract, 12, appendix | synthesis | supported as stated | high | [128, 137, 142, 145, 134, 130] |
| LOC-HYP-H7 | H7 (location signals in production use in agent-driven buying) is unresolved: the highest state any location signal reaches is an implemented connector (the Prebid Sales Agent, targeting AdCP 3.1.0-beta.3), no interoperability test result is public, and geographic cells reach no state. | abstract, 12, appendix | synthesis | supported as stated | medium | [198, 200, 201, 205] |

### The corrections

Every substantive change after the verification passes, with what questioned it, what was checked, the verdict, the change and how it was verified.

**Corrections ledger**

| ID | Original claim or score | Challenge | Evidence checked | Verdict | Change made | Reason | Affected |
|---|---|---|---|---|---|---|---|
| R2-01 | MRC's only location-specific standard is the 'MRC Location-Based Advertising Measurement Guidelines', Final Version 1.0, dated March 2017 and sponsored by MRC, IAB and MMA; it was issued in draft for a 30-day public comment period in November 2016, finalised on 29 March 2017, and MRC's standards index at 24 October 2026 shows it as issued March 2017 with no later update. | The March 2017 version, three sponsoring associations, November 2016 consultation and 29 March 2017 final release are confirmed. Bound the update claim to the MRC index opened on 24 October 2026; that index lists this dedicated location guideline without a later update. | Reviewer opened: https://mediaratingcouncil.org/standards-and-guidelines https://mediaratingcouncil.org/sites/default/files/Standards/MRC%20Location-Based%20Advertising%20Measurement%20Guidelines%20Final%20March%202017.pdf https://www.iab.com/wp-content/uploads/2017/03/March2017-Location-Based-Guidelines-PR_final.pdf | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F01; data/claims-overlay.json |
| R2-02 | H6 test: on 24 October 2026 none of MRC's six accredited-services lists (Digital, Multimedia, Out-of-Home, Television, Radio, Outside the U.S.) carries an accredited visit, visitation, foot-traffic, store-visit or location-determination metric for any product, none of the five under-review lists (Digital, Out-of-Home, Television, Radio, Services Outside the U.S.) shows a location or visit audit in progress, and MRC's two most recent decision communications (April-June and July-September 2026) record no location or visit action; the Out-of-Home and Outside-the-U.S. accredited lists are empty. | All six accredited lists and five under-review lists were opened. I found no listed visit, visitation, footfall or location-determination metric; Q2 and Q3 decisions add none. Kevel conversions and Kroger sales are not visit accreditation. The verifier is right. Date this conclusion 24 October, not the still-future 30 October freeze. The empty OOH and outside-US accredited lists are confirmed. The digest's television enumeration is incomplete because the live list also includes Nielsen, without changing H6. | Source check G1 (24 October 2026, primary sources opened): R2-02: H6 zero confirmed on re-fetch (24 Oct). Fix the enumeration: TV accredited = Comscore TV, iSpot.tv and Nielsen Media Research, and correct the under-review counts to match the verifier (Digital ~15, TV 1, Radio 2). https://mediaratingcouncil.org/accreditation/services-under-review/digital https://mediaratingcouncil.org/accreditation/services-under-review/digital https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/MRC_Accreditation_Status_Quarterly_Update_Q3_2026.pdf https://mediaratingcouncil.org/accreditation/out-of-home | supported with qualification | On 24 October 2026, none of MRC's six accredited-services lists and none of its five under-review lists carries a visit, visitation, foot-traffic or location-determination metric. The closest under-review entries are Kevel's 'Conversions and incrementality' and Kroger's attributed and incremental sales. MRC's April–June and July–September 2026 communications record no location or visit action, and the Out-of-Home and Outside-the-U.S. accredited lists are empty. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F10; data/claims-overlay.json |
| R2-03 | Foursquare's Visits data set was granted MRC accreditation in July 2020 for 'location information reported in Visits data set' (the first accreditation under the 2017 guidelines), was suspended in October 2022 and was revoked in December 2022; Foursquare's own release said the accreditation did not extend to its Attribution product, and MediaPost reported that Placed powered by Foursquare was not audited. | MRC records confirm July 2020 grant, October 2022 suspension and December 2022 revocation. Foursquare explicitly excludes Attribution from the accreditation. Preserve the narrower description: the first since the 2017 guidelines, not the first location-related accreditation of any kind. | Reviewer opened: https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/101320%20MRC%20Accreditation%20Status%20Communication.pdf https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/012423%20MRC%20Accreditation%20Status%20Communication.pdf https://foursquare.com/resources/blog/news/foursquare-visits-receives-first-media-rating-coun/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F11; data/claims-overlay.json |
| R2-04 | GroundTruth's 'Location, Place, and Visit data set' was granted MRC accreditation in July 2021, placed on a six-month hiatus at GroundTruth's request in September 2023 (not accredited during the hiatus), and revoked after the hiatus 'due to withdrawal from the MRC accreditation process' (row labelled February, reported in MRC's communication of 26 July 2024); GroundTruth's 2021 release claiming the accreditation is still online. | MRC confirms July 2021 grant, September 2023 six-month hiatus and no accreditation during hiatus. The July 2024 communication labels revocation February; preserve both the action label and communication date rather than silently assigning a new month. GroundTruth's 2021 announcement remains online. | Reviewer opened: https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/102021%20MRC%20Accreditation%20Status%20Communication.pdf https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/101623_MRC_Accreditation_Status_Communication.pdf https://mediaratingcouncil.org/sites/default/files/News/MRC%20Accreditation%20Status%20Quarterly%20Update_Q2%2024%207.26.pdf https://www.groundtruth.com/article/groundtruth-earns-mrc-accreditation-for-visitation-data/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F12; data/claims-overlay.json |
| R2-05 | OpenRTB 2.6's Geo object carries lat and lon as floats, a 'type' for the source of location (AdCOM Location Types: 1 GPS/Location Services, 2 IP Address, 3 User Provided), 'accuracy' in metres as reported by the device (recommended for type 1), 'lastfix' in seconds since the fix, and 'ipservice' (AdCOM IP Location Services, beginning ip2location, Neustar (Quova), MaxMind), and it states that lat/lon should be passed only if they conform to the accuracy of the stated type and that a postal-code centroid should not be passed; the 2.6 text contains no rule on truncating or rounding coordinates. | The Geo fields, units, source codes and postal-centroid warning match the opened specifications. Section 3.2.19 provides no coordinate-rounding rule. Record a commit hash or archived copy because both cited branches can change; coordinate precision is not guaranteed accuracy. | Reviewer opened: https://github.com/InteractiveAdvertisingBureau/openrtb2.x/blob/main/2.6.md https://github.com/InteractiveAdvertisingBureau/AdCOM/blob/main/AdCOM%20v1.0%20FINAL.md | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F16; data/claims-overlay.json |
| R2-06 | NAI's Precise Location Information Solution Provider Voluntary Enhanced Standards were first issued on 22 June 2022 with Cuebiq, Foursquare and Precisely PlaceIQ as the initial signatories, and reissued as 'Updated and effective October 2024'; they apply only to signatories' US activities, bar any use, licensing, sale or transfer of device or user activity correlated to a known Sensitive POI (eleven definitive categories including places of worship, correctional facilities, shelters, addiction treatment and sensitive-condition medical facilities, plus two non-definitive categories), bar use o… | The verifier correctly adds contractual downstream restrictions and limited exceptions. Apply the bans subject to section IV: consented direct-relationship services requested by consumers, compliance processing and tightly limited technical services. The 2022 release confirms Cuebiq, Foursquare and Precisely PlaceIQ. Visual inspection of the current PDF's final page identifies five logos: Cuebiq, Precisely PlaceIQ, Foursquare, Outlogic and Veraset; label this list last updated March 2025, rather than saying it cannot be read or treating it as a fresh October 2026 audit. Eleven definitive and two non-definitive POI categories, US scope and annual NAI staff assessment are confirmed. | Source check G1 (24 October 2026, primary sources opened): R2-06: the reviewer's five-logo list is wrong. The March 2025 page shows four logos (Cuebiq, Foursquare, InMarket, Outlogic), with no PlaceIQ and no Veraset. Replace 'could not be read' with these four and add the section IV exceptions. https://thenai.org/wp-content/uploads/2025/03/NAI-Precise-Location-Information-Solution-Provider-Voluntary-Enhanced-Standards.pdf https://thenai.org/press/leading-companies-commit-to-privacy-first-nai-enhanced-standards-for-precise-location-information/ https://thenai.org/wp-content/uploads/2025/03/NAI-Precise-Location-Information-Solution-Provider-Voluntary-Enhanced-Standards.pdf https://thenai.org/wp-content/uploads/2025/03/NAI-Precise-Location-Information-Solution-Provider-Voluntary-Enhanced-Standards.pdf | supported with qualification | NAI's Precise Location Enhanced Standards were first issued on 22 June 2022 with Cuebiq, Foursquare and Precisely PlaceIQ as signatories. They were reissued 'Updated and effective October 2024', and the PDF's signatory page, 'Last updated March 2025', shows four logos: Cuebiq, Foursquare, InMarket and Outlogic. Within US activities they bar using or transferring activity correlated to a known Sensitive POI (11 definitive and 2 non-definitive categories) and law-enforcement, national-security or bounty-hunting uses, subject to section IV's exceptions for consented direct-relationship services, blocking directories, technical providers and compliance processing. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F20; data/claims-overlay.json |
| R2-07 | An independent audit of SafeGraph's point-of-interest data (about 47 million US devices) against North Carolina's 2018 general-election voter file found that polling places with older and with more non-white voters had lower coverage: rank correlations of coverage with the share of voters over 65 of -0.14 (p<0.001) and with the share of non-white voters of -0.11 (p=0.0067). | The 2018 North Carolina SafeGraph study reports age-65-plus correlation −0.14 (p<0.001) and non-white share correlation −0.11 (p=0.0067). These concern polling-place coverage, not current retail-visit accuracy. The verifier is right. | Reviewer opened: https://arxiv.org/pdf/2011.07194 | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-B-F01; data/claims-overlay.json |
| R2-08 | Across 2018-2022 SafeGraph's monthly panel of residing devices averaged 7.5% of the US population (range 4.5% to 14.5%); device counts tracked census population closely at county level (Pearson r 0.986 for urban and 0.934 for rural counties) and more loosely at tract level (r above 0.683 for urban and 0.830 for rural tracts). | Accept the verifier's correction. The 4.5–14.5% extrema span the national, urban and rural monthly series; the national extremes cited in the paper are 4.9% and 13.4%. The quoted county and tract correlations describe the 2019 figure, not a pooled five-year fit; urban-tract r² falls to 0.375 in 2022. These are representativeness comparisons against population, not validation of projected store visits. | Source check G2 (24 October 2026, primary sources opened): R2-08 accepted with one change. All statistics match the paper, but the paper never labels 13.4%/4.9% as the national series, so the wording now says 'overall trend' and drops 'national'. https://pmc.ncbi.nlm.nih.gov/articles/PMC10798630/ https://pmc.ncbi.nlm.nih.gov/articles/PMC10798630/ https://pmc.ncbi.nlm.nih.gov/articles/PMC10798630/ https://pmc.ncbi.nlm.nih.gov/articles/PMC10798630/ | supported with qualification | Across 2018-2022, SafeGraph's monthly sampling rate (residing devices divided by total population) averaged 7.5%. Monthly values ranged from 4.5% to 14.5% across the national, urban and rural series, and the overall trend peaked at 13.4% in May 2022 before falling to 4.9% in July 2022. In 2019, device counts tracked census population closely at county level (Pearson r 0.986 urban, 0.934 rural) and moderately at tract level (r above 0.683 urban, above 0.830 rural). The urban-tract fit weakened to r-squared 0.375 by 2022. These results compare the panel with population counts and do not validate visit counts. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-B-F05; data/claims-overlay.json |
| R2-09 | In Yellowstone, SafeGraph Patterns visits (January 2018-November 2020) correlated with NPS monthly recreation visits at r=0.88 and r=0.85 for the two most-visited POIs, with a trail counter's daily counts at r=0.76 (2018) and 0.77 (2019), and with park-wide gate vehicle counts at r=0.79 in 2019 but only 0.47 in 2020, when the compared facility imposed indoor capacity limits. | Downloaded and read the primary PDF directly after the browser fetch failed. All six quoted correlations and the January 2018–November 2020 Patterns range match. Specify that individual facility POI series are compared with wider park reference counts; these correlations do not establish count agreement. Capacity limits are offered as a potential explanation for the 2020 mismatch, not an experimentally identified cause. | Reviewer opened: https://par.nsf.gov/servlets/purl/10451340 | supported with qualification | Downloaded and read the primary PDF directly after the browser fetch failed. All six quoted correlations and the January 2018–November 2020 Patterns range match. Specify that individual facility POI series are compared with wider park reference counts; these correlations do not establish count agreement. Capacity limits are offered as a potential explanation for the 2020 mismatch, not an experimentally identified cause. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-B-F10; data/claims-overlay.json |
| R2-10 | The only independent comparison of Placer.ai found by the cutoff, a University of Minnesota Extension case study at one Minnesota recreation area in summer 2021, recorded 10,138 visits on automated trail counters against mobile-platform estimates of 13,233 and 14,544 (Placer.ai and StreetLight; the slide does not make the assignment legible) with no statistically significant difference, but the platforms showed weekends 7 times busier than weekdays against 2.1 times in the counter data, and their inferred demographics differed significantly from the intercept survey on race, education and inc… | Downloaded the April 2024 primary report and visually checked slide 19: Placer.ai is 13,233 and StreetLight 14,544 versus 10,138 counter visits during 29 May–6 September 2021. The verifier assigns the figures correctly. State that these are 30.5% and 43.5% higher totals, not proof of equivalent accuracy. The report identifies unequal polygons, event handling and activity populations; StreetLight trips are multiplied by three passengers. The 7× versus 2.1× weekend comparison and demographic differences are reported, with these comparability limits. Bound any only-study claim to the sources screened. | Source check G2 (24 October 2026, primary sources opened): R2-10 accepted. Slide 19 visually confirms Placer.ai 13,233 and StreetLight 14,544, and the report states the three-passenger adjustment. https://conservancy.umn.edu/server/api/core/bitstreams/fdf3b6a9-7824-4e64-b330-f6392f4491b4/content https://conservancy.umn.edu/server/api/core/bitstreams/fdf3b6a9-7824-4e64-b330-f6392f4491b4/content https://conservancy.umn.edu/server/api/core/bitstreams/fdf3b6a9-7824-4e64-b330-f6392f4491b4/content | supported with qualification | In the only independent comparison of Placer.ai found in the sources screened, a University of Minnesota Extension case study (April 2024) at one Minnesota recreation area, Placer.ai estimated 13,233 visits and StreetLight 14,544 for 29 May-6 September 2021. Those totals are 30.5% and 43.5% above the 10,138 recorded by trail counters. StreetLight's figure is vehicle trips multiplied by a median of three passengers. The authors found no statistically significant difference, but the comparison is not like for like: polygons differ, counters cover only the cross-country trails, and events are handled differently. The 7x versus 2.1x weekend ratio and the demographic differences should be read with those limits. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-B-F12; data/claims-overlay.json |
| R2-11 | Across 15 US Facebook advertising campaigns run as randomised controlled trials in 2015 (500 million user-experiment observations, 1.6 billion impressions), matching and regression methods using rich individual-level data mostly overestimated lift relative to the experiments, and in half of the studies the estimated percentage increase in purchases was off by a factor of three across all methods. | The experiment year, 15 US campaigns, 500 million user-experiment observations, 1.6 billion impressions and factor-of-three conclusion match the paper. Qualify the evidence label: this is an academic–Facebook collaboration with two Facebook-employed coauthors; the two academics disclose no financial interest or compensation. It is useful methodological evidence, but not wholly independent of the platform studied. The internal verifier checked the numbers but missed this authorship qualification. | Source check G2 (24 October 2026, primary sources opened): R2-11 accepted with one change: relabel the evidence class from independent measurement to academic-platform collaboration, since two coauthors list Facebook affiliations. The figures are all correct, and the researcher had already noted the coauthors. https://www.kellogg.northwestern.edu/faculty/gordon_b/files/fb_comparison.pdf https://www.kellogg.northwestern.edu/faculty/gordon_b/files/fb_comparison.pdf https://www.kellogg.northwestern.edu/faculty/gordon_b/files/fb_comparison.pdf | supported with qualification | Fifteen US Facebook advertising campaigns run as randomised controlled trials in 2015 (500 million user-experiment observations, 1.6 billion impressions) were used to test observational methods. Matching and regression methods using rich individual-level data mostly overestimated lift relative to the experiments, and in half of the studies the purchase lift was off by a factor of three across all methods. The study is an academic-Facebook collaboration on Facebook's own data: two coauthors are affiliated with Facebook, and the two Kellogg academics declare no financial interest or compensation. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-B-F23; data/claims-overlay.json |
| R2-12 | This review found no independent, published comparison of any named US vendor's projected visit counts with door counts or transactions across many retail places and months, and no vendor-published held-out validation with error statistics, in the sources screened by the cutoff; the independent ground-truth evidence consists of park-visitation correlations for SafeGraph (0.47-0.88), Advan (0.875) and a single-site counter comparison for Placer.ai, so H2 is unresolved for every product in the register. | The internal verifier is right that the summary omits the 2018 SafeGraph polling-place audit and that blanket H2-unresolved wording fails the registered systematic-bias rule. Include that audit and record a historical, use-specific adverse result. Retain unresolved status for current retail projections lacking matched validation; do not carry an old panel verdict to all current products. The sources opened here do not independently establish the digest's entire bounded absence claim. | Reviewer opened: https://arxiv.org/pdf/2011.07194 https://conservancy.umn.edu/server/api/core/bitstreams/fdf3b6a9-7824-4e64-b330-f6392f4491b4/content https://par.nsf.gov/servlets/purl/10451340 | contradicted | The internal verifier is right that the summary omits the 2018 SafeGraph polling-place audit and that blanket H2-unresolved wording fails the registered systematic-bias rule. Include that audit and record a historical, use-specific adverse result. Retain unresolved status for current retail projections lacking matched validation; do not carry an old panel verdict to all current products. The sources opened here do not independently establish the digest's entire bounded absence claim. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-B-F31; data/claims-overlay.json |
| R2-13 | Azira's methodology page (footer 2025) is the only cohort A vendor document opened in this stream that documents a randomised control as the default: 10-15% of impressions serve public-service creative as a PSA holdout, with a synthetic control from its identity graph as the alternative, polygon-based visits, extrapolation of observed visits by a detection probability capped at 20x, minimums of about 1,000 observed visits and 100,000 exposed devices for breakouts, 99% confidence intervals and a two-tailed t-test. | The source supports the stated methodology as Azira's own account. Its footer year is not a publication date. Keep the claim labelled vendor assertion. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology | supported with qualification | The source supports the stated methodology as Azira's own account. Its footer year is not a publication date. Keep the claim labelled vendor assertion. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-C-F05; data/claims-overlay.json |
| R2-14 | PlaceIQ's undated A/B case for a national big-box retailer reports a Place Visit Rate of 2.75% for exposed devices against 2.13% for a standing 10% holdout never served a PlaceIQ ad (29% lift) and 0.67% for the open population (310% 'open lift'), so the same campaign shows a 10.7-fold larger lift when the comparator is unexposed non-targeted users rather than the randomised holdout. | The reported rates and lifts match: exposed 2.75%, holdout 2.13%, open population 0.67%; rounded lifts 29% and 310%, a 10.7-fold ratio. The verifier is right to remove 'randomised': the PDF does not state the assignment procedure. The undated case does not establish a current default. | Source check G2 (24 October 2026, primary sources opened): R2-14 accepted. The PDF never calls the holdout random, and all figures recompute. https://mmaglobal.com/files/a-b_control_case_study_final.pdf https://mmaglobal.com/files/a-b_control_case_study_final.pdf https://mmaglobal.com/files/a-b_control_case_study_final.pdf | supported with qualification | PlaceIQ's undated A/B case for a national big-box retailer reports a Place Visit Rate of 2.75% for exposed devices against 2.13% for a standing 10% holdout never served a PlaceIQ ad (29% lift). The document says how the holdout is maintained but not how it is selected. Against the open population the rate is 0.67% (310% 'open lift'). The same campaign therefore shows a 10.7-fold larger lift when the comparator is unexposed non-targeted users rather than the holdout. The case does not show whether PlaceIQ (now Precisely) runs this design today. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-C-F11; data/claims-overlay.json |
| R2-15 | Vistar Media's two US DOOH cases measured by Foursquare report, for Target (flight 2021-08-15 to 2022-01-29; 22 states; $1.2M+ budget), 72 million+ impressions, 533,000 store visits inside a four-day look-back and a 6.51% lift in visitation 'compared with non-exposed consumers', and for Jack in the Box (twelve markets; undated), 49M+ impressions, 1.3M+ store visits and an 8.8% lift; exposure is inferred from ad-log timestamps, device location, venue type and screen location, and the OAAA document says Vistar's studies 'compare control versus exposed groups' without describing the control. | The opened vendor cases support the Target flight, budget, states, 72M-plus impressions, 533,000 visits, four-day lookback and 6.51% lift, and Jack in the Box's twelve markets, 49M-plus impressions, 1.3M-plus visits and 8.8% lift. Retain their vendor-case evidence class: the PDFs do not establish randomized controls or independently validated causal effects. | Reviewer opened: https://oaaa.org/wp-content/uploads/2022/12/Target_MPA_2022_FINAL.pdf https://oaaa.org/wp-content/uploads/2024/02/Vistar-Media-US-Case-Study-DOOH-QSR-Jack-in-the-Box.pdf | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-C-F14; data/claims-overlay.json |
| R2-16 | Arrivalist's 2021 Lee County analysis scaled 492,897 panel trips to 9,861,477 projected US adult trips (a factor of 20.0) and 932,706 panel point-of-interest trips to 22,730,788, with only 60,609 'panel attributed arrivals' tied to media exposure, from a panel described as GPS data 'accurate to 30 feet' balanced monthly to the US Census and excluding international visitors, under-18s, trips under 50 miles or under 4 hours, one-way trips and commuters. | The panel and projected trip counts, approximately 20.0-fold overall expansion, geographic/trip exclusions and vendor accuracy claim are supported. Preserve the slide's caveat: 60,609 panel-attributed arrivals are a small subset of all attributed arrivals, not the total attributed-arrival figure. The inclusion slide also lists connected-vehicle panel trips. | Reviewer opened: https://www.visitfortmyers.com/sites/default/files/2022-03/LVCB%20-%20Industry%20Meeting%20-%20Arrivalist%20Distribution%20Version.pdf | supported with qualification | The panel and projected trip counts, approximately 20.0-fold overall expansion, geographic/trip exclusions and vendor accuracy claim are supported. Preserve the slide's caveat: 60,609 panel-attributed arrivals are a small subset of all attributed arrivals, not the total attributed-arrival figure. The inclusion slide also lists connected-vehicle panel trips. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-C-F22; data/claims-overlay.json |
| R2-17 | Across the cohort A seed vendors whose documentation states a default control design in the sources opened by this stream, one (Azira) documents a randomised PSA holdout as the default, four (Foursquare, InMarket, Cuebiq, adsquare) document a modelled, matched or synthetic control, and one (GroundTruth) reports exposed-only observed visits with no control; Simpli.fi, Vistar, Blis and Reveal either report exposed-only visitation or assert a control without a method; PlaceIQ documents a standing holdout in one undated case whose current status is unknown. | The limited Stream C screen and Stream E's broader screen need reconciling, not competing randomization counts. Blis's older smoothie case does not state a method, while its dedicated product page describes a machine-balanced geographic holdout. Neither opened Blis page establishes random assignment or use by default. Code available design separately from documented default. InMarket's cited root opened only a sign-in page here, so its contribution to the exact count remains unresolved in this review. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://cuebiq.com/blog/advanced-measurement-targeting-platform/ https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://help.groundtruth.com/hc/en-us/articles/360000695267-Observed-Visits https://blis.com/tropical-smoothie-cafe-case-study/ https://blis.com/omnichannel-measurement/ https://lciadminconsole.inmarket.com/ | supported with qualification | The limited Stream C screen and Stream E's broader screen need reconciling, not competing randomization counts. Blis's older smoothie case does not state a method, while its dedicated product page describes a machine-balanced geographic holdout. Neither opened Blis page establishes random assignment or use by default. Code available design separately from documented default. InMarket's cited root opened only a sign-in page here, so its contribution to the exact count remains unresolved in this review. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-C-F23; data/claims-overlay.json |
| R2-18 | Only two of the vendor documents opened disclose minimum-sample and significance rules for visit lift: InMarket (10 million or 25 million impressions per campaign depending on channel; '>95%' confidence; standard error reported) and Azira (about 1,000 observed visits and 100,000 exposed devices; 99% confidence intervals; two-tailed t-test); Vistar states only that it requires 'enough scale', Reveal shows a '95%' label, and Foursquare, Cuebiq, adsquare, Blis, Simpli.fi and GroundTruth state none on the pages opened. | Azira's numeric minima and significance rules are verified. InMarket's cited glossary was not accessible: both direct and search-result opens redirected to sign-in, so its thresholds and the exact 'only two' total remain unresolved. Reveal's current measurement page discloses confidence intervals, p-values and computation using formulas/simulations, beyond a dashboard's 95% label; no numeric minimum is stated there. Bound all absence claims to the pages actually screened. | Reviewer opened: https://lciadminconsole.inmarket.com/ https://business.azira.com/azira-location-measurement-and-attribution-methodology https://oaaa.org/wp-content/uploads/2022/12/Target_MPA_2022_FINAL.pdf https://revealmobile.com/solutions/measurement https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://cuebiq.com/blog/advanced-measurement-targeting-platform/ https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://blis.com/omnichannel-measurement/ | supported with qualification | InMarket's minimum-sample and significance thresholds sit in a glossary behind a sign-in and are marked unverified from public sources; Choozle documents a different minimum (30 days and 1 million impressions a month) for its InMarket Visit Stream integration only (see LOC-C-F24 final wording). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-C-F24 |
| R2-19 | T-Mobile US agreed on 20 December 2024 to acquire 100% of Vistar Media Inc., completed the acquisition on 3 February 2025 and transferred cash of US$621 million, per its FY2025 Form 10-K; the event is an acquisition (completed). | The agreement date, 100% stake, completion date and cash transferred are supported by T-Mobile’s September 2025 10-Q acquisition note. The cited FY2025 10-K exceeded the retrieval size limit; cite the opened note instead. Preserve “cash transferred”; it includes settlement of pre-existing relationships and is not identical to acquisition consideration. | Source check G3 (24 October 2026, primary sources opened): LOC-D-F01: added 'cash transferred' vs US$617m consideration (10-K Note 2); kept the 10-K citation, since the reviewer's claim that it was unretrievable is wrong (curl with SEC UA works). https://www.sec.gov/Archives/edgar/data/1283699/000128369925000154/R11.htm https://www.sec.gov/Archives/edgar/data/1283699/000128369926000010/tmus-20251231.htm https://www.sec.gov/Archives/edgar/data/1283699/000128369926000010/tmus-20251231.htm | supported with qualification | T-Mobile US agreed on 20 December 2024 to acquire 100% of Vistar Media Inc. and completed the acquisition on 3 February 2025, transferring cash of US$621 million, per its FY2025 Form 10-K. Note 2 of that filing puts the fair value of consideration transferred at US$617 million, because part of the payment settled preexisting relationships with Vistar. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-D-F01; data/claims-overlay.json |
| R2-20 | The Delaware court approved the sale of substantially all of Near Intelligence's assets to BTC Near Holdco LLC (Blue Torch) on 22 February 2024 for a credit bid; the sale closed on 1 March 2024; the plan of liquidation was confirmed on 15 March 2024 and became effective on 27 March 2024, when all common stock and warrants were cancelled; a Form 15 deregistration was filed on 28 March 2024. | The sale, credit bid, court approval, confirmation, effective date and cancellation are supported. The actual Form 15 was opened and is signed 27 March 2024; the claimed filing date of 28 March could not be independently verified because the SEC submissions JSON and filing-index routes were inaccessible. Keep signature and filing dates distinct; the verifier’s unqualified support is too broad. | Source check G3 (24 October 2026, primary sources opened): LOC-D-F07: kept the 28 March filing date (EDGAR index: Filing Date 2024-03-28, accepted 2024-03-27 17:52) and added the 27 March signature date. The reviewer's 'unverifiable' is wrong. https://www.sec.gov/Archives/edgar/data/1826671/000121390024026702/ea0202622-1512g_nearintel.htm https://www.sec.gov/Archives/edgar/data/1826671/000121390024026702/0001213900-24-026702-index.htm | supported with qualification | ... a Form 15-12G deregistration, signed 27 March 2024, was filed with the SEC on 28 March 2024 (EDGAR filing date; accepted the evening of 27 March). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-D-F07; data/claims-overlay.json |
| R2-21 | Foursquare announced its acquisition of Placed from Snap on 30 May 2019 with terms undisclosed; Snap's 10-Q records that it divested its membership interest in Placed, LLC to Foursquare Labs, Inc. in June 2019 for total cash of US$77.8 million, of which US$66.9 million was purchase consideration, recognising a US$39.9 million gain; the event is an acquisition (completed June 2019). | Retain. Foursquare’s dated 30 May 2019 announcement describes signing, while Snap’s filing establishes June completion, $77.8m total cash, $66.9m purchase consideration and $39.9m gain. Cite the original Foursquare announcement in place of trade reporting for announcement timing. | Reviewer opened: https://www.sec.gov/Archives/edgar/data/1564408/000156459019037449/R12.htm https://foursquare.com/resources/blog/news/foursquare-to-acquire-placed-from-snap-inc-to-deepen-its-location-technology-platform/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-D-F10; data/claims-overlay.json |
| R2-22 | ZeroToOne.AI announced on 25 March 2026 that it had acquired GroundTruth (terms and seller undisclosed), GroundTruth's site now describes itself as 'a ZeroToOne.AI company', and the acquirer's counsel confirmed the transaction on 24 April 2026; the event is an acquisition, announced as completed. | Retain “acquisition, announced as completed”: the company release dated 25 March 2026 and acquirer counsel’s 24 April release support this classification. Neither opened source supplies purchase terms or the seller. | Reviewer opened: https://www.groundtruth.com/article/zerotooneai-acquires-groundtruth/ https://www.nelsonmullins.com/insights/news/press_releases/nelson-mullins-advises-zerotoone-ai-in-acquisition-of-groundtruth | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-D-F15; data/claims-overlay.json |
| R2-23 | The FTC's January 2024 complaint describes InMarket Media, LLC as a Delaware LLC based in Austin, Texas, whose SDK was incorporated into more than 300 third-party apps downloaded onto over 390 million unique devices since 2017 (its own apps onto over 30 million), and alleges that from 2016 'about 100 million unique devices sent Respondent location data each year'. | Retain, attributed to the FTC complaint as allegations. The 390m third-party-app downloads and 30m own-app downloads are cumulative since 2017; the approximately 100m figure is annual devices, not monthly active devices or a current 2026 panel. | Reviewer opened: https://www.ftc.gov/system/files/ftc_gov/pdf/Complaint-InMarketMediaLLC.pdf | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-D-F19; data/claims-overlay.json |
| R2-24 | In the filings, official statements and vendor pages opened by this review, no US location vendor discloses a US panel series on a constant definition that spans a pre-2021 figure and a post-2023 figure, and no filing or official statement attributes a supply change to a named Apple or Google permission change; the dated figures found use incompatible units (monthly devices, opted-in panel, devices worldwide, unique devices per year, unique user IDs, daily signals, active devices), so H1 is unresolved on workstream D's evidence. | The opened records do not establish a comparable pre-2021/post-2023 US panel series: SafeGraph gives a July 2020 carrier snapshot, Placer a September 2024 presentation with active-device figures lacking a stated observation period, InMarket annual devices, and Gravy daily signals/devices. Keep H1 unresolved on that evidence. The broader absence claim across all filings and vendor pages was not independently reproduced; retain an explicit screened-source scope. No market-wide contraction or substitution follows from these figures. | Source check G2 (24 October 2026, primary sources opened): R2-24 accepted with one addition. All four unit descriptions are confirmed, and the Utah deck carries two different Placer device counts (30M active and 25M+), neither with a period. https://www.utah.gov/pmn/files/1194509.pdf https://docs.safegraph.com/docs/monthly-patterns https://www.ftc.gov/system/files/ftc_gov/pdf/Complaint-InMarketMediaLLC.pdf https://www.ftc.gov/news-events/news/press-releases/2024/12/ftc-takes-action-against-gravy-analytics-venntel-unlawfully-selling-location-data-tracking-consumers | supported with qualification | In the filings, official statements and vendor pages this review opened, no US location vendor discloses a US panel series on one definition spanning a pre-2021 and a post-2023 figure. The dated figures use incompatible units: SafeGraph's July 2020 carrier breakdown, InMarket's 'about 100 million unique devices' a year, Gravy's daily signals from about a billion devices, and Placer's undated '30M Active Devices' and '25M+ devices' in a September 2024 deck. H1 is therefore unresolved on this evidence. The absence holds only for the sources screened, and none of these figures shows a market-wide contraction or substitution. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-D-F41; data/claims-overlay.json |
| R2-25 | GroundTruth's Observed Visits metric is an observed, exposed-only count with no control group: a visit is a user visiting the brand or drive-to location within 14 days of exposure, credited last-touch, verified after 7 days, and the help centre states that observed visits may be lower than actual visits because not all users are trackable. | The stated fourteen-day window, seven-day verification and incomplete observation caveat are supported. This help page describes an exposed-only visit metric and specifies no control comparison. Do not turn that bounded observation into proof that every GroundTruth measurement product lacks a control, or count Observed Visits as a falsely labelled lift metric without evidence of that label. | Reviewer opened: https://help.groundtruth.com/hc/en-us/articles/360000695267-Observed-Visits https://help.groundtruth.com/hc/en-us/articles/12770162930707-Projected-Visits | supported with qualification | The stated fourteen-day window, seven-day verification and incomplete observation caveat are supported. This help page describes an exposed-only visit metric and specifies no control comparison. Do not turn that bounded observation into proof that every GroundTruth measurement product lacks a control, or count Observed Visits as a falsely labelled lift metric without evidence of that label. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F10; data/claims-overlay.json |
| R2-26 | GroundTruth's Projected Visits is a panel-based probabilistic projection: a panel of impressed devices with frequent location signals is built per campaign and its visitation applied to all impressed devices, anchored in Observed Visits including first-party WeatherBug data, shown only once a campaign reaches about 40 Observed Visits and 350,000 impressions, cumulative, with no uncertainty reported. | The campaign-specific impressed-device panel, WeatherBug input, usual thresholds of forty observed visits and 350,000 impressions, and cumulative reporting are supported. Replace 'with no uncertainty reported' with 'this help page publishes no uncertainty interval'; absence from a help article does not establish absence from all delivered reports. The page says panels refresh daily. | Reviewer opened: https://help.groundtruth.com/hc/en-us/articles/12770162930707-Projected-Visits | supported with qualification | The campaign-specific impressed-device panel, WeatherBug input, usual thresholds of forty observed visits and 350,000 impressions, and cumulative reporting are supported. Replace 'with no uncertainty reported' with 'this help page publishes no uncertainty interval'; absence from a help article does not establish absence from all delivered reports. The page says panels refresh daily. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F11; data/claims-overlay.json |
| R2-27 | Blis's Smart Holdout Groups, released globally on 10 October 2024, is a geographic holdout design: machine learning splits an audience into statistically balanced target and control zip codes on demographics and income, allows for commuting between zip codes, and inserts buffer zones where ads are neither served nor measured; footfall lift is 'Only available in the US', and the page documents no data source, visit definition, minimum size or significance rule. | Blis's own release dates its global-release announcement to 9 October 2024, not 10 October. The product page supports machine-balanced geographic target/control groups, commuting adjustments, buffer zones and US-only footfall lift. The verifier correctly says randomization is not stated; neither source establishes that this design is the default for all Blis visit-lift reporting. Use the primary release rather than its next-day syndication to date the announcement. | Source check G2 (24 October 2026, primary sources opened): R2-27 accepted: Blis's own release is dated 9 October 2024. The 10 October date came from a MarTech Series syndication. https://blis.com/blis-rolls-out-omnichannel-measurement-solution-into-broader-market/ https://blis.com/omnichannel-measurement/ | contradicted | Blis announced the global release of Smart Holdout Groups on 9 October 2024. It is a geographic holdout design: machine learning splits an audience into statistically balanced target and control zip codes on demographics and income, allows for commuting, and inserts buffer zones where ads are neither served nor measured. Blis does not describe the assignment as randomised and does not say the design is the default for all its visit-lift reporting. Footfall lift is 'Only available in the US', and the page documents no data source, visit definition, minimum size or significance rule. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F14; data/claims-overlay.json |
| R2-28 | Azira's technical white paper documents a randomised PSA-based control as 'the default approach' (typically 10 to 15 per cent of impressions served public-service content), a synthetic control weighted to Census block groups when PSAs are not used, 99 per cent confidence intervals, a two-tailed t-test, minimum samples of about 100,000 exposed devices and 1,000 observed visits, and attribution windows of 7 days or 14 to 30 days by category. | Accept the verifier's qualification: sample minimums apply to lift breakouts, and attribution windows are configurable. Retain vendor-assertion status and do not derive a publication date from the copyright year. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology | supported with qualification | Accept the verifier's qualification: sample minimums apply to lift breakouts, and attribution windows are configurable. Retain vendor-assertion status and do not derive a publication date from the copyright year. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F16; data/claims-overlay.json |
| R2-29 | Azira's technical white paper documents a randomised PSA-based control as 'the default approach' (typically 10 to 15 per cent of impressions served public-service content), a synthetic control weighted to Census block groups when PSAs are not used, 99 per cent confidence intervals, a two-tailed t-test, minimum samples of about 100,000 exposed devices and 1,000 observed visits, and attribution windows of 7 days or 14 to 30 days by category. | Separate mathematical caution: the paper defines lift as a relative rate increase, then multiplies it directly by exposed devices. Under that definition incremental visits require the control visit rate too, or the absolute rate difference times exposed devices. Confirm the vendor's notation before endorsing its formula. | Source check G2 (24 October 2026, primary sources opened): R2-29 accepted. The paper defines lift as a fraction over the control rate, then multiplies it by exposed devices, so its incremental-visits formula is dimensionally wrong as written. https://business.azira.com/azira-location-measurement-and-attribution-methodology https://business.azira.com/azira-location-measurement-and-attribution-methodology | supported with qualification | Azira's white paper defines lift as (exposed rate - control rate) / control rate, but then computes incremental visits as 'Exposed Group Lift * Exposed Group Devices'. Read literally, that multiplies a relative lift by a device count. Under the paper's own definition, incremental visits are (exposed rate - control rate) x exposed devices. Do not reproduce or endorse the published formula without that correction. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F16; data/claims-overlay.json |
| R2-30 | Across the fourteen cohort A seed products, nine document a default control design: two randomised or holdout (Azira's PSA control; Blis's zip-code holdout), six matched or synthetic (Foursquare, InMarket LCI, Reveal, Adsquare, Cuebiq, Arrivalist) and one exposed-only (GroundTruth); five document none (Unacast, Lifesight, Vistar, Kochava, Precisely). Seven of nine documented defaults are matched, synthetic or exposed-only, which meets H3's 'contradicted' rule as coded by this stream. | Do not publish nine documented defaults or two randomized defaults from these records. Blis documents an available, algorithmically balanced geographic holdout; default use is not established. Its nonrandom assignment does not exclude it from the preregistered 'randomised or holdout' category if default use is later established. The verifier's proposed eight-or-nine-of-nine stricter coding changes the registered rule. InMarket's method could not be reopened here. GroundTruth's exposed visit count must be kept distinct from lift. Rebuild the exact count at current product/metric level; the opened Foursquare, Reveal, Adsquare, Cuebiq and Arrivalist materials do describe modeled/matched methods. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology https://blis.com/omnichannel-measurement/ https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://revealmobile.com/solutions/measurement https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://cuebiq.com/blog/ctv-ltv-attribution/ https://www.arrivalist.com/media-insights https://help.groundtruth.com/hc/en-us/articles/360000695267-Observed-Visits https://lciadminconsole.inmarket.com/ | contradicted | Do not publish nine documented defaults or two randomized defaults from these records. Blis documents an available, algorithmically balanced geographic holdout; default use is not established. Its nonrandom assignment does not exclude it from the preregistered 'randomised or holdout' category if default use is later established. The verifier's proposed eight-or-nine-of-nine stricter coding changes the registered rule. InMarket's method could not be reopened here. GroundTruth's exposed visit count must be kept distinct from lift. Rebuild the exact count at current product/metric level; the opened Foursquare, Reveal, Adsquare, Cuebiq and Arrivalist materials do describe modeled/matched methods. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F28; data/claims-overlay.json |
| R2-31 | Under the Q1 coding rule, the visit numbers sold by cohort A products are documented as projected for Foursquare Attribution, InMarket LCI (projected store visits), GroundTruth Projected Visits and Azira (extrapolated total visits); as observed for GroundTruth Observed Visits and InMarket's exposed device visits; as declared for Arrivalist MediaInsights (card transactions); and as 'not disclosed' for Blis, Reveal, Adsquare, Cuebiq, Vistar, Kochava, Lifesight and Precisely. | Accept the central verifier correction: Arrivalist's card-based spend metric is declared, but this does not establish that its arrival counts are transactions. Its Media Insights page lists location pings as well as card data. Foursquare and GroundTruth describe projected counts; retain Azira's observed/projected distinction. InMarket's specific formula could not be reopened behind sign-in. Code each metric separately rather than assigning one basis to a product that reports visits and sales. | Reviewer opened: https://foursquare.com/resources/blog/ad-tech/location-foursquares-attribution-methodology-takes-your-media-dollars-further/ https://help.groundtruth.com/hc/en-us/articles/12770162930707-Projected-Visits https://business.azira.com/azira-location-measurement-and-attribution-methodology https://www.arrivalist.com/media-insights https://lciadminconsole.inmarket.com/ | contradicted | Accept the central verifier correction: Arrivalist's card-based spend metric is declared, but this does not establish that its arrival counts are transactions. Its Media Insights page lists location pings as well as card data. Foursquare and GroundTruth describe projected counts; retain Azira's observed/projected distinction. InMarket's specific formula could not be reopened behind sign-in. Code each metric separately rather than assigning one basis to a product that reports visits and sales. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F29; data/claims-overlay.json |
| R2-32 | GroundTruth Ads Manager (live, US, $0 minimum spend) sells four location-targeting layers documented in its help centre as of 23 September 2026: Target Locations (city to precise business address, lat/long or nationwide), Target by Audience (historic visitation, with Deterministic first-party, Predictive and Datonics 'Premium' tiers), Target Neighborhoods (trade-area lookalikes) and Target On-Premise (real-time, mobile only, with In Store, On Lot, Retail Block and Radial boundaries); the platform calls geofencing 'Location Targeting'. | Retain as vendor-documented functionality. The help article dated 23 September 2026 supports the geography plus three optional layers, the Datonics premium tier and four on-premise boundaries. The product page advertises $0 minimum spend. These are product claims, not independently verified targeting accuracy. | Reviewer opened: https://help.groundtruth.com/hc/en-us/articles/53523927168659-Location-Audience-Targeting https://www.groundtruth.com/solutions/ads-manager/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F02; data/claims-overlay.json |
| R2-33 | The FTC's final order against Mobilewalla (14 January 2025, vote 4-1) bans the sale of sensitive location data, bans collecting consumer data from real-time bidding exchanges 'for purposes other than participating in those auctions', and bars using, transferring, selling or disclosing location data from sensitive locations 'such as health clinics, religious organizations, correctional facilities'; it is the only record in this stream that documents bidstream as a location source for a cohort B vendor. | State: the order was issued 13 January 2025 and its finalization was publicly announced 14 January (4–1 vote). It restricts auction-derived collection/retention to auction participation and prohibits sensitive-location use or disclosure under its definitions and conversion proviso. Accept the verifier’s change from “only record” to “only regulator record in this stream”: Azira’s current data page separately names bid-stream as a source. | Source check G4 (24 October 2026, primary sources opened): R2-33 accepted: Mobilewalla D&O issued 13 Jan 2025 (release 14 Jan, 4-1); Provision II auction rule and Provision III conversion proviso verified at the order PDF; "only record" narrowed to "only regulator record" (Azira names bid-stream). https://www.ftc.gov/system/files/ftc_gov/pdf/Mobilewalla-D&O.pdf https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-finalizes-order-banning-mobilewalla-selling-sensitive-location-data https://www.ftc.gov/system/files/ftc_gov/pdf/Mobilewalla-D&O.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Mobilewalla-D&O.pdf | supported with qualification | The FTC's Mobilewalla Decision and Order (Docket C-4811) was issued 13 January 2025 and its finalisation announced 14 January 2025 on a 4-1 vote; it bars Mobilewalla from collecting or retaining information it accesses in online advertising auctions for any purpose other than participating in them, and from selling, sharing, disclosing or otherwise using Sensitive Location Data for the sensitive locations it must identify, except to convert that data into non-sensitive or non-location data. It is the only regulator record in this stream that documents bidstream as a cohort B location source; Azira's own data page also names bid-stream among its sources. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F10; data/claims-overlay.json |
| R2-34 | Vistar Media's DOOH audience targeting (the product URL returned 404 on 24 October 2026) is documented in its own guide as screen selection from 'aggregated and anonymized location data' plus third-party audience partners (Epsilon with 'over 3,000 readily-available audiences', Eyeota, InMarket, Affinity Answers), and since 5 October 2026 T-Mobile segments 'built from app ownership data, from devices within T-Mobile's network', which trade reporting says are exclusive to Vistar's DSP 'because of tightening regulations around location data and the use of mobile device IDs for ad targeting'. | Accept the verifier’s qualification. Vistar documents aggregated movement/location targeting and the named partners. The opened 5 October 2026 report describes both mobile location data and app installation/use inferred from network requests, with OOH-proximity exposure for users opting into location tracking. Keep those distinct inputs separate. Exclusivity and the regulatory rationale are attributed to T-Mobile’s executive, not an independent legal conclusion. A past 404 is an access observation, not evidence of product retirement. | Source check G3 (24 October 2026, primary sources opened): LOC-F-F12: kept location and app-usage inputs separate, attributed the regulatory rationale to Zhu, recorded the 404 as an access note only; AdExchanger's 'January 2025' Vistar date conflicts with the 10-K (3 Feb 2025). https://www.vistarmedia.com/blog/vistar-media-dooh-audience-targeting https://www.adexchanger.com/digital-out-of-home/t-mobile-brings-its-mobile-data-exclusively-to-vistar-to-scale-up-dooh-targeting/ https://www.adexchanger.com/digital-out-of-home/t-mobile-brings-its-mobile-data-exclusively-to-vistar-to-scale-up-dooh-targeting/ https://www.vistarmedia.com/blog/vistar-media-dooh-audience-targeting | supported with qualification | Vistar Media's own guide documents DOOH audience targeting from 'aggregated and anonymized location data' plus third-party partners (Epsilon, Eyeota, InMarket, Affinity Answers). Since 5 October 2026 T-Mobile's data has been available for OOH targeting exclusively through Vistar; per AdExchanger it combines mobile location data with app installation and usage that T-Mobile infers from network requests, with OOH-proximity exposure only for users who opted into location tracking. T-Mobile's Jess Zhu attributes the exclusivity to tightening regulation of location data and mobile device IDs. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F12; data/claims-overlay.json |
| R2-35 | Broadsign Ads' DOOH audience targeting (found in discovery; not in the register) is the only DOOH audience product opened whose documentation names both the mobility supplier and the collection route: Native Touch data 'pulled from active application SDKs and APIs', collected 'from smartphones once audiences enter the defined location boundaries', including 'device IDs, languages, app IDs, age, gender', translated into 6 targeting groups and over 190 personas per screen; no panel size, consent route or region is stated. | Replace the failed old URL with the opened /audiences/ page. The page supports Native Touch, SDK/API collection inside venue boundaries, the listed device attributes, six groups and more than 190 personas. It states no panel period/size or US coverage on that page. Keep the discovery as a cohort-B candidate; do not score it as US-available until a source establishes US availability. “Only opened DOOH product” is limited to the original screen, not the whole market. | Reviewer opened: https://broadsign.com/audiences/ | supported with qualification | Replace the failed old URL with the opened /audiences/ page. The page supports Native Touch, SDK/API collection inside venue boundaries, the listed device attributes, six groups and more than 190 personas. It states no panel period/size or US coverage on that page. Keep the discovery as a cohort-B candidate; do not score it as US-available until a source establishes US availability. “Only opened DOOH product” is limited to the original screen, not the whole market. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F17; data/claims-overlay.json |
| R2-36 | None of the cohort B product documentation opened by 24 October 2026 describes suppression of sensitive places (health facilities, places of worship, shelters or the categories named in FTC orders and state laws); the only sensitive-location handling on record for these vendors is imposed by FTC orders (InMarket 2024, Mobilewalla 2025, Kochava and Collective Data Solutions 2026), GroundTruth's policy states it may sell sensitive personal data, and Ads Manager's one documented category control is an alcohol age gate. | The internal verifier is right that the original “only handling on record is imposed by FTC orders” is false. Foursquare publishes sensitive-place restrictions, Mobilewalla disclaims knowing collection at sensitive sites, and Azira describes sensitive-location removal. Describe these as public policy statements, not verified enforcement. GroundTruth’s sensitive-personal-data sale notice is present but does not establish sale of sensitive-place visits. Retain any absence claim only for the exact product pages originally screened. | Source check G2 (24 October 2026, primary sources opened): R2-36 accepted. All four privacy pages say what the reviewer reports, so 'only FTC-imposed handling' is false. GroundTruth's notice is the generic state-law sensitive-data notice. https://foursquare.com/legal/privacy-center/foursquare-privacy-policy-consumer-services/ https://www.mobilewalla.com/privacy-center https://www.azira.com/platform/privacy https://www.groundtruth.com/privacy-policy/ | contradicted | None of the cohort B product pages opened describes sensitive-place suppression, but vendor privacy documents do. Foursquare commits to the NAI enhanced standard and lists 13 sensitive-place categories. Mobilewalla says it does not knowingly collect visits to places of worship, shelters, treatment centres or reproductive health clinics. Azira describes a sensitive-location data program that removes such data. These are self-attested policy statements with no verified enforcement, alongside FTC orders imposing programs on InMarket, Mobilewalla, and Kochava with Collective Data Solutions. GroundTruth's notice that it 'may sell your sensitive personal data' is a state-law disclosure and does not show that it sells sensitive-place visits. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F19; data/claims-overlay.json |
| R2-37 | H4 is unresolved on the brief's rule: of the fourteen cohort B seed products, twelve were readable and only one (GroundTruth Ads Manager, via its privacy policy) documents its location sources, as a blend without proportions; eleven are coded 'not disclosed' from current documentation (opt-in, consented, proprietary, GPS or mobility signals without a route) and two could not be read (Blis bot-blocked; Kochava Collective absent from its URL), so fewer than four products document a primary source and neither the bidstream/IP count (0) nor the SDK/panel count (0) can be compared. | The original source-disclosure count is wrong, but the verifier’s replacement is also too strong. Foursquare describes SDK/API partners AND suppliers/brokers/ad networks providing location data. InMarket’s location-data page lists its own apps, partner SDK network AND trusted third-party sources. Neither proves the primary location source or proportions. Code both “multiple disclosed sources; primary source not established,” unless product-specific method documentation settles dominance. Azira expressly discloses an SDK/bid-stream/transaction blend. H4 remains unresolved; rebuild counts from primary-source evidence and the revised register rather than reusing 0/0 or the verifier’s two SDK votes. | Source check G2 (24 October 2026, primary sources opened): R2-37 accepted. InMarket's data page lists a third item, 'Trusted & certified third-party sources', and Foursquare's policy names brokers and ad networks supplying location data, so neither vendor can be coded as SDK-primary. https://foursquare.com/legal/privacy-center/foursquare-privacy-policy-consumer-services/ https://inmarket.com/data/ https://www.azira.com/platform/data | contradicted | H4 remains unresolved, and the cohort B source counts must be rebuilt. Foursquare (SDK/API partners plus data suppliers, brokers and ad networks) and InMarket (own app, partner SDK network and 'trusted & certified third-party sources') disclose multiple sources without naming a primary one, so code both 'multiple disclosed sources; primary source not established'. Azira (SDK, bid-stream and transactions) and GroundTruth disclose blends without proportions. No product documents a primary source, so neither the earlier 0/0 count nor the verifier's two SDK votes stands. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F20; data/claims-overlay.json |
| R2-38 | iOS 14 (announced 22 June 2020, released 16 September 2020) let users grant apps approximate instead of precise location; Apple's WWDC20 session states apps cannot opt out, that reduced-accuracy fixes are quantised to a region with a typical radius of about five kilometres (a couple of kilometres in dense cities, ten kilometres or more in rural areas), recomputed about four times per hour, and that region monitoring requires full accuracy. | Retain. Apple’s dated announcement/release and WWDC transcript support user-controlled reduced accuracy, variable quantization region size, approximately four updates per hour, no app opt-out and full-accuracy requirement for region notifications. These describe permission behavior, not an observed vendor supply decline. | Reviewer opened: https://www.apple.com/newsroom/2020/06/apple-reimagines-the-iphone-experience-with-ios-14/ https://www.apple.com/newsroom/2020/09/ios-14-is-available-today/ https://developer.apple.com/videos/play/wwdc2020/10660/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-G-F02; data/claims-overlay.json |
| R2-39 | App Tracking Transparency was announced on 22 June 2020 as an iOS 14 requirement, deferred on 3 September 2020 to 'early next year', and became mandatory for all apps on 26 April 2021 with iOS 14.5; Apple's definition of tracking includes sharing user or device data with data brokers. | Retain the announcement/deferment/enforcement sequence and broker-sharing scope. Keep ATT permission separate from Core Location permission; ATT does not itself remove an app’s location entitlement. | Reviewer opened: https://www.apple.com/newsroom/2020/06/apple-reimagines-the-iphone-experience-with-ios-14/ https://developer.apple.com/news/?id=hx9s63c5 https://developer.apple.com/news/?id=ecvrtzt2 https://www.apple.com/newsroom/2021/04/ios-14-5-offers-unlock-iphone-with-apple-watch-diverse-siri-voices-and-more/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-G-F03; data/claims-overlay.json |
| R2-40 | iCloud Private Relay, released as a public beta with iOS 15 on 20 September 2021 for iCloud+ subscribers, replaces the client IP with a relay IP that maps to a roughly city-level area by default ('Maintain general location', a four-character geohash of roughly 800 square kilometres) or to country and time zone, and covers Safari browsing, DNS queries and unencrypted (insecure HTTP) app traffic only, leaving encrypted in-app traffic and its IP geolocation untouched. | Accept the verifier’s qualification: the four-character geohash is the location hint for selecting the relay IP, not a guarantee that every relay IP maps to that exact area. Private Relay requires an eligible user to enable it, preserves coarse location by default and covers Safari, DNS and unencrypted app connections; do not imply that it masks HTTPS in-app transport or changes Core Location permissions. The 20 September 2021 release/public-beta timing is supported. | Reviewer opened: https://www.apple.com/newsroom/2021/09/ios-15-is-available-today/ https://www.apple.com/privacy/docs/iCloud_Private_Relay_Overview_Dec2021.PDF https://developer.apple.com/icloud/prepare-your-network-for-icloud-private-relay/ | supported with qualification | Accept the verifier’s qualification: the four-character geohash is the location hint for selecting the relay IP, not a guarantee that every relay IP maps to that exact area. Private Relay requires an eligible user to enable it, preserves coarse location by default and covers Safari, DNS and unencrypted app connections; do not imply that it masks HTTPS in-app transport or changes Core Location permissions. The 20 September 2021 release/public-beta timing is supported. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-G-F06; data/claims-overlay.json |
| R2-41 | Android 12 (source pushed to AOSP on 4 October 2021) lets users grant only approximate location even when an app requests precise location, offers Precise and Approximate options in the permission dialog, lets users set the accuracy for any app in Settings regardless of its target SDK, and Google documents approximate location as accurate to within about 3 square kilometres. | Retain, with the scope already supplied by Google: the approximate estimate applies to LocationManagerService/FusedLocationProvider. The 4 October 2021 AOSP date, approximate/precise choice and Settings control regardless of target SDK are supported. This is an approximate area, not a 3km radius. | Reviewer opened: https://android-developers.googleblog.com/2021/10/android-12-is-live-in-aosp.html https://developer.android.com/develop/sensors-and-location/location/permissions/runtime https://developer.android.com/develop/sensors-and-location/location/permissions | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-G-F09; data/claims-overlay.json |
| R2-42 | On 17 October 2025 Google announced it would retire the Attribution Reporting API, Protected Audience and Topics on both Chrome and Android, together with Protected App Signals, SDK Runtime, IP Protection, On-Device Personalization, Private Aggregation, Related Website Sets and SelectURL, with phase-outs to follow Chrome and Android processes and no retirement dates given; CHIPS, FedCM and Private State Tokens continue. | Retain the named retirement decisions and continuing CHIPS/FedCM/Private State Tokens. Attribute “no retirement dates” to this 17 October 2025 announcement, which says phaseout will follow Chrome/Android processes; it is not a claim that no later implementation dates exist. | Reviewer opened: https://privacysandbox.google.com/blog/update-on-plans-for-privacy-sandbox-technologies | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-G-F15; data/claims-overlay.json |
| R2-43 | In the sources screened by the cutoff, this review found no filing, official platform statement or independent measurement that attributes a quantified change in US SDK-sourced location supply to a named Apple or Google permission change; the Telenav FY2019 and FY2020 10-Ks do not mention iOS 13 or location permissions, the Near S-4/A attributes no figure, and EDGAR full-text searches for 10-Ks pairing location data with App Tracking Transparency, iOS 13 or background location returned no location vendor. | The opened Telenav FY2020 filing yielded no matches for “iOS 13” or “location permissions”; Google’s policy page states policy changes, not quantified US vendor-panel effects. The cited Near S-4/A was inaccessible, and the original EDGAR search inventory was not reproduced. The verifier correctly narrows the absence, but this review cannot support the entire negative claim. Keep H1 unresolved and record the unreviewed filings/searches explicitly. | Reviewer opened: https://www.sec.gov/Archives/edgar/data/1474439/000147443920000029/tnav630202010k.htm https://android-developers.googleblog.com/2020/02/safer-location-access.html | supported with qualification | The opened Telenav FY2020 filing yielded no matches for “iOS 13” or “location permissions”; Google’s policy page states policy changes, not quantified US vendor-panel effects. The cited Near S-4/A was inaccessible, and the original EDGAR search inventory was not reproduced. The verifier correctly narrows the absence, but this review cannot support the entire negative claim. Keep H1 unresolved and record the unreviewed filings/searches explicitly. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-G-F20; data/claims-overlay.json |
| R2-44 | The InMarket order (Docket C-4803, proposed package dated 18 January 2024, issued 29 April 2024, finalisation announced 1 May 2024 by a 3-0-2 vote) is the only one of the five that bans selling or licensing precise Location Data outright ('in exchange for any valuable consideration'); it separately bars products that categorise or target consumers by sensitive location data across nine categories, requires deletion of Historic Location Data within 90 days and audience segments within 120 days, and runs 20 years. | The sale prohibition and dates are supported, but the verifier missed material remediation qualifications. XII.A requires deletion of historic data collected through InMarket apps within 90 days, with an option to obtain affirmative consent for retention. XII.B gives 120 days to delete, deidentify or render non-sensitive third-party historic data. XII.C requires deletion of segments made from historic data within 120 days. Replace the blanket 90-day deletion statement with these distinct duties. Historic data means data obtained without affirmative express consent. | Source check G4 (24 October 2026, primary sources opened): R2-44 accepted: InMarket XII split verified at the order PDF (A: own-app data 90 days + consent option; B: third-party data 120 days delete/deidentify/render non-sensitive; C: segments 120 days); blanket "delete within 90 days" was wrong. https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf | supported with qualification | The InMarket order (C-4803, issued 29 April 2024) bans selling or licensing Location Data and bars sensitive-location products; for Historic Location Data (data collected without affirmative express consent) it requires deletion within 90 days of data collected through InMarket's own apps unless the consumer gives affirmative express consent to retention, deletion, deidentification or rendering non-sensitive within 120 days of data obtained from third parties, and deletion of audience segments built from historic data within 120 days. The order runs 20 years. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F03; data/claims-overlay.json |
| R2-45 | FTC v. Kochava (D. Idaho, No. 2:22-cv-00377-BLW) was litigated for almost four years: complaint filed 29 August 2022; first motion to dismiss granted with leave to amend 4 May 2023 (Dkt. 24); First Amended Complaint 5 June 2023; motion to dismiss it denied 3 February 2024 (Dkt. 71); Kochava transferred its data-broker business to subsidiary Collective Data Solutions, LLC, so a Second Amended Complaint naming CDS was filed 15 July 2024 (Dkt. 86); motion to dismiss it denied 3 February 2025 (Dkt. 101); a stipulated order was announced 4 May 2026 (Commission vote 2-0) and entered 25 June 2026 (D… | The opened docket decisions support the dismissal and amendment history, the transfer of the broker business to CDS, and June 25, 2026 entry of the stipulated order. Preserve the distinction between FTC approval/announcement and court entry. No correction to the chronology identified. | Reviewer opened: https://www.ftc.gov/system/files/ftc_gov/pdf/1.%20Complaint.pdf https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-0.pdf https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-2.pdf https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-3.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf https://www.ftc.gov/news-events/news/press-releases/2026/05/ftc-ban-kochava-subsidiary-selling-sensitive-location-data-settle-charges-they-sold-location-data | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F06; data/claims-overlay.json |
| R2-46 | The Kochava stipulated order bans Kochava and Collective Data Solutions from selling, licensing, transferring, sharing or disclosing Sensitive Location Data (Precise Location Data tied to five categories of Sensitive Locations, narrower than the administrative orders), with the same direct-relationship consent exception, a 1,850-foot coarse carve-out, a Supplier Assessment Program, third-party incident reports within 30 days, deidentification (not deletion) of Historical Location Data within 90 days, a deletion-request mechanism and a 10-year term; defendants neither admit nor deny, and sale … | The central prohibition is supported. Amend historical remediation to deidentify OR render non-sensitive within 90 days; XII also allows retention where records confirming consent are obtained within that period. The definition of precise data additionally excludes specified security, federal national-security and imminent-harm uses. Record these defined exclusions when comparing orders. The verifier missed these qualifications. | Source check G4 (24 October 2026, primary sources opened): R2-46 accepted with changes: Kochava XII.A "deidentify or render non-sensitive" + consent-records retention proviso and Def. G exclusions verified; reviewer overstates that the verifier missed them (verifier quoted XII.A and the exclusions; only the retention proviso was missing). https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf | supported with qualification | The Kochava order (entered 25 June 2026, 10-year term) requires defendants within 90 days to deidentify or render non-sensitive all Historical Location Data, unless they obtain within that period records showing consumers consented to its collection, use and disclosure. Its definition of Precise Location Data excludes coarse data (radius of at least 1,850 feet) and data used for security purposes, federal national-security purposes, or a law enforcement response to an imminent risk of death or serious bodily harm. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F07; data/claims-overlay.json |
| R2-47 | Gravy Analytics (a Unacast subsidiary) told the Norwegian Data Protection Authority that on 4 January 2025 it identified unauthorised access to its AWS cloud storage, apparently through a misappropriated access key, that the intruder 'obtained some files' whose contents and personal-data status were under investigation, that it took its data-processing platform offline and restored it by 9 January 2025, and that it was also informing the UK ICO; this is a security incident, not an enforcement action. | The cited DocumentCloud PDF and its viewer/asset alternatives returned inaccessible or HTTP 403 responses, including a direct download attempt. I could not independently check the notice, its dates, the intruder-notification account or the verifier correction. Keep unresolved pending an accessible copy of the primary notice. | Reviewer opened: | supported with qualification | The cited DocumentCloud PDF and its viewer/asset alternatives returned inaccessible or HTTP 403 responses, including a direct download attempt. I could not independently check the notice, its dates, the intruder-notification account or the verifier correction. Keep unresolved pending an accessible copy of the primary notice. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F08; data/claims-overlay.json |
| R2-48 | Maryland's Online Data Privacy Act of 2024 (Chapter 455, SB 541, approved 9 May 2024, effective 1 October 2025) prohibits a controller from selling sensitive data, which includes precise geolocation data defined by a 1,750-foot radius, making it a flat ban on sale of precise geolocation; it also bars geofences within 1,750 feet of mental health or reproductive or sexual health facilities for consumer-health-data purposes. | The sensitive-data sale prohibition and 1,750-foot health-facility geofence restriction are supported, subject to statutory coverage and exemptions. The verifier correctly distinguishes the any-person geofence provision. Its April 1, 2026 note belongs specifically to section 14-4612, as the chapter says; do not recast it as a general postponement of the entire law or sale prohibition. | Source check G4 (24 October 2026, primary sources opened): R2-48 accepted: MD ch. 455 Section 2 ties the 1 April 2026 date to 14-4612 alone (the exemptions section in the enacted text); verifier wording already correct; keep "subject to coverage and exemptions". https://mgaleg.maryland.gov/2024RS/Chapters_noln/CH_455_sb0541e.pdf https://mgaleg.maryland.gov/2024RS/Chapters_noln/CH_455_sb0541e.pdf https://mgaleg.maryland.gov/2024RS/Chapters_noln/CH_455_sb0541e.pdf | supported with qualification | Maryland's Online Data Privacy Act (2024 ch. 455, SB 541, approved 9 May 2024, effective 1 October 2025) bars a controller from selling sensitive data, which includes precise geolocation defined by a 1,750-foot radius, with no consent exception, subject to the act's coverage thresholds and exemptions; it also bars any person from using a geofence within 1,750 feet of a mental health or reproductive or sexual health facility to identify, track, collect data from or notify consumers about their consumer health data. Section 2 of the chapter makes section 14-4612 as enacted apply only prospectively, not to processing before 1 April 2026; that clause does not postpone the act or the sale ban. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F10; data/claims-overlay.json |
| R2-49 | Under the preregistered rule, H5 is contradicted: four of the five FTC-ordered companies (Outlogic; Gravy Analytics as Unacast with Venntel; Mobilewalla; Kochava through Collective Data Solutions) document at the cutoff continued sale or licensing of precise or device-level location data to third parties, in each case alongside sensitive-location suppression language that tracks their orders; InMarket, the only company whose order bans sale of precise location data outright, documents no such product while disclosing CCPA-sense sale or sharing of precise geolocation. | H5 has current marketed counterexamples: Outlogic explicitly identifies precise-coordinate sales and Collective Data Solutions markets precision-location feeds. Unacast currently markets location products and explicitly discloses precise-geolocation sales over the preceding twelve months, a qualified third disclosure rather than conclusive evidence of a precise feed still sold on 24 October. Mobilewalla documents licensing and device-linked coordinates but not the precision or currently licensed fields. Replace the unqualified four-company current precise-seller count with this company-level evidence. Foundry visit reporting is not the proof for CDS raw-data licensing; continued sale alone establishes no order violation. | Source check G5 (24 October 2026, primary sources opened): R2-49: four-company count cut to three documented (Outlogic, CDS, Unacast) plus Mobilewalla unclear; reviewer's Unacast 'qualified' downgrade rejected (policy dated Sep 2026 plus current device-level product); CDS strengthened by its 1 Oct 2026 'resold ... precise geolocation' disclosure. https://outlogic.io/ https://collectivedata.io/ https://www.unacast.com/privacy-policy https://www.mobilewalla.com/california-privacy-notice-2023 | supported with qualification | Under the preregistered rule H5 is contradicted. Three of the five ordered companies document at the cutoff the sale or licensing of precise geolocation to third parties: Outlogic, Collective Data Solutions for Kochava, and Gravy Analytics trading as Unacast, each in its own current policy or product pages. Mobilewalla documents licensing of device-linked lat/long data but not the precision of what it licenses, and InMarket documents no such product. Compliance with the orders is not assessed. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F23; data/claims-overlay.json |
| R2-50 | Geopath's methodology documentation (published 31 May 2023) names its inputs as a proprietary OOH inventory database, roadway data from Teralytics ('Terralytics' on one page) with Streetlytics and HERE Maps, vehicle-travel data from HERE Maps, 'recurring activity patterns from anonymized mobile device panels (from MotionWorks)' and demographics from government agencies, MotionWorks and Claritas; it gives no proportions. | The 31 May 2023 pages support the named historical inputs, including MotionWorks and the roadway suppliers. The verifier is right to date that list: Geopath’s 19 October 2026 announcement names Bentley Systems for roadside data and says the 2026 dataset has updated sources. Describe the list as 2023 documentation; the 2026 planning dataset is announced to become the default on 12 November, after both this review and the proposed freeze. | Source check G1 (24 October 2026, primary sources opened): R2-50: accepted. Label the provider list 'as documented 31 May 2023'. Bentley Systems is the roadside provider per the 19 Oct 2026 post, and the 2026 dataset is the default from 12 Nov (post-cutoff). https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323447855/Geopath+High-Level+Media+Measurement+Process https://blog.geopath.org/2026/10/19/2026-planning-dataset-to-become-default-on-november-12/ https://blog.geopath.org/2026/10/19/2026-planning-dataset-to-become-default-on-november-12/ https://blog.geopath.org/2026/10/19/2026-planning-dataset-to-become-default-on-november-12/ | supported with qualification | Geopath's methodology documentation, published 31 May 2023, names MotionWorks device panels, Teralytics/Streetlytics and HERE Maps for roadways, HERE Maps for vehicle travel, and government agencies, MotionWorks and Claritas for demographics, with no proportions. That list is as documented in 2023: Geopath's 19 October 2026 post names Bentley Systems as its roadside data provider, and its 2026 Planning Dataset, which credits 'updated data sources' for pedestrian-impression changes, becomes the default only on 12 November 2026, after the cutoff. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F01; data/claims-overlay.json |
| R2-51 | Geopath's Mobile Device Panel page (31 May 2023) describes a weekly sample of GPS-sighted devices selected for roughly eight active hours a day and a consistent Census Block Group home, states that about 80% of GPS sightings come from 20% of devices, and quantifies the panel only through one example (week of 9 August 2020: panel devices were 19.5% of devices processed and 78% of locations processed); it discloses no absolute US panel size and names no provider on the page. | The original page supports the weekly selection criteria, the approximate 80/20 distribution and the August 2020 19.5%/78% example, but it explicitly names Standard Data, contrary to the original “names no provider” wording. The verifier is right on that omission. Its broader implication of no published absolute US panel size also needs qualification: a separate 31 May 2023 Geopath page states over 100M unique US devices monthly and a selected sample of 7–10M qualified devices weekly, without a clear observation vintage. Record this as documentation-dated disclosure, not a current audited panel count. The figure’s absolute counts and the separate 2% assertion were not independently checked. | Source check G1 (24 October 2026, primary sources opened): R2-51: accepted. Standard Data is named on the panel page, and page 2323449264 (31 May 2023) gives >100M monthly devices and 7–10M qualified devices weekly. Code the panel size as documentation-dated, vintage unstated. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449407 https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449264/Geopath+Mobile+Device+Location+Movement+Patterns https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449407 | supported with qualification | Geopath's Mobile Device Panel page (31 May 2023) describes a weekly panel of GPS-sighted devices, active about eight hours a day with a consistent Census Block Group home, culled by Standard Data, and illustrates it with a 2020 example (19.5% of devices, 78% of locations). A companion 31 May 2023 page says the feeds cover 'more than 100 million unique devices in the United States every month' and that the sample is 'between seven and ten million qualified devices each week'. That is a documentation-dated disclosure with no stated observation vintage, not a current audited panel count. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F02; data/claims-overlay.json |
| R2-52 | On 31 March 2026 Geopath and the OAAA announced that their measurement modernisation 'began in early 2025', that 'Ipsos has been selected to support an industry pilot launching in the second half of 2026' to 'evaluate a next-generation OOH audience measurement system across multiple formats and environments' 'including updated methods for estimating impressions, reach, and frequency', with 'an orderly transition from the current Geopath measurement system to a next-generation framework beginning in 2027' and 'full adoption in 2028'; the release names no data sources. | Retain as an announced plan: the opened syndicated Geopath/OAAA release is datelined 31 March 2026 and supports the early-2025 initiative, second-half-2026 pilot, anticipated 2027 transition and 2028 adoption. It does not establish that the pilot has actually begun. The verifier’s supported verdict is appropriate. | Reviewer opened: https://martechedge.com/press-release/ooh-industry-advances-effort-to-modernize-audience-measurement-and-selects-ipsos-for-pilot-program | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F05; data/claims-overlay.json |
| R2-53 | OpenRTB 2.6's Qty object carries 'multiplier' (float; required: the quantity of billable events deemed to occur, e.g. 'a DOOH opportunity may be considered to be 14.2 impressions'), 'sourcetype' (integer; recommended) and 'vendor' (string, 'required if sourcetype is present and type = 1': 'The top level business domain name of the measurement vendor providing the quantity measurement'); AdCOM's 'DOOH Multiplier Measurement Source Types' list enumerates 0 Unknown, 1 Measurement Vendor Provided, 2 Publisher Provided and 3 Exchange Provided; neither document records the vendor's mobility source,… | The opened Qty table and AdCOM list support multiplier, recommended sourcetype, conditional vendor domain and source types 0–3. They do not define a standard field for a named provider’s underlying mobility panel or sampling method. Bound the absence claim to these standard objects; vendor extensions remain possible. Pin a release or commit for the evidence freeze because both cited URLs follow main. | Reviewer opened: https://raw.githubusercontent.com/InteractiveAdvertisingBureau/openrtb2.x/main/2.6.md https://raw.githubusercontent.com/InteractiveAdvertisingBureau/AdCOM/main/AdCOM%20v1.0%20FINAL.md | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F09; data/claims-overlay.json |
| R2-54 | Across the US DOOH currency and SSP products screened by the cutoff, only Geopath names its mobility-data providers in its own methodology documentation (MotionWorks device panels; Teralytics and HERE for roadway and vehicle data); Vistar Media documents named partner sources on its own partner pages (InMarket SDK data, Foursquare/Placed); Place Exchange and Broadsign, Perion (Hivestack) and Talon describe 'mobile data', 'anonymised device-level location data' or 'mobility and partner data' without a provider; and no screened product or currency discloses a panel size with a period, so every … | The original 0-panel-disclosed tally is wrong under Q2’s registered size-plus-period rule: Geopath’s 2023 documentation discloses 7–10M qualified devices weekly and over 100M unique US input devices monthly, with observation vintage unstated. Code disclosure present but documentation-dated, with current panel size unverified. The verifier is also right that Vistar’s own role- and region-tagged partner pages document audience-targeting sources, including GroundTruth’s mixed first-party/partner/bidstream supply and Adsquare mobile location data. Do not silently change the denominator: its visible correction splits Vistar into attribution and targeting while dropping Talon from the original six. Publish an explicit per-product row set before recomputing totals; retain Talon or explain its exclusion. The full verifier correction is truncated in the digest, so only its visible proposal could… | Source check G1 (24 October 2026, primary sources opened): R2-54: accept with changes. Recode Geopath to panel disclosed (7–10M weekly, documentation-dated 2023) and Vistar targeting to source disclosed (3/3; 1/5). Reject the claim that Talon was dropped: the original derivation already excluded Talon, and the claim prose should say so. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449264/Geopath+Mobile+Device+Location+Movement+Patterns https://www.vistarmedia.com/partner-groundtruth https://www.vistarmedia.com/partner-groundtruth | supported with qualification | Of six US DOOH products and currencies coded (Geopath; Vistar attribution; Vistar audience targeting; Place Exchange PerView; Broadsign; Perion; Talon is excluded as coded only from a UK-hosted global page), three name their location-data sources in their own or the operator's documentation: Geopath, Vistar attribution (InMarket, Foursquare/Placed) and Vistar audience targeting (Vistar-hosted partner pages naming GroundTruth's first-party, third-party and bid-stream data and Adsquare mobile location data). Place Exchange, Broadsign and Perion do not. Only Geopath states a panel size with a period: 7–10M qualified devices weekly from more than 100M US devices monthly, in 31 May 2023 documentation with no observation vintage stated. Counts: 3 of 6 disclose a source and 1 of 6 a panel size. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F23; data/claims-overlay.json |
| R2-55 | None of the DOOH attribution products or case studies screened documents a randomised or holdout control as its default: Hivestack's documented control is unexposed passers-by at the same screens by playlog timing (UK, 2022), Perion's current page asserts 'a matched control group' without method, Broadsign cites 'non-exposed control groups' without method, and Vistar's InMarket and Foursquare partner pages and GroundTruth's Place Exchange release describe no control; coded under H3 that is 0 randomised or holdout, 1 matched asserted without method (Perion), 2 unexposed comparisons without mat… | Opened sources confirm Perion asserts a matched control, Broadsign’s Netherlands Decathlon case names non-exposed controls, and the opened Vistar and GroundTruth releases give no randomization procedure. These are bounded documentation findings, not proof of each product’s default. The IAB UK Hivestack PDF was inaccessible on repeated opens, so the six-case tally and its UK control method were not independently verified. Preserve the verifier’s geography qualification and mark that component unresolved; do not treat a campaign-specific comparator as a product default. | Reviewer opened: https://perion.com/channels/dooh/ https://broadsign.com/measurement-attribution/ https://www.vistarmedia.com/partner-inmarket https://www.vistarmedia.com/news/foursquare-dooh-measurement-partner https://www.groundtruth.com/article/groundtruth-expands-into-digital-out-of-home-with-measurable-foot-traffic-attribution/ | supported with qualification | Opened sources confirm Perion asserts a matched control, Broadsign’s Netherlands Decathlon case names non-exposed controls, and the opened Vistar and GroundTruth releases give no randomization procedure. These are bounded documentation findings, not proof of each product’s default. The IAB UK Hivestack PDF was inaccessible on repeated opens, so the six-case tally and its UK control method were not independently verified. Preserve the verifier’s geography qualification and mark that component unresolved; do not treat a campaign-specific comparator as a product default. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F24; data/claims-overlay.json |
| R2-56 | Hierarchical geographic cells are not part of AdCP at the cutoff versions: issue #5969 (geo_cells / geo_cells_exclude with an h3, s2, geohash or custom cell-system enum), opened by the paper's author on 21 July 2026, is OPEN with milestone 3.4.0 and labels rfc and needs-wg-review, its two companions (#6013 conversion, #6014 cell-grain delivery reporting) are also OPEN on milestone 3.4.0, no geo_cells pull request was found, and neither v3.2.6 nor v3.3.0-beta.0 contains a geo_cells field; under the brief's rules the proposal earns no protocol state. | Verified through the GitHub connector: issue 5969 and companions 6013/6014 are open on milestone 3.4.0; creation dates and labels match. Neither the v3.2.6 nor v3.3.0-beta.0 targeting properties contain geo_cells. A repository PR search for geo_cells returned zero results. Retain the proposal/no-state distinction and bound the PR absence to that search. | Reviewer opened: https://github.com/adcontextprotocol/adcp/issues/5969 https://api.github.com/repos/adcontextprotocol/adcp/issues/5969 https://api.github.com/repos/adcontextprotocol/adcp/issues/6013 https://api.github.com/repos/adcontextprotocol/adcp/issues/6014 https://github.com/adcontextprotocol/adcp/blob/v3.2.6/static/schemas/source/core/targeting.json https://github.com/adcontextprotocol/adcp/blob/v3.3.0-beta.0/static/schemas/source/core/targeting.json https://api.github.com/search/issues?q=repo%3Aadcontextprotocol%2Fadcp%20is%3Apr%20geo_cells | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-J-F05; data/claims-overlay.json |
| R2-57 | An implemented AdCP connector maps administrative geo targeting into an ad server: the Prebid Sales Agent v2.0.0 (1 July 2026) translates geo_countries, geo_regions and geo_metros (Nielsen DMA only) and their exclusions into Google Ad Manager targetedLocations, but rejects geo_postal_areas with an explicit error ("not implemented in GAM static mapping") and contains no handling for geo_proximity, store_catchments or geo_places in the targeting manager read. | The verifier is correct: release v2.0.0 was published 1 July 2026; pyproject pins adcp Python SDK 5.7.0 and explicitly identifies spec 3.1.0-beta.3. The manager maps country, region and Nielsen-DMA targets and exclusions, raises on postal targeting/exclusion, and contains none of geo_proximity, store_catchments or geo_places. This establishes a versioned implementation, not a successful live buy or an observed failed postal-targeted transaction. | Reviewer opened: https://github.com/prebid/salesagent/blob/v2.0.0/src/adapters/gam/managers/targeting.py https://github.com/prebid/salesagent/blob/v2.0.0/pyproject.toml https://api.github.com/repos/prebid/salesagent/releases/tags/v2.0.0 | supported with qualification | The verifier is correct: release v2.0.0 was published 1 July 2026; pyproject pins adcp Python SDK 5.7.0 and explicitly identifies spec 3.1.0-beta.3. The manager maps country, region and Nielsen-DMA targets and exclusions, raises on postal targeting/exclusion, and contains none of geo_proximity, store_catchments or geo_places. This establishes a versioned implementation, not a successful live buy or an observed failed postal-targeted transaction. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-J-F06; data/claims-overlay.json |
| R2-58 | In the public AAO agent registry on 24 October 2026, none of the 34 registered agents shows a passing geo or DOOH conformance test: geo_region_targeting, geo_region_all_values_support, geo_region_exclusion_support and geo_place_targeting appear only on the two agents graded on AdCP 3.2.6 (goTom Sandbox and Publink) and are untested (0/0 steps) on both, no agent declares the sales-dooh specialism, and the 11 agents holding a verified badge hold it only in spec mode, none in the live-traffic mode. | The API at https://agenticadvertising.org/api/registry/agents?limit=500 was inaccessible through the web tool, and direct HTTP returned 403. The registry UI opened only an empty shell. Therefore none of the 34/11/0 counts, per-agent storyboards, three-versus-two spec-version correction or badge-mode tally was independently verified. Retain as unresolved until the dated raw list and per-agent compliance responses are supplied; neither the original nor the verifier’s correction earns support from this review. | Reviewer opened: https://agenticadvertising.org/registry/?tab=agents | supported with qualification | The API at https://agenticadvertising.org/api/registry/agents?limit=500 was inaccessible through the web tool, and direct HTTP returned 403. The registry UI opened only an empty shell. Therefore none of the 34/11/0 counts, per-agent storyboards, three-versus-two spec-version correction or badge-mode tally was independently verified. Retain as unresolved until the dated raw list and per-agent compliance responses are supplied; neither the original nor the verifier’s correction earns support from this review. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-J-F07; data/claims-overlay.json |
| R2-59 | The strongest public claims of location signals in agent-driven AdCP buys are vendor releases that do not record the signal or the protocol version: PubMatic and Butler/Till (15 December 2025) describe an AdCP-enabled CTV campaign for Clubtails "delivering geo-targeted, premium CTV inventory" without naming a field or geographic unit, and Broadsign (27 May 2026) describes a Netherlands OOH campaign in which agents using AdCP informed "audience and venue targeting" without naming identifiers or delivery metrics; neither reaches documented production use of a named location signal under H7's ru… | The full PubMatic/Butler-Till release and full Broadsign release were opened. They report executed/live campaigns using AdCP and geographic or venue targeting, but neither identifies a protocol version or a specific carried location field. The verifier’s distinction is correct under H7’s strict rule. Keep the positive vendor production claims visible while stating that they fall short of this paper’s field-and-version evidence requirement; do not describe them as merely future launches. | Reviewer opened: https://www.businesswire.com/news/home/20251215976339/en/PubMatic-and-ButlerTill-Launch-AdCP-Enabled-Agentic-AI-Campaign-Across-CTV https://broadsign.com/blog/broadsigns-sell-side-ai-agent-joins-with-global-netherlands-and-draft-digital-to-deliver-first-ever-end-to-end-agentic-ai-powered-ooh-campaign/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-J-F08; data/claims-overlay.json |
| R2-60 | The non-agentic baseline is further along than any agentic protocol: OpenRTB 2.6 defines a Geo object (lat, lon, type, accuracy, lastfix, ipservice, country, region, metro, city, zip, utcoffset) plus DOOH venue types and a qty impression multiplier, and Google Authorized Buyers documents production bid requests in which geo type is always set to IP (2) and accuracy reflects coarse IP-based geolocation, i.e. documented production use of coarse, inferred location rather than device-observed location. | The OpenRTB Geo/Qty/DOOH definitions and Google’s documented coarse IP-based request population support the baseline claim. Google marks the type field as optional, so “always IP” describes its value when populated, not presence in every request. Replace the universal comparison with “the opened Google baseline reaches documented use, whereas the screened agentic evidence has not satisfied H7’s field-and-version criterion.” These sources alone do not establish superiority over every agentic protocol. | Reviewer opened: https://raw.githubusercontent.com/InteractiveAdvertisingBureau/openrtb2.x/main/2.6.md https://developers.google.com/authorized-buyers/rtb/openrtb-guide | supported with qualification | The OpenRTB Geo/Qty/DOOH definitions and Google’s documented coarse IP-based request population support the baseline claim. Google marks the type field as optional, so “always IP” describes its value when populated, not presence in every request. Replace the universal comparison with “the opened Google baseline reaches documented use, whereas the screened agentic evidence has not satisfied H7’s field-and-version criterion.” These sources alone do not establish superiority over every agentic protocol. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-J-F11; data/claims-overlay.json |
| R2-61 | Inference for H7 at 24 October 2026: across AdCP 3.2.6, AAMP components, ARTF 1.0 and the location-vendor interfaces screened, the highest documented state for any location signal in agent-driven buying is an implemented connector (AdCP administrative geo to Google Ad Manager in the Prebid Sales Agent); geographic cells reach no state, geo_proximity, store catchments, geo_places, venue identifiers, DOOH metrics and foot_traffic reach schema representation only, no interoperability test result is public, and the two vendor-claimed agentic campaigns do not record a location field, so H7 is unre… | The highest state independently established here is the Prebid administrative-geo connector against AdCP 3.1.0-beta.3; v3.2.6 contains additional geographic and DOOH schema fields. The verifier’s version qualification is sound. The campaign releases do not meet H7’s version-and-signal test. However, the registry API could not be checked, vendor-interface and all-protocol absence was not exhaustively re-established, and source-code postal refusal creates a rule-interpretation issue. Report H7 unresolved on the verified positive record, with these explicit limits; bound all no-interoperability/no-production statements to the searched evidence. | Reviewer opened: https://github.com/adcontextprotocol/adcp/blob/v3.2.6/static/schemas/source/core/targeting.json https://github.com/adcontextprotocol/adcp/blob/v3.2.6/static/schemas/source/core/delivery-metrics.json https://github.com/adcontextprotocol/adcp/blob/v3.2.6/static/schemas/source/core/placement.json https://github.com/prebid/salesagent/blob/v2.0.0/src/adapters/gam/managers/targeting.py https://github.com/prebid/salesagent/blob/v2.0.0/pyproject.toml https://github.com/IABTechLab/OpenProposal/blob/main/spec/openproposal-3.0.md https://github.com/IABTechLab/agentic-real-time-framework/blob/main/proto/agenticrtbframework.proto https://www.businesswire.com/news/home/20251215976339/en/PubMatic-and-ButlerTill-Launch-AdCP-Enabled-Agentic-AI-Campaign-Across-CTV https://broadsign.com/blog/broadsigns-sell-side-ai-agent-joins-with-global-netherlands-and-draft-digital-to-deliver-first-eve… | supported with qualification | The highest state independently established here is the Prebid administrative-geo connector against AdCP 3.1.0-beta.3; v3.2.6 contains additional geographic and DOOH schema fields. The verifier’s version qualification is sound. The campaign releases do not meet H7’s version-and-signal test. However, the registry API could not be checked, vendor-interface and all-protocol absence was not exhaustively re-established, and source-code postal refusal creates a rule-interpretation issue. Report H7 unresolved on the verified positive record, with these explicit limits; bound all no-interoperability/no-production statements to the searched evidence. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-J-F13; data/claims-overlay.json |
| R2-62 | Mobvista (HKEX: 1860) presents itself at the cutoff as a mobile app-growth company whose product lines are Mintegral (user acquisition and monetisation), GameAnalytics, Nativex, Mindworks, Playturbo and XMP; its homepage names no location product, so it is excluded from the location cohorts. | The opened homepage’s listed businesses match the claim. No separately named location-advertising product was identified in that page’s product menu. Exclusion on this bounded product-admission basis is supported; it is not evidence that no product exists anywhere. | Reviewer opened: https://www.mobvista.com/en/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F01; data/claims-overlay.json |
| R2-63 | Infillion announced on 18 September 2026 an agreement to acquire Foursquare (terms undisclosed); the release refers to leadership joining 'after the transaction closes', so on the settling party's own record the event is an acquisition announced and pending, even though Foursquare's homepage banner reads 'Foursquare has joined Infillion' at 24 October 2026. | The September 18 announcement establishes an announced acquisition and describes closing as future then. The currently opened Foursquare page says it has joined Infillion. Those sources do not establish a closing date or prove that the transaction remained pending on October 24. Record announced September 18; current company website says joined; closing date/status not independently settled. The verifier’s definite pending classification at the later date is too strong. | Source check G3 (24 October 2026, primary sources opened): LOC-U-F02: reclassified from 'announced, pending' to 'announced; target says joined; closing date unconfirmed'. No closing release found 24 Oct; secondary 'acquired' claims were unopenable. https://infillion.com/blog/infillion-to-acquire-foursquare-closing-the-loop-between-advertising-and-real-world-results/ https://foursquare.com/ https://finance.yahoo.com/media-advertising/articles/exclusive-infillion-acquires-location-data-192246542.html | supported with qualification | Infillion announced on 18 September 2026 that it is acquiring Foursquare (no terms disclosed), and the release treats closing as a future event. By 24 October 2026 Foursquare's homepage says it 'has joined Infillion', but no closing announcement or date was found, so completion is unconfirmed rather than pending. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F02; data/claims-overlay.json |
| R2-64 | T-Mobile US owns both a DOOH SSP and a location-led DSP from the seed list: it completed the acquisition of Vistar Media on 3 February 2025 for US$621 million in cash (agreement 20 December 2024) and of Blis Holdco Limited on 3 March 2025 for US$180 million (agreement 18 February 2025), and Blis's homepage title reads 'Blis, part of T-Ads \| Omnichannel DSP'. | Closing dates and gross cash amounts are supported. Label $621m for Vistar and $180m for Blis as cash transferred, not acquisition fair values. The 10-K excludes settlement of preexisting relationships: acquisition consideration is $617m for Vistar and $174m for Blis. The verifier identified only the Blis distinction. Both full SEC filings were downloaded and read after the web fetcher’s size limit. | Source check G3 (24 October 2026, primary sources opened): LOC-U-F05: labelled 621/180 'cash transferred' and added 617/174 consideration for both deals (verifier had only Blis). https://www.sec.gov/Archives/edgar/data/1283699/000128369926000010/tmus-20251231.htm https://www.sec.gov/Archives/edgar/data/1283699/000128369926000010/tmus-20251231.htm evidence/verify/U.json | supported with qualification | T-Mobile US completed its acquisitions of Vistar Media on 3 February 2025 (agreement 20 December 2024) and of Blis Holdco Limited on 3 March 2025 (agreement 18 February 2025), transferring cash of US$621 million and US$180 million respectively. Excluding payments that settled preexisting relationships, the 10-K puts the fair value of consideration at US$617 million (Vistar) and US$174 million (Blis), both provisional. Blis's homepage is titled 'Blis, part of T-Ads \| Omnichannel DSP'. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F05; data/claims-overlay.json |
| R2-65 | Bluedot, the seed developer-geofencing vendor, is no longer independent: Rezolve AI plc agreed on 4 February 2025 to buy the share capital of Bluedot Industries, Inc. (Delaware) and Bluedot Industries Pty. Ltd. and closed on 20 February 2025 for 819,737 ordinary shares, a related-party deal (the seller DBLP is wholly owned by a Rezolve director) accounted for as a common-control transfer; bluedot.io now redirects to rezolve.com. | The opened full 20-F supports the named entities, February 4 agreement, February 20 close, 819,737 shares and related-party/common-control accounting; the domain redirects to Rezolve. Preserve the entity scope: Rezolve separately acquired Bluedot Innovation in March 2025, so this transaction is not necessarily the full consideration for every business operating under the Bluedot brand. | Reviewer opened: https://www.sec.gov/Archives/edgar/data/1920294/000119312526132456/rzlv-20251231.htm https://bluedot.io/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F07; data/claims-overlay.json |
| R2-66 | Outlogic (formerly X-Mode, under an FTC order banning sale of sensitive location data) still licenses an 'Observation Panel', attribution data and audiences at the cutoff through Snowflake, daily and hourly AWS S3 exports, a REST API, Eyeota and Adsquare, and its homepage notice under Texas law says it 'may receive and sell the precise latitude and longitude coordinates of your mobile device'. | The opened homepage confirms the Observation Panel and delivery routes, the no-bidstream assertion, and the Texas notice about possible precise-coordinate sales. Attribute these to Outlogic. Availability plus that notice supports continued marketed licensing; it does not establish any prohibited sensitive-location sale. | Reviewer opened: https://outlogic.io/ | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F10; data/claims-overlay.json |
| R2-67 | Product names in the seed register have drifted by the cutoff in ways that change rows: Reveal sells the 'Mira Console' OOH measurement platform (no 'VISIT Local'); Azira sells Azira One, Azira Insights, Azira Audience, Azira Campaigns and Azira Measurement; Foursquare adds Proximity and the Spatial suite; Cuebiq sells Footfall Measurement and Cuebiq Audiences; Advan sells FIT, REI, REVEAL, SPENDVIEW and Patterns+; Kochava sells store visitation through Kochava Foundry; Bluedot is owned by Rezolve; and the name 'Reveal' now denotes three different products (Reveal Mobile, Advan REVEAL, Radar … | The current names are largely confirmed, with two qualifications beyond the seed: Mira Console is a self-serve interface to Reveal’s measurement service, not proof that the whole product was renamed Mira; Kochava describes store visitation as an outcome available through partnerships/Foundry, not a separately named visitation SKU. Current menus establish present names, not a dated one-to-one rename history. Keep vendor-qualified names for the three different Reveal brands. | Source check G3 (24 October 2026, primary sources opened): LOC-U-F26: Mira Console = self-serve UI, not a product rename; Kochava visitation = outcome, not a SKU; Advan has six products (REP added). https://revealmobile.com/ https://advanresearch.com/ https://www.kochava.com/ https://www.sec.gov/Archives/edgar/data/1920294/000119312526132456/rzlv-20251231.htm | supported with qualification | By the cutoff, vendors' current pages use these names: Reveal Mobile offers OOH measurement with reporting through its self-serve 'Mira Console' (no 'VISIT Local'); Azira lists Azira One, Insights, Audience, Campaigns and Measurement; Foursquare adds Proximity and the Spatial suite; Cuebiq lists Footfall Measurement and Cuebiq Audiences; Advan lists FIT, REI, REVEAL, SPENDVIEW, Patterns+ and REP. Kochava lists store/footfall visitation as a measurable outcome via data partnerships and Kochava Foundry reporting rather than a named product, Bluedot is owned by Rezolve (acquisitions closed February and March 2025), and 'Reveal' denotes three vendors' products (Reveal Mobile, Advan REVEAL, Radar Reveal). These are current names, not a dated rename history. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F26; data/claims-overlay.json |
| R2-68 | IAB's location-data guidance is a 2016 publisher guide and a 2014 buyer's guide: the 'IAB Mobile Location Data Guide for Publishers' (February 2016, revised; IAB Mobile Location Working Group, with contributors including Factual, Foursquare, NinthDecimal, Thinknear and xAd) defines accuracy and precision and gives the two-decimal (about 1 km) and four-decimal (about 10 m) equivalences, while the 'Mobile Location Data Buyer's Guide' (1 July 2014) is gated; this review found no later IAB or IAB Tech Lab location-data buyer or seller guide in the sources screened by the cutoff. | The internal verifier is right: IAB published the buyer-facing Location-based Marketing Playbook for Retailers in April 2017. It asks about panel versus ad-exchange collection and control selection, and cites February 2015 Marketing ROI and Location Data plus 12 Questions Every Buyer Should Ask. The original bounded search missed a later guide; do not present 2016 as the newest located IAB guidance. | Source check G1 (24 October 2026, primary sources opened): R2-68: accepted. The April 2017 IAB retail playbook supersedes the '2016 is newest' framing. Disclose Verve as a contributor. https://www.iab.com/wp-content/uploads/2017/04/IAB_Location-Playbook-for-Retail-Marketers_FINAL.pdf https://www.iab.com/wp-content/uploads/2017/04/IAB_Location-Playbook-for-Retail-Marketers_FINAL.pdf https://www.iab.com/wp-content/uploads/2017/04/IAB_Location-Playbook-for-Retail-Marketers_FINAL.pdf | contradicted | IAB's location-data guidance found by the cutoff runs from the gated 'Mobile Location Data Buyer's Guide' (July 2014) and the February 2016 'Mobile Location Data Guide for Publishers' to the April 2017 'Location-based Marketing Playbook for Retailers'. The playbook is buyer-facing and asks whether device data come 'through panels or pass through ad exchanges' and how store-visit-lift control groups are selected. It cites 'IAB Marketing ROI and Location Data' (February 2015) and '12 Questions Every Buyer Should Ask about Location Data'. No IAB or IAB Tech Lab location guide later than April 2017 was found in the sources screened. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-A-F22; data/claims-overlay.json |
| R2-69 | None of the opened documentation for the fourteen cohort A products publishes a comparison of its visit numbers with an independent reference (door counts, transactions or official visitor statistics) with error statistics; the validation claims found are vendor assertions without a document: InMarket's 'Comscore-Verified Visit Methodology' and GroundTruth's Blueprints 'verified using a selection of trusted location sources'. | The verifier is right: Azira reports MAPE 10% and correlation 0.85 against sales/counters. The page supplies no reproducible sample, period, places or held-out design. Report this as a vendor assertion, not independent validation. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology | contradicted | The verifier is right: Azira reports MAPE 10% and correlation 0.85 against sales/counters. The page supplies no reproducible sample, period, places or held-out design. Report this as a vendor assertion, not independent validation. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-E-F31; data/claims-overlay.json |
| R2-70 | Azira Audience (live, 2026) builds segments 'based on real-world visitation, purchase behavior, and consumer intent' from '13T Global Mobility Signals', '35.6 Billions consumer transactions' and '43M Points of Interest', activates through Azira's own Advance DSP (claimed 80-90% addressability) across desktop, mobile, CTV, audio, DOOH and social, supports custom polygons and visit duration, and uses device graphs that 'analyze IP addresses and home or daytime locations'; no collection route for the mobility signals is stated. | The verifier correctly rejects “no collection route”: Azira discloses a blend without proportions. Keep the 80–90% addressability number attributed to its 5 January 2025 marketing blog, without treating it as measured accuracy. The current data page shows 45m+ global POIs while the digest cites 43m from a different page; retain URL/access-date-specific values. The proposed “within 3m precision” addition was not found on the opened current Audience page and should remain unresolved without its exact captured source. | Reviewer opened: https://www.azira.com/platform/data https://www.azira.com/solutions/azira-audience https://www.azira.com/platform/privacy https://www.azira.com/blogs/how-to-use-a-next-generation-demand-side-platform-powered-by-single-data-set-to-boost-return-on-ad-spend | contradicted | The verifier correctly rejects “no collection route”: Azira discloses a blend without proportions. Keep the 80–90% addressability number attributed to its 5 January 2025 marketing blog, without treating it as measured accuracy. The current data page shows 45m+ global POIs while the digest cites 43m from a different page; retain URL/access-date-specific values. The proposed “within 3m precision” addition was not found on the opened current Audience page and should remain unresolved without its exact captured source. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F07; data/claims-overlay.json |
| R2-71 | GroundTruth Ads Manager documents the widest channel set of any cohort B product opened (mobile, CTV and OTT, audio, DOOH with '1M+ digital out-of-home screens across the US', and direct mail) but names no third-party DSP activation, only reporting integrations with TapClicks, NinjaCat and Agency Analytics; Foursquare, Adsquare, Precisely, Reveal and Mobilewalla name external DSPs or marketplaces, while Azira, Vistar and Simpli.fi activate through their own platforms. | The verifier is right to add GroundTruth’s desktop/tablet channel and reject unique “widest” status. GroundTruth documents six categories including direct mail; Azira documents desktop/mobile, audio, video, CTV, DOOH and social. Under the brief’s common channel taxonomy both have mobile, desktop/web, CTV, audio and DOOH, plus one additional category. State the channel sets; remove the superlative unless all cohort rows are recoded consistently. GroundTruth’s named reporting integrations do not by themselves establish audience activation into third-party DSPs. | Reviewer opened: https://www.groundtruth.com/solutions/ads-manager/ https://www.azira.com/solutions/azira-campaigns https://www.azira.com/blogs/how-to-use-a-next-generation-demand-side-platform-powered-by-single-data-set-to-boost-return-on-ad-spend | contradicted | The verifier is right to add GroundTruth’s desktop/tablet channel and reject unique “widest” status. GroundTruth documents six categories including direct mail; Azira documents desktop/mobile, audio, video, CTV, DOOH and social. Under the brief’s common channel taxonomy both have mobile, desktop/web, CTV, audio and DOOH, plus one additional category. State the channel sets; remove the superlative unless all cohort rows are recoded consistently. GroundTruth’s named reporting integrations do not by themselves establish audience activation into third-party DSPs. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-F-F24; data/claims-overlay.json |
| R2-72 | Connecticut's Public Act 25-113 (SB 1295) amends the Connecticut Data Privacy Act effective 1 July 2026 so a controller shall 'not sell the sensitive data of a consumer without the consumer's consent' (sensitive data includes precise geolocation data, defined by a 1,750-foot radius), a consent rule rather than a sale ban, and restricts collection of minors' precise geolocation to what is strictly necessary with a persistent signal. | The original consent-only description is outdated at this review date. Connecticut’s Attorney General states that the October 1, 2026 changes under PA 26-64 include a precise-geolocation sale ban, confirming the verifier’s core correction. The official act PDF repeatedly returned 502/503, so I have not independently checked every subsection, exception or approval-date detail in its proposed replacement; cite the opened AG notice for the confirmed change. | Source check G4 (24 October 2026, primary sources opened): R2-72 accepted: PA 26-64 Secs. 14-15 sale ban (eff. 1 Oct 2026, approved 27 May 2026) read at cga.ct.gov (opened fine, no 502) and matches the AG notice of 16 Sep 2026; consent-only reading is outdated. https://portal.ct.gov/ag/press-releases/2026-press-releases/rights-and-requirements-related-to-new-and-updated-privacy-laws https://www.cga.ct.gov/2026/ACT/PA/PDF/2026PA-00064-R00SB-00004-PA.PDF https://www.cga.ct.gov/2026/ACT/PA/PDF/2026PA-00064-R00SB-00004-PA.PDF | contradicted | Connecticut bans the sale of precise geolocation data: Public Act 26-64 (SB 4, approved 27 May 2026, effective 1 October 2026) provides that no controller and no third party shall sell any consumer's precise geolocation data (1,750-foot radius; content of communications and utility metering data excepted), replacing the consent-to-sell rule PA 25-113 had put in place from 1 July 2026. Data brokers selling or licensing brokered personal data in Connecticut must be registered from 1 January 2027. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F13; data/claims-overlay.json |
| R2-73 | Virginia's Consumer Data Protection Act (Code 59.1-575) defines precise geolocation data by a 1,750-foot radius and counts it as sensitive data, the category to which the act's consent requirement attaches; no Virginia sale ban or geofencing ban on precise geolocation was found in the section opened. | The verifier is right on the operative rule. Section 59.1-578(A)(6) expressly bars sale or offers for sale of precise geolocation; the adjacent sensitive-data processing consent rule does not supply a consent exception to that sale ban. A definitions-only lookup cannot support the original current-law reading. | Source check G4 (24 October 2026, primary sources opened): R2-73 accepted: VA 59.1-578(A)(6) sale ban verified at law.lis.virginia.gov; ch. 820 (SB 338) approved 13 Apr 2026, effective 1 Jul 2026 per LIS bill history (replaces the Willkie-sourced date). https://law.lis.virginia.gov/vacode/title59.1/chapter53/section59.1-578/ https://lis.virginia.gov/bill-details/20261/SB338/text/CHAP0820 https://lis.virginia.gov/bill-details/20261/SB338 | contradicted | Virginia bans the sale of precise geolocation data: Code of Virginia 59.1-578(A)(6), added by 2026 Acts ch. 820 (SB 338, approved 13 April 2026, effective 1 July 2026), requires a controller to "not sell or offer for sale precise geolocation data concerning a consumer", with no consent exception; other processing of sensitive data, including precise geolocation, still requires consent under 59.1-578(A)(5). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F16; data/claims-overlay.json |
| R2-74 | GroundTruth's DOOH attribution, announced on 9 October 2025 with Place Exchange, 'captures mobile device IDs within viewing range of a screen when an ad plays' and links them to later store visits using GroundTruth's own 'location and purchase data' based on 'observed real-world consumer behavior'; the release cites a belVita campaign with Target reported at 221,000 in-store visits in one month and a 3.44% visitation rate, with 'more than half of those visits being incremental', and describes no control design beyond offering 'third-party lift studies'. | The verifier is correct. The 221,000 Target visits and 3.44% rate belong to the belVita omnichannel DOOH-plus-mobile case. The claim that over half the visits were incremental belongs to a separate unnamed national retailer’s 42,000-plus visits in New York, Chicago and Los Angeles. Keep the two cases separate and vendor-attributed; the release gives no comparator design for the incremental assertion. | Reviewer opened: https://www.groundtruth.com/article/groundtruth-expands-into-digital-out-of-home-with-measurable-foot-traffic-attribution/ https://billboardinsider.com/ground-truth-partners-with-place-exchange-for-foot-traffic-attribution/ | contradicted | The verifier is correct. The 221,000 Target visits and 3.44% rate belong to the belVita omnichannel DOOH-plus-mobile case. The claim that over half the visits were incremental belongs to a separate unnamed national retailer’s 42,000-plus visits in New York, Chicago and Los Angeles. Keep the two cases separate and vendor-attributed; the release gives no comparator design for the incremental assertion. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-I-F17; data/claims-overlay.json |
| R2-75 | Three seed data suppliers describe their supply as SDK or server-to-server first-party app data with explicit counts at the cutoff: Quadrant ('an Appen company', '1+ billion opted in devices', '50+ billion daily location events', compiled 'through a server to server (S2S) integration'), Complementics ('over 1 billion global devices', location events 'from multiple supply sources') and Outlogic ('No Bidstream Data'); none gives a US-only count or a period. | The verifier’s correction is supported. Quadrant supplies the S2S route and global counts; Complementics supplies a global count and multiple supply sources without establishing that route; Outlogic supplies first-party/no-bidstream claims without a count. Complementics is absent from the seed register. Do not turn those three different disclosures into three documented SDK/S2S sources with counts. | Reviewer opened: https://www.quadrant.io/ https://complementics.com/ https://outlogic.io/ | contradicted | The verifier’s correction is supported. Quadrant supplies the S2S route and global counts; Complementics supplies a global count and multiple supply sources without establishing that route; Outlogic supplies first-party/no-bidstream claims without a count. Complementics is absent from the seed register. Do not turn those three different disclosures into three documented SDK/S2S sources with counts. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-U-F15; data/claims-overlay.json |
| R2-76 | In the California Data Broker Registry download of 24 October 2026 (603 rows, covering brokers that operated in 2025), Collective Data Solutions, LLC (Sandpoint, Idaho), Outlogic, LLC, InMarket Media, LLC, Mobilewalla, Inc., Unacast, Inc. and Venntel, Inc. are all registered and each reports collecting consumers' precise geolocation; Unacast and Venntel also report having shared or sold consumers' data to the federal government and to other state governments in the past year; Kochava Inc. itself is not listed. | The downloaded CSV contains 603 rows. CDS, Outlogic, InMarket, Mobilewalla, Unacast and Venntel each self-report collecting precise geolocation; Unacast and Venntel report past-year federal and state-government sale/sharing. Kochava Inc. is not a row. These are submitted disclosures covering prior activity, not regulatory verification or proof of precise-data sales at the October 2026 cutoff. In particular, a collection checkbox cannot resolve Mobilewalla's sold-product precision. | Source check G5 (24 October 2026, primary sources opened): R2-76: every number and flag reproduced from registry.csv; qualification (self-reported, collection not sale, 2025 activity) added; 115/603 denominator added. https://cppa.ca.gov/data_broker_registry/registry.csv https://cppa.ca.gov/data_broker_registry/registry.csv https://cppa.ca.gov/data_broker_registry/registry.csv https://cppa.ca.gov/data_broker_registry/registry.csv | supported with qualification | In the California Data Broker Registry file (603 registrants that operated as data brokers in 2025; file dated 29 July 2026, downloaded 24 October 2026), Collective Data Solutions, Outlogic, InMarket, Mobilewalla, Unacast, Inc. and Venntel each self-report collecting precise geolocation (115 of the 603 registrants do), and Unacast and Venntel report past-year sale or sharing to the federal government and other state governments. Kochava Inc. is not listed. These are the brokers' own declarations about collection in 2025, not evidence of what they sold at the cutoff. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | evidence/research LOC-H-F15; data/claims-overlay.json |
| R2-77 | H1 | The unresolved reading follows the registered rule on the opened evidence: no comparable falling US panel series plus a documented primary-source shift was established. Permission announcements cannot supply the missing outcome. A constant-definition pre/post series for at least two vendors, or a qualifying official attribution combined with a documented shift away from SDK primary supply, would change the reading. Do not infer contraction from corporate exits or unlike reach units. | Reviewer opened: https://www.ftc.gov/system/files/ftc_gov/pdf/Complaint-InMarketMediaLLC.pdf https://docs.safegraph.com/docs/monthly-patterns https://www.utah.gov/pmn/files/1194509.pdf | supported as stated | None: H1 stays unresolved under its registered rule. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H1 |
| R2-78 | H2 | The registered rule produces a contradicted reading for the historical SafeGraph polling-place use tested, because published demographic bias is documented without a stated error. It does not produce one verdict for every vendor, product and vintage. Current retail projections remain unresolved where comparable validation is absent. | Reviewer opened: https://arxiv.org/pdf/2011.07194 | supported with qualification | Superseded by the reviewer's own revision (R2B-35): the 2018 SafeGraph polling-place coverage bias is reported as adverse historical evidence; its formal contradicted reading for that panel is conditional on the error claim and on no equal-or-better rebuttal; H2 is unresolved overall, with no single verdict across products and vintages. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H2; LOC-B-F31 |
| R2-79 | H3 | The exact documented-default count is not securely established by this review. Five opened product descriptions describe modeled/matched methods and Azira explicitly calls its PSA design the default, but the former need product-level evidence that the description governs default reporting. H3 would be contradicted if those five qualify; report that conditional result separately from an adjudicated default count. Blis documents a geographic holdout without proof of random assignment or default use, and that holdout can qualify under the broad registered category. Rebuild the product/metric default table and report disclosure coverage before finalizing the reading. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology https://blis.com/omnichannel-measurement/ https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://revealmobile.com/solutions/measurement https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://cuebiq.com/blog/ctv-ltv-attribution/ https://www.arrivalist.com/media-insights | supported with qualification | Superseded by the 14-product recount (R2B-09 to R2B-23, checked in G7): 6 matched or synthetic, 2 randomised or holdout (Azira randomised PSA; Blis geographic holdout, randomisation not documented), 1 exposed-only (GroundTruth), 5 not documented. 7 of 9 documented defaults are non-holdout, so H3 reads contradicted; the strict lift-claim count (GroundTruth excluded) is 6 of 8. Every sensitivity variant (without Adsquare; without the InMarket mapping) stays contradicted. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H3; LOC-E-F28; LOC-C-F23 |
| R2-80 | H4 | The unresolved reading is defensible, but the original and verifier denominators are not ready to publish. Source presence is not documented dominance. Treat Foursquare, InMarket and Azira as mixed/disclosed-but-primary-unknown on the opened records. Rebuild product-specific coding after U changes and count Broadsign only once US eligibility is established. H4 needs at least four products with an established primary source; a supplier list or opt-in claim cannot meet that threshold. | Reviewer opened: https://foursquare.com/legal/privacy-center/foursquare-privacy-policy-consumer-services/ https://inmarket.com/data/ https://www.azira.com/platform/data https://broadsign.com/audiences/ | supported as stated | None: H4 stays unresolved; Foursquare, InMarket and Azira recoded 'multiple sources disclosed, primary not established' (R2-37). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H4; LOC-F-F20 |
| R2-81 | H5 | The contradicted reading follows the existential rule from Outlogic’s and CDS’s current explicit precise-data marketing. Unacast adds a qualified preceding-year precise-sale disclosure alongside current location products; Mobilewalla’s sold-product precision remains unknown. The exact four-company current-seller count does not follow. Clarify precedence: a verified counterexample refutes universal withdrawal even if other product sets remain unreadable or precision-unclear; keep those company-specific unknowns unresolved. Record this as a rule clarification, not a compliance judgment. | Source check G5 (24 October 2026, primary sources opened): R2-81: precedence note added (counterexample beats partial unreadability); count is 3 documented + Mobilewalla unclear, not 4 and not 2. research/location-2026/data/hypotheses.json (H5) https://www.unacast.com/privacy-policy https://www.mobilewalla.com/california-privacy-notice-2023 https://www.mobilewalla.com/privacy-center | supported with qualification | H5 is contradicted: Outlogic and Collective Data Solutions (for Kochava) market precise-location offers at the cutoff, and Gravy Analytics dba Unacast documents a current device-level data offer plus a preceding-year precise-sale notice; Mobilewalla is unclear on precision; InMarket documents none (check G5). Precedence clarified in hypotheses.json (dated). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H5; LOC-H-F23 |
| R2-82 | Gravy Analytics, Inc. dba ‘Unacast | The current Unacast menu markets location-data products, and Gravy dba Unacast’s current policy lists precise-geolocation sales to third parties over the preceding twelve months. This supports a continuing location business plus a time-window precise-sale disclosure; it does not by itself settle the precision of a feed still sold on 24 October. Record a qualified third disclosure, counting the group once. Current precise-data products under Venntel and their fit to the order’s government-use exclusions were not separately established. | Source check G5 (24 October 2026, primary sources opened): R2-82: Unacast kept as a documented seller (Gravy dba Unacast, policy Sep 2026 + current device-level product); reviewer's 'qualified' downgrade only partly accepted; Venntel left open. https://www.unacast.com/privacy-policy https://www.unacast.com/privacy-policy https://www.unacast.com/products/location-data https://www.unacast.com/products/location-data | supported with qualification | Gravy Analytics, Inc., trading as Unacast, documents at the cutoff the sale of precise geolocation: its policy dated September 2026 lists 'Precise Geolocation Data' among categories sold in the preceding twelve months, and its current product page offers 'device-level location data'. The group counts once. Venntel's current precise-data products and their fit to the order's federal national-security exclusion were not established. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | H5: Gravy Analytics / Unacast / Venntel |
| R2-83 | we do not sell (in the plain language sense of that word) or license to customers | The policy also lists precise geolocation among data sold or shared for targeted advertising in the preceding year. It expressly denies ordinary-language precise-data sale or customer licensing. The broad statutory sale/share notice therefore does not establish a licensed precise-data product or an order breach. Record product withdrawal as not contradicted by this notice, rather than claiming public proof that no prohibited disclosure occurs. | Source check G5 (24 October 2026, primary sources opened): R2-83: both InMarket passages and order Provision II confirmed verbatim; coded 'not contradicted by the notice', not 'proven withdrawn'. https://inmarket.com/privacy/ https://inmarket.com/privacy/ https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf | supported with qualification | InMarket documents no product that sells or licenses precise location data at the cutoff: its policy (updated 30 June 2026) says it does 'not sell (in the plain language sense of that word) or license to customers' precise geolocation. The same policy discloses state-law 'sale (or sharing for interest-based advertising)' of precise geolocation to third-party partners. That notice neither establishes a licensed product nor shows that no prohibited disclosure occurs. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | H5: InMarket |
| R2-84 | H5: Kochava / Collective Data Solutions | Counts through Collective Data Solutions’ explicit marketed precision-location data feeds, which were read via direct HTML download after the web fetch failed. Use that legal entity and feed disclosure, not Kochava Foundry’s outcome-reporting page. The sensitive-location prohibition and direct-service consent exception do not ban all non-sensitive precise-location licensing. Compliance is not assessed. | Source check G5 (24 October 2026, primary sources opened): R2-84: CDS-Kochava link sourced to Dkt. 101 and Dkt. 138; CDS evidence strengthened with its own 'resold ... precise geolocation' disclosure; Foundry not relied on. https://collectivedata.io/ https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-3.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf https://collectivedata.io/ | supported with qualification | Kochava counts through its subsidiary Collective Data Solutions, LLC, to which it transferred its data-broker business (D. Idaho Dkt. 101) and which is a co-defendant bound by the June 2026 stipulated order. CDS markets 'precision location' data feeds and states in its 1 October 2026 policy that it has resold precise geolocation, including lat-long coordinates, in the preceding twelve months. The order bans only sensitive-location data, with a direct-relationship consent exception, so this is not a compliance finding. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | H5: Kochava / Collective Data Solutions |
| R2-85 | Mobilewalla does process “movement data” that includes mobile advertising ID (MAID), latitude and longitude | Commercial licensing and movement-data processing are documented, but coordinates/device IDs alone do not disclose spatial precision or prove which fields are licensed. The order excludes coarse location. Do not count this as a fourth confirmed precise-data seller without a current feed schema, precision threshold or product statement tying that data to licensing. Do not infer withdrawal either. | Source check G5 (24 October 2026, primary sources opened): R2-85: unclear kept; California notice (30 Jun 2026) added, which states the held data's precision but not the licensed fields. https://www.mobilewalla.com/privacy-center https://www.mobilewalla.com/privacy-center https://www.mobilewalla.com/california-privacy-notice-2023 https://www.ftc.gov/system/files/ftc_gov/pdf/Mobilewalla-D&O.pdf | supported with qualification | Mobilewalla is coded unclear. Its current notices say it licenses personal information to clients, processes MAID-linked lat/long 'movement data', holds CCPA-precise geolocation (1,850 ft) as its only sensitive category, and 'may use Sensitive Personal Information as a part of its marketing and advertising products'. None says that the licensed products deliver precise coordinates, so it is neither a confirmed precise-data seller nor shown to have withdrawn. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | H5: Mobilewalla |
| R2-86 | we may receive and sell the precise latitude and longitude coordinates of your mobile device. | Counts as a documented marketed precise-data seller: the product is licensed and the company explicitly identifies precise-coordinate sales. The order restricts sensitive-location data and imposes other controls; it is not a blanket prohibition on every precise-data product. This is an existential counterexample to universal withdrawal, not a compliance judgment. | Source check G5 (24 October 2026, primary sources opened): R2-86: confirmed verbatim on outlogic.io and its privacy notice (past-12-months sold = Yes for lat/long); no change beyond adding the privacy-notice table. https://outlogic.io/ https://outlogic.io/ https://www.ftc.gov/system/files/ftc_gov/pdf/X-ModeSocialDecisionandOrder.pdf | supported with qualification | Outlogic (formerly X-Mode) documents at the cutoff the licensing of location data through Snowflake, S3 exports and an API, and states that it 'may receive and sell the precise latitude and longitude coordinates of your mobile device'; its privacy notice reports selling lat/long geolocation in the past 12 months. Its order bans sensitive-location data, not all precise data, so this is an existential counterexample to H5, not a compliance finding. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | H5: Outlogic |
| R2-87 | H6 | The registered rule yields contradicted as of 24 October 2026: the opened MRC lists contain no accredited visit or visit-lift metric. Neither the former Foursquare nor GroundTruth status changes that. A final cutoff verdict requires rereading the lists on or after the planned 30 October freeze. | Reviewer opened: https://mediaratingcouncil.org/accreditation/digital https://mediaratingcouncil.org/accreditation/multimedia https://mediaratingcouncil.org/accreditation/out-of-home https://mediaratingcouncil.org/accreditation/television https://mediaratingcouncil.org/accreditation/radio https://mediaratingcouncil.org/accreditation/outside-US https://mediaratingcouncil.org/accreditation/services-under-review/digital https://mediaratingcouncil.org/accreditation/services-under-review/out-of-home https://mediaratingcouncil.org/accreditation/services-under-review/television https://mediaratingcouncil.org/accreditation/services-under-review/radio https://mediaratingcouncil.org/accreditation/services-under-review/services-outside-US | supported as stated | None now: H6 contradicted as of 24 October 2026; MRC's lists are re-read at the 30 October freeze. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H6 |
| R2-88 | adcp==5.7.0 | The supported branch is not met: the opened campaign releases omit protocol version and the carried field, and source code alone is below production use. The contradicted branch cannot be assigned from source code alone if “in a workflow or interoperability test” requires an observed attempt. Report unresolved provisionally with the verified connector state and the postal-branch ambiguity. A dated runtime trace or test recording a postal-targeting request and rejection would change the reading under the present literal negative rule. | Reviewer opened: https://github.com/prebid/salesagent/blob/v2.0.0/src/adapters/gam/managers/targeting.py https://github.com/prebid/salesagent/blob/v2.0.0/pyproject.toml https://www.businesswire.com/news/home/20251215976339/en/PubMatic-and-ButlerTill-Launch-AdCP-Enabled-Agentic-AI-Campaign-Across-CTV https://broadsign.com/blog/broadsigns-sell-side-ai-agent-joins-with-global-netherlands-and-draft-digital-to-deliver-first-ever-end-to-end-agentic-ai-powered-ooh-campaign/ | supported as stated | None: H7 unresolved. Author ruling (24 October 2026) recorded as a dated clarification: a by-design refusal seen only in code is not a signal that 'could not be carried'. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H7 |
| R2-89 | n/a comes only from the cohort's applicability mask (cohorts[].applicable), never vendor by vendor. | The masks do not fit the admitted outputs. Cohort C includes place-only datasets, raw movement feeds and projected footfall dashboards, while every C row must be scored for visit definition, visit-count validation and projection. Cohort D includes Places APIs alongside visit-detection SDKs, and A includes visitation reporting alongside lift. Register output-specific subcohorts or fixed task masks before scoring; otherwise a product that does not purport to project visits is treated as undocumented on an irrelevant capability. Do not use n/e to stand for non-applicability. | Reviewer opened: | structural | Cohorts narrowed by product, not company: cohort D keeps geofencing and visit-detection SDKs and APIs; places and maps APIs and place-only datasets are listed; cohort C split into C1 (foot-traffic and visit analytics) and C2 (raw movement data) with fix-level anchors; cohort A kept whole with causal design scored as causal capability (R2B-01, R2B-37). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json cohorts; data/product-register.json |
| R2-90 | A product that reports observed counts only and documents that it does not project is scored against the 'documented as observed-only' note | scoring-rules.json/dimensions/projection refers to an observed-only note but supplies no observed-only numeric anchors: anchors 1–5 all require a projection method. Supply explicit anchors or a preregistered subcohort mask before scoring raw feeds and observed-only products. Documenting that no projection is performed cannot be mapped reproducibly to the existing 0–5 ladder. | Reviewer opened: | structural | Projection removed from the Ground Quotient for every row and kept as a separate module for products making population claims (R2B-02). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json dimensions, modules.projection, composite_index.groups |
| R2-91 | Daily figures are not converted to monthly; they are scored on the monthly thresholds divided by four, marked derived. | Remove the divide-by-four rule for daily unique devices. It assumes an undocumented daily-to-monthly turnover ratio. A panel with 12.5M identical devices every day has 12.5M monthly devices yet clears the 50M monthly tier under this rule. Additive API request counts need a separate rule: 1B requests a day over a stated 30-day month is 30B monthly requests, whereas the current divide-by-four ladder gives it tier 4 rather than tier 5 before evidence caps. Keep unique-device periods distinct; aggregate request counts only over a documented period. | Reviewer opened: | structural | Divide-by-four rule removed. Daily uniques set a labelled lower bound on monthly uniques on the same basis; daily API requests summed over a documented month; generic annualisation removed (R2B-03). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json scale (renamed documented production volume) |
| R2-92 | Bidstream 'devices seen' or 'reach' is an attribute, never a panel, and caps the cell at 2. | Keep the distinction between a panel and bidstream reach, but do not call the resulting number a pure scale score. The anchor imposes a source-quality penalty even when production volume is documented, while source quality, precision and validation already have their own dimensions. Report volume and collection route separately or state explicitly that this is qualified panel scale. The campaign-count alternative in A also measures a different unit from devices; retain the unit and do not interpret the two as equivalent measurement coverage. | Reviewer opened: | structural | Dimension renamed 'Documented production volume'; the bidstream ceiling at 2 is stated as an author policy and bidstream reach is never a panel. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json scale |
| R2-93 | 5,000,000 or more US places with polygons documented, with the place-base definition and date stated | Treat the scale and place-count thresholds as author-set breadth categories, not demonstrated thresholds of accuracy or buyer suitability. A bespoke product that correctly measures the buyer's 200 stores can score 1 for coverage, while a much broader place base can score 5 without a stated minimum observation density per place. Separate mapped POI inventory from places with usable observations and a stated minimum reportability rule. Keep US scope, but report how bands change at the threshold boundaries and avoid implying that a larger panel or POI catalogue is intrinsically better for every buying task. | Reviewer opened: | structural | Coverage thresholds described as author-set breadth categories; places with reportable observations used where disclosed, otherwise the mapped base, and the cell says which. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json coverage |
| R2-94 | Group means plus shrinkage avoid both. | The quotient still rewards omission of a documented weakness; change this claim. Worked example using the registered formula: six admitted A rows, all cells 3 except one row's privacy=0. That row has Raw=2.7, field mean F=2.95, k=12 and GQ=55.25. Replacing its privacy 0 with n/e gives Raw=3, recomputed F=3, k=11 and GQ=60; it remains band-eligible. Deletion sensitivity exposes this 4.75-point gain but does not prevent it. Describe the quotient as conditional on observed evidence and surface missingness beside it; do not claim shrinkage removes the incentive to omit poor evidence. | Reviewer opened: | structural | Reproduced exactly (GQ 55.25 with the privacy 0; 60.00 with it as n/e). The quotient text now says shrinkage limits but does not remove the reward for omission and that the quotient is conditional on the evidence observed; the formula is the shared method's and is unchanged. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json composite_index.why_not_a_sum |
| R2-95 | a regulator order, court finding or documented breach involving the product's location data | Separate the three-year adverse-record flag from current privacy controls. The current privacy anchor equates a breach, a consent order and an adverse court finding, locks the score at 0 despite independently verified remediation, and permits an abrupt change when the event leaves the window. Record event type, product scope, responsible entity, event date and remediation separately; score current controls from current evidence and retain a visible recent-event warning. If the author instead wants a non-compensable risk rule, label it as an author-defined exclusion or flag rather than a measure of current consent posture. Do not infer liability or noncompliance from a consent order alone. | Reviewer opened: | structural | Author ruling (24 October 2026), refined by R2B-05: the cell records current controls (anchors 1-5) and an event flag; the GQ uses an author-defined risk-adjusted value, min(current controls, 1), while a qualifying event is inside the three-year window, shown beside the uncapped score with its effect on GQ; 0 only for a positively documented current failure; n/e is never imputed as 1. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json privacy |
| R2-96 | a commissioned third-party certification or audit of the product's segments or data practices | The B/C/D assurance ladders are not monotone in assurance achieved: a completed commissioned audit scores 2, while an independent audit merely in progress scores 3. Payment, conflict of interest, auditor independence, audit scope, published result and currency are distinct attributes. Define them separately and require an achieved, scoped result to outrank a process with no result. Keep security-only certifications under privacy. A's top anchor should also say every visit or lift metric actually sold, rather than literally requiring both visit and lift metrics from a visitation-only product. | Reviewer opened: | structural | Assurance ladders made monotone in achieved scope (R2B-04); a pending audit is a status flag, not a score; withdrawal or lapse is a history flag, not 0. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json assurance |
| R2-97 | Independent means produced by a party with no financial interest; unknown funding is not independent. | Keep commissioned evidence labelled and distinguish it from a study with no vendor funding, but do not silently apply a different independence test across cohorts. MRC assurance is judged from an accreditor record in A, while paid certifier audits are capped at 2 in B/C/D. Specify when an independently governed external auditor can provide assurance despite being paid, and when a sponsor's influence instead makes the evidence vendor-commissioned validation. Otherwise the score ranks institutional form and funding model as well as audit quality. | Reviewer opened: | structural | One independence test in every cohort: verified scheme governance, impartiality controls, published criteria and a public scope and result; normal fees do not disqualify. cell_rules.evidence_grade amended to match. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json assurance, cell_rules |
| R2-98 | all of those documented, with a published detection error statistic | Several anchors measure the completeness of documentation rather than the level of capability: publishing any detection-error statistic advances visit-definition scoring without a required performance threshold. Ground-truth evidence also counts under ground_truth, assurance and method_disclosure, while source disclosure counts under both source_transparency and method_disclosure. With A's equal within-group averaging, direct source transparency and method disclosure alone carry 19% of Raw; the same validation can affect another 19% through ground_truth and assurance. Retain this as an evidence matrix, make repeated use of evidence explicit, and avoid presenting its quotient as accuracy or product-performance quality. | Reviewer opened: | structural | Overlap between source transparency and disclosure element 1, and between ground truth and disclosure element 5, disclosed in the quotient limits; no performance thresholds set because no evidence base exists for them. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json composite_index.limits |
| R2-99 | Among cohort A products with a documented default method, more than half | The H3 rule is a valid descriptive test of disclosed defaults, not the whole market. On the prompt's provisional counts, 7 matched/synthetic/exposed-only of 9 documented, under the prompt’s provisional coding, equals 77.8%, but 7 of the full 14 equals 50%; assigning all five unknowns to randomised removes a population majority. Publish both denominators and missing-default bounds. Add an explicit minimum of four to supported and contradicted, and an unresolved tie rule: as written, 2 of 3 can simultaneously satisfy supported and fewer-than-four unresolved, while a 2–2 tie has no stated verdict. Preserve the original rule and log any amendment after seeing evidence. Keep random assignment, geographic holdout and matched-control attributes distinct: the registered randomised-or-holdout category is broader than randomization alone. An alternative randomization-only test is a sensitivity an… | Reviewer opened: | structural | H3 decision tree recorded as a dated clarification (minimum of four in both branches; exactly half unresolved; both denominators published). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H3.clarifications |
| R2-100 | more document bidstream or IP-derived location as the primary source than document SDK or panel observation. | H4 tests a relative product count between two source families, not whether most cohort-B products or most location observations are bidstream/IP-derived. For example, among four documented products, one bidstream, zero SDK and three telco products would satisfy supported although bidstream represents only one quarter. Retitle the estimand as bidstream/IP versus SDK/panel among products documenting either, or prospectively require a majority of all classified products for the word mainly. Keep blends and other source families visible in the denominator; do not infer source volume from vendor counts. | Reviewer opened: | structural | H4 estimand recorded as a dated clarification: bidstream or IP versus SDK or panel among products documenting either; the eligibility restriction stated as substantive. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H4.clarifications |
| R2-101 | The product sets of two or more of the five cannot be read from public sources | H5 needs explicit precedence. A verified precise-location seller is an existential counterexample and satisfies contradicted even if two other vendors are unreadable, which also satisfies the present unresolved rule. State: any verified continuation within the registered precision definition => contradicted; otherwise all five verified exits/no relevant sale => supported; otherwise unresolved. Record the other companies separately. Apply each order's own definitions and do not treat an audience or visit-attribution service by itself as proof that device-level precise data is sold, nor continued lawful sale as evidence of an order breach. | Reviewer opened: | structural | H5 precedence recorded as a dated clarification: a verified current offer within the precise definition decides contradicted; the original contradicted branch is also met. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H5.clarifications |
| R2-102 | a source documents that results or data are not released to the buyer outside the vendor's own interface | The data_access 0 anchor overlaps score 1, dashboards or reports only: a dashboard-only product that releases nothing outside its own interface meets both. Define 0 as no buyer access to any usable output, 1 as dashboard/report access without export, and subsequent anchors by documented export or activation routes. Resolve overlaps before the scoring implementation chooses a value. | Reviewer opened: | structural | Data access: 0 = an opened source establishes no usable output; 1 = dashboard or report without structured export; undocumented access is n/e (R2B-06). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json data_access |
| R2-103 | a large-platform developer product described for context, not scored against specialist SDKs | Leaving developer products unbanded prevents a false ranking across unlike APIs and SDKs. However, excluding Google Maps Platform by owner size while admitting other Places APIs is a market-scope choice, not a buying-task distinction. Label cohort D as the selected specialist sample and show the platform alternative alongside comparable task attributes, or apply the same task-based admission rule to it. Do not claim that one 0–5 score or quotient is comparable across cohorts with different masks, units and anchors. | Reviewer opened: | structural | Places and maps APIs listed whoever owns them; cohort D labelled a selected specialist sample; Android's Geofencing API listed for comparison. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json developer cohort; data/product-register.json |
| R2-104 | design | Accept Reveal OOH measurement, Azira Measurement and Cuebiq Footfall Measurement in cohort A; Azira Audience, Cuebiq Audiences and Foursquare Proximity in B. Keep Mira Console as the Reveal service’s self-serve delivery interface, with API/managed service attributes, rather than creating interface-based duplicate products. Classify products by purchased task and document any separately purchasable SKU before splitting rows. | Reviewer opened: https://revealmobile.com/ https://www.azira.com/ https://cuebiq.com/ https://foursquare.com/products/proximity/ | structural | Register: Reveal OOH Measurement (Mira Console is its interface), Azira Measurement and Azira Audience, Cuebiq Footfall Measurement and Cuebiq Audiences, Foursquare Proximity entered by purchased task. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/product-register.json |
| R2-105 | SpendView™ delivers credit and debit card transaction insights | Advan’s five current named offers can be entered in the register, but do not score all five as interchangeable mobility-panel products. Patterns+ is explicitly a foot-traffic feed; SpendView is a transaction-panel product, with card-based coverage and optional integration into REI/REveal. Apply scope, source-type and applicability decisions per product; prohibit inheriting the vendor’s 45m-device panel into SpendView. Foursquare Spatial tools are geospatial analytics in C if the purchased task meets the brief; distinguish tool access from an underlying mobility dataset. | Source check G2 (24 October 2026, primary sources opened): R2-105 accepted with one correction: Advan lists six named offers, not five, because REP is missing. The product distinctions the reviewer draws are confirmed. https://advanresearch.com/ https://advanresearch.com/products/patternsplus https://advanresearch.com/products/spendview https://advanresearch.com/ | supported with qualification | Enter Advan's six current named offers (FIT, REI, REveal, SpendView, Patterns+ and REPerspectives) in the register, each with its own scope, source type and applicability decision. Patterns+ is a weekly US and Canada foot-traffic feed. SpendView is a card-transaction product (120M+ cards) and must not inherit the 45M-US-device location panel. Foursquare Spatial is analytics tooling: score it only if the purchased task meets the brief, and score any mobility dataset it carries, such as H3 Hub mobility layers, separately. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | design |
| R2-106 | design | Mobvista can be excluded because the bounded opened product menu did not show a location SKU. Lifesight’s opened current homepage supports MMM/incrementality positioning and no location product on that page, so move its seed row to listed/unconfirmed rather than asserting a settled market exit or retirement. A current menu’s omission does not date or establish cessation of a previously sold product. | Reviewer opened: https://www.mobvista.com/en/ https://lifesight.io/ | structural | Mobvista excluded (no location product); Lifesight listed as unconfirmed current, not as a dated exit. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/product-register.json |
| R2-107 | GroundTruth is now part of ZeroToOne.AI. | Accept T-Mobile ownership of Vistar and Blis and Rezolve ownership of the named Bluedot entities from filings. GroundTruth’s acquisition by ZeroToOne.AI was announced as completed on 25 March 2026, confirmed by the opened company and acquirer-counsel releases in R2-22; the underlying legal closing date and consideration were not established. For Foursquare retain announced deal plus current joined statement and unresolved closing evidence. Bluedot’s redirect alone does not establish current availability of the old standalone SDK/API; keep ownership and purchasability as separate fields. | Source check G3 (24 October 2026, primary sources opened): design: accepted the ownership fields. The Nelson Mullins release is dated 24 Apr 2026 and does not itself confirm the 25 Mar date; only the GroundTruth release does. https://www.groundtruth.com/article/zerotooneai-acquires-groundtruth/ https://www.nelsonmullins.com/insights/news/press_releases/nelson-mullins-advises-zerotoone-ai-in-acquisition-of-groundtruth https://www.sec.gov/Archives/edgar/data/1283699/000128369926000010/tmus-20251231.htm https://bluedot.io/ | supported with qualification | Ownership: T-Mobile owns Vistar and Blis (10-K), Rezolve owns the Bluedot entities (20-F), and ZeroToOne.AI announced on 25 March 2026 that it has acquired GroundTruth (company release; counsel Nelson Mullins confirmed its role on 24 April 2026; no legal closing date or consideration disclosed). Foursquare's acquisition by Infillion is announced and the target says it 'has joined', but no closing date was found. Ownership and current purchasability (for example, Bluedot's standalone SDK) are recorded as separate fields. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | design |
| R2-108 | Cannot fulfill buyer contract for postal areas | H7 postal ruling: intended fail-closed refusal is an implementation gap, and “by design” alone is not a registered exception to “could not be carried.” But the opened code does not document an actual attempted workflow or interoperability test. Therefore do not treat this code branch as observed failed execution. Keep H7 unresolved on this evidence; explicitly state that qualifying contradiction requires an attempted workflow/test, or report the sensitivity that a broader source-documented-capability interpretation would yield contradicted. Any exemption for all intentional scope limits must be logged as a rule amendment, not passed off as the original rule. Distinguish a connector’s unsupported signal from protocol-wide impossibility. | Reviewer opened: https://github.com/prebid/salesagent/blob/v2.0.0/src/adapters/gam/managers/targeting.py | supported as stated | None: author ruling recorded; H7 unresolved. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H7 |
| R2-109 | mobile audience data | Accept discovery of a distinct Broadsign Ads audience-targeting product in cohort B: the vendor describes venue profiling from active app SDKs and APIs and more than 190 personas for activation. It belongs separately from Broadsign footfall measurement and PerView audience measurement; do not inherit their evidence. The opened audience page does not establish US availability, US panel size/period or a current sale under the newer OutMoove brand. Add it to the register with those fields unresolved and score only if US eligibility and the required numeric-cell minimum are established. Do not award capability or validation from its claim of precise audience composition alone. | Source check G2 (24 October 2026, primary sources opened): R2-109 accepted with one addition: broadsign.com/broadsign-ads/ now 301-redirects to /outmoove/. That supports a probable rename the reviewer did not find, but no page confirms the rename or US availability of the Native Touch audience data. https://broadsign.com/audiences/ https://broadsign.com/audiences/ https://broadsign.com/broadsign-ads/ | supported with qualification | Add Broadsign's audience-targeting product to cohort B as a separate register entry. Broadsign describes venue profiling from active app SDKs and APIs, supplied by Native Touch, translated into over 190 personas. The broadsign.com/broadsign-ads/ URL now redirects to the OutMoove DSP page, but no page states that the Native Touch audience data is sold under OutMoove. US availability, US panel size and period remain unresolved, so the product is not scored until US eligibility and the numeric-cell minimum are established. It inherits no evidence from Broadsign's footfall or PerView measurement, and its 'precise audience composition' claim earns no capability or validation credit. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | design |
| R2-110 | Then check as many as you can of the 16 findings | The supplied digest has 292 findings and 12, not 16, marked exactly 'Internal verifier: contradicted': A-F22; B-F31; E-F31; F-F07/F19/F20/F24; H-F13/F16; I-F17/F23; U-F15. Reconcile the prompt with this frozen input; four are already among the 66 headlines, so the combined requested fact-check set contains 74 unique findings. | Reviewer opened: | supported as stated | The prompt's '16 contradicted' counted data-table and summary rows; 12 findings were contradicted. No data change. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | private review prompt only |
| R2-111 | Research cutoff (evidence freeze): 30 October 2026. | This review observes sources on 24 October 2026. Treat all current-status conclusions as provisional at that date. A future 30 October freeze cannot yet be certified; refresh time-sensitive ownership, products, laws, accreditations and protocol versions at the actual freeze without rewriting registered decision rules. | Reviewer opened: | structural | Every current-status conclusion is dated 24 October 2026 and re-read at the 30 October freeze (MRC lists, AdCP tags and registry, ownership, state laws, H5 product pages, Geopath). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | freeze checklist |
| R2-112 | triage comments by … | Literal, case-insensitive searches of all six supplied files found neither Ether Data nor Nathan Woodman in the digest or registers. Both names occur only in the prompt's preflight note, which describes public commenters and asserts no client relationship. No additional confidential-client exclusion list was supplied; this check does not identify undisclosed relationships or certify that every public source omits the names. | Reviewer opened: | supported as stated | None: the two names were not in the digest; the author's pre-send note was wrong about that. No client exposure. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | private review prompt only |
| R2-113 | Passages are truncated at 450 characters | The digest also truncates some verifier corrections with ellipses. Do not accept an incomplete proposed correction wholesale: retain the full verifier record and exact source version for ledger adjudication. This review evaluates the visible correction and reopened sources; omitted wording is unreviewed. | Reviewer opened: | supported as stated | None: the ledger is built from the full verifier records, not the truncated digest. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | process |
| R2-114 | Location – 6 Applicants; 1 Accredited | Add this historical MRC record: its September 2018 ACA presentation, slide 9, reports six location applicants and one accreditation. It does not name the service or define its metric. Reconcile this with the narrower 2020 first-since-2017-guidelines announcement before making a complete history or first-ever claim; do not invent a pre-2020 visit accreditation. | Source check G1 (24 October 2026, primary sources opened): R2-114: add. MRC's Sept 2018 ACA slide 9 shows 6 location applicants and 1 accredited (unnamed). Narrow any 'first-ever' framing of Foursquare 2020; the 2018 service is unidentified. https://acaweb.ca/en/wp-content/uploads/sites/2/2018/10/ACA-Executive-Forums-MRC.pdf | supported with qualification | MRC's own presentation to the ACA Executive Forums (26–27 September 2018, slide 9) reports 'Location – 6 Applicants; 1 Accredited' without naming the service or the metric. Any claim that Foursquare's July 2020 grant was the first MRC location accreditation must therefore be narrowed to 'first under the 2017 guidelines' as stated by the parties, or the 2018 accreditation must be identified. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | missing |
| R2-115 | will not use geofencing technology at or near Massachusetts healthcare facilities to infer the health status | Add Massachusetts AG's Copley Advertising settlement, announced 4 April 2017 and resolved by an Assurance of Discontinuance filed in Suffolk Superior Court. It bars specified healthcare geofencing uses in Massachusetts; do not describe it as a nationwide statutory ban. The official release was retrieved and read directly after the web opener returned 403. | Source check G6 (24 October 2026, primary sources opened): R2-115 confirmed via Wayback copy (mass.gov 403): 4/04/2017, AOD in Suffolk Superior Court, purpose-limited bar on one company, no payment; predates the 2023-26 window. https://www.mass.gov/news/ag-reaches-settlement-with-advertising-company-prohibiting-geofencing-around-massachusetts-healthcare-facilities https://web.archive.org/web/20220505002332id_/https://www.mass.gov/news/ag-reaches-settlement-with-advertising-company-prohibiting-geofencing-around-massachusetts-healthcare-facilities https://web.archive.org/web/20220505002332id_/https://www.mass.gov/news/ag-reaches-settlement-with-advertising-company-prohibiting-geofencing-around-massachusetts-healthcare-facilities | supported with qualification | On 4 April 2017 the Massachusetts Attorney General settled with Copley Advertising, which had geofenced reproductive-health and methadone clinics in five other states' cities to serve anti-abortion ads. Under an Assurance of Discontinuance filed in Suffolk Superior Court, Copley agreed not to use geofencing at or near Massachusetts healthcare facilities to infer anyone's health status, condition or treatment; no money was paid, and the order binds Copley only. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | missing |
| R2-116 | August 13, 2024 \| Press Release | Add Texas v. General Motors/OnStar as another location-data enforcement case: the August 2024 complaint alleges deceptive vehicle-data collection and sale to insurers. Distinguish allegations and filing from any later judgment or settlement; the opened announcement alone establishes no adjudicated violation or current case outcome. | Source check G6 (24 October 2026, primary sources opened): R2-116 accept with changes: date and filing confirmed; location rests on petition para. 28 (GPS lat/long), not the release; 1.5m (release) vs 1.8m Texans (petition); unresolved per GM 10-Q of 2026-10-20. https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-sues-general-motors-unlawfully-collecting-drivers-private-data-and https://www.texasattorneygeneral.gov/sites/default/files/images/press/General%20Motors%20Original%20Petition%20Filestamped.pdf https://www.sec.gov/Archives/edgar/data/1467858/000146785826000059/gm-20260930.htm | supported with qualification | On 13 August 2024 Texas sued General Motors and OnStar under its deceptive trade practices law (Montgomery County, cause 24-08-12392). The suit alleges that GM collected and sold vehicle 'Driving Data', including GPS latitude and longitude, to companies that built driving scores for insurers, without drivers' informed consent. The claims are allegations only; GM's 20 October 2026 quarterly filing reports only FTC and California settlements and says it is defending the remaining state suits. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | missing |
| R2-117 | $391.5 million multistate settlement; remains subject to court approval; October 31, 2025 \| Press Release | Add Google location-enforcement context absent from this digest: the 40-state settlement announced 14 November 2022 ($391.5m); California's $93m settlement announced 14 September 2023 (then proposed); and Texas's agreement signed and finalization announced 31 October 2025 ($1.375bn). The Texas sum resolves combined geolocation, incognito and biometric claims: do not attribute the full amount to location. The California announcement does not establish the final judgment's entry date. | Source check G6 (24 October 2026, primary sources opened): R2-117 accept with changes: all three amounts and dates confirmed at AG releases; add the 9 May 2025 in-principle date and the 24 Jan 2022 Victoria County filing; 'subject to court approval' belongs to California, not Minnesota; no court-entry dates established. https://ag.state.mn.us/Office/Communications/2022/11/14_Google.asp https://oag.ca.gov/news/press-releases/attorney-general-bonta-announces-93-million-settlement-regarding-google%E2%80%99s https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-finalizes-historic-settlement-google-and-secures-1375-billion-big-tech https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-secures-historic-1375-billion-settlement-google-related-texans-data | supported with qualification | State attorneys general have repeatedly penalised Google for location tracking that continued after users turned Location History off. Those actions are: a 40-state US$391.5m settlement announced 14 November 2022; California's US$93m settlement announced 14 September 2023 (a proposed stipulated judgment then awaiting court approval); and Texas's US$1.375bn settlement, agreed in principle on 9 May 2025 and signed by Google, as announced 31 October 2025. The Texas figure also resolves separate incognito-browsing and biometric claims, so it cannot be attributed to location alone. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | missing |
| R2-118 | searchable, central; Data brokers are required to register starting January 1, 2024.; Effective January 1, 2027 | The supplied digest checks California but contains no Texas, Oregon or Vermont registry records. Add those separate registers for entity/alias and reporting-year cross-checks, without treating registration as product accreditation or inferring unregistered status from a failed search. Texas's official index links its registry; Oregon links verification and opt-out records. Vermont's statute separates current registration rules from changes effective 1 January 2027. Individual Texas/Oregon/Vermont vendor entries were not verified in this review. | Source check G5 (24 October 2026, primary sources opened): R2-118: three registers added; Texas list read in full (4 of the ordered groups present, Mobilewalla/InMarket absent); Oregon and Vermont listings sit behind search forms and were not checked. https://www.sos.texas.gov/statdoc/data-brokers.shtml https://texas-sos.appianportalsgov.com/data-broker-registry https://dfr.oregon.gov/business/licensing/data-broker-registry/Pages/index.aspx https://legislature.vermont.gov/statutes/section/09/062/02446 | supported with qualification | Texas (Secretary of State, Bus. & Com. Code ch. 510), Oregon (DFR, HB 2052, registration required from 1 January 2024) and Vermont (9 V.S.A. 2446, revised from 1 January 2027 to require disclosure of precise-geolocation collection) keep their own data broker registers. The Texas register's 456 records on 24 October 2026 include Outlogic, Collective Data Solutions, Unacast, Inc. and Venntel but not Mobilewalla, InMarket or Kochava. Registration is a filing, not accreditation, and absence does not show a company is unregistered. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | missing |
| R2-119 | missing | The specific validation sources reopened in this review cover historical polling places, parks and a trail site; Azira publishes vendor accuracy assertions without a reproducible held-out method. I identified no additional independent, multi-site retail comparison of current projected visits against matched door counts or transactions in those opened sources. Record this as an unresolved evidence gap, not proof that no such validation exists, and do not recast historical demographic-bias evidence as a current retail accuracy test. | Reviewer opened: https://arxiv.org/pdf/2011.07194 https://par.nsf.gov/servlets/purl/10451340 https://conservancy.umn.edu/server/api/core/bitstreams/fdf3b6a9-7824-4e64-b330-f6392f4491b4/content https://business.azira.com/azira-location-measurement-and-attribution-methodology | supported as stated | None: recorded as a bounded evidence gap (no independent multi-site retail comparison found). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-B-F31 |
| R2B-01 | exposed-only visitation products are still scored on causal design, because the buyer's task is to learn whether the media worked. | Add: 'Cohort A scores causal capability for assessing media effects. A documented exposed-only default scores 1 for causal design when it supplies no causal comparison; it is labelled descriptive, without implying a causal claim was made. Score 0 requires the separately documented false or failed causal claim in the zero anchor. Unknown control design is n/e, not 1.' For C, register raw-movement and visit-estimate task masks before scoring, or restrict the composite to a common applicable core and show visit/projection scores separately. List transaction-only and general geospatial tools unless a separately purchasable in-scope task is established. Move named places/maps products rather than excluding every product of HERE or Mapbox by company name; a separately established geofencing SKU should face the same task rule. Keep the full A diagnostic table, but do not silently change H3's p… | Reviewer opened: | structural | Causal design wording adopted (exposed-only default = 1, labelled descriptive; unknown = n/e); cohort C split into C1 and C2; transaction-only and general geospatial tools listed; places and maps products moved by product. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json causal_design, cohorts |
| R2B-02 | A product that documents observed-only counts, and makes no population claim, gets n/e on projection | Use: 'A documented observed-only output that makes no population inference has projection_status=not_applicable_to_output, with source and reason. A product claiming estimates, normalisation, extrapolation or population totals cannot receive that status merely because its projection method is undisclosed; that case is n/e.' The simplest common-composite fix is to remove projection from the GQ mask for every A and C row and retain it as a separately scored module for products making population claims. Alternatively create preregistered output subcohorts and compute quotients only within a common mask. Neither option requires invented observed-only numeric anchors. Do not call omission neutral or claim the remaining missing-evidence incentive has disappeared; retain deletion and missingness sensitivity for the common core. Log the changed masks/formula as a dated post-research, pre-scorin… | Reviewer opened: | structural | Projection removed from every A and C mask and kept as a separate module (modules.projection). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json |
| R2B-03 | A daily unique-device figure counts as a lower bound against the monthly thresholds | Use: 'A stated daily unique-device count establishes monthly uniques of at least that count only on the same defined population, geography and deduplication basis. Score the highest minimum tier it proves and label the cell lower-bound; do not restate it as measured MAU or infer an upper bound. A figure below the first threshold means only the demonstrated lower bound is below that threshold. Daily API requests may be summed over the documented month, or a documented mean over every day of that month may be multiplied by its day count. A peak, typical day or unspecified run-rate is an attribute, not an actual monthly aggregate.' Remove generic annualisation. Keep measured annual campaign counts; put partial-period annualised run-rates and historical cumulative averages in separate attributes. Apply age caps to the observation period, not merely the page's publication date. Either name t… | Reviewer opened: | structural | Lower-bound rule for daily uniques, monthly summation for API requests, no generic annualisation, age cap on the observation period, dimension renamed. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json scale |
| R2B-04 | Accreditation provides independent confirmation of competence. | Keep pending_audit as an auditor-confirmed status, not a positive assurance result; retain any current completed result, otherwise use self-attestation=1 or n/e as appropriate. One transparent ladder is: 0=published material failed criterion relevant to the assessed product/metric, still applicable to its current version; 1=self-attestation only; 2=completed commissioned partial-scope assessment; 3=completed commissioned full-metric assessment OR independent partial-scope assessment, with the incomparability disclosed; 4=completed independent assessment of all relevant product metrics in at least one sold environment, with other sold environments explicitly excluded; 5=completed independent assessment covering all relevant metrics and sold environments. Require public scope, date, result and tested version at 2–5. Voluntary withdrawal, lapse or a historical failure is a status/history f… | Reviewer opened: https://www.iso.org/certification.html | structural | Reviewer's ladder adopted for B, C1, C2 and D; cohort A ladder rebuilt on the same principles; pending audits and withdrawals are flags. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json assurance, cell_rules |
| R2B-05 | This Order will terminate 20 years from the date of its issuance | Preferred wording: 'Report current_controls_score from anchors 1–5, recent_event_flag and event type/date/scope/entity, and active_legal_obligations separately. If the author retains the three-year cap, show risk_adjusted_score=min(current_controls_score,1) beside the uncapped score, label it an author-defined risk adjustment, and report its effect on GQ; do not label it current controls. Date a legal event from final order/entered judgment and a breach from confirmed discovery, retain conduct dates separately, and do not reset the clock for a repeated press release. Define current_failure=established only where a dated primary regulator/court finding or a company/regulator incident update expressly identifies an ongoing material prohibited practice, uncontained access, or a specified overdue remedial act affecting this product during the assessed current period. Record the exact failur… | Reviewer opened: https://www.ftc.gov/system/files/ftc_gov/pdf/X-ModeSocialDecisionandOrder.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf | structural | Current controls and the author's risk-adjusted value reported side by side; event dating and current-failure test adopted; n/e never imputed. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json privacy |
| R2B-06 | 0 means no buyer access to any usable output; 1 means dashboard or report access without export. | Use: '0: an opened source explicitly establishes that the buyer receives no usable output. 1: the buyer can access a dashboard or human-readable report, including a static report file, but no structured-data export or external activation route is documented. n/e: the opened record does not establish the buyer's access. Higher tiers require the named, documented operational routes already defined.' Apply the normal availability and evidence caps; no other ladder change is required to resolve this particular overlap. | Reviewer opened: | structural | Reviewer's wording adopted. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json data_access |
| R2B-07 | Our Data is Available to License | Publish the original rules and a dated amendment/clarification diff, with rationale and old/new verdicts on the same frozen coding. H3 decision tree: if D<4, unresolved; otherwise R>D/2 => supported, M+E>D/2 => contradicted, else unresolved. D is codable documented defaults, N is the full eligible-product population; publish R/M/E/unknown and D/N, and retain the broad registered randomised-or-holdout category. Explicitly adjudicate whether truthful non-lift reporting is outside the lift claim or a separate diagnostic category; do not silently count it as mislabeled lift. H4: let b and s be products with a documented primary source in the two named families; if b+s<4 or b=s, unresolved, otherwise compare b and s. List other sources/blends/unknowns separately and identify this eligibility restriction as a substantive population clarification. H5: first test whether any verified current US… | Reviewer opened: https://outlogic.io/ https://www.ftc.gov/system/files/ftc_gov/pdf/X-ModeSocialDecisionandOrder.pdf | structural | Clarifications recorded with original rules preserved, decision trees as proposed, and the H5 same-reading claim restated as precedence over an overlapping original rule. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H3, H4, H5 clarifications |
| R2B-08 | Outlogic: we may receive and sell the precise latitude and longitude coordinates of your mobile device. CDS: precision location (protected by Privacy Block™). | State: under the original contradicted criterion, current publicly documented precise-data licensing offers supply a counterexample; the clarified precedence assigns contradicted despite remaining company-specific unknowns. Otherwise all five verified exits/no relevant offers yields supported; otherwise unresolved. Preserve the original wording and report its overlapping branch issue. Define verified current sale consistently with the registered documented-product-set test: verification of a current public offering to sell/license, not proof of a consummated transaction. If actual paid transactions are now required, that is a new evidentiary test and the same-verdict claim must be re-evaluated. Do not interpret continued licensing as an order breach. | Reviewer opened: https://outlogic.io/ https://www.collectivedata.io/ https://www.unacast.com/privacy-policy https://www.ftc.gov/system/files/ftc_gov/pdf/X-ModeSocialDecisionandOrder.pdf https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf | structural | A current public offer, not a consummated transaction, is the test; a preceding-year sale notice alone is not a current offer. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H5 |
| R2B-09 | Step 3: We create a modeled control group | Code the documented product method as matched_or_synthetic. Record alternative designs as not documented in the page screened, rather than claiming none is available. | Reviewer opened: https://docs.foursquare.com/campaign-products/docs/what-is-attribution | supported with qualification | Code the documented product method as matched_or_synthetic. Record alternative designs as not documented in the page screened, rather than claiming none is available. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-10 | Build a statistical model of expected visitation based on historical patterns — a synthetic control | Code the current InMarket visit-lift offering as matched_or_synthetic, cite the dated public method explanation, and update the LCI product-name mapping explicitly. Keep Visit Stream's exposed conversion feed separate from visit-lift reporting. | Reviewer opened: https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ https://inmarket.com/solutions/measurement/ https://go.inmarket.com/hubfs/sellsheets/NinthDecimal-LCI%20Powered%20by%20InMarket.pdf https://www.globenewswire.com/news-release/2019/07/11/1881759/0/en/ninthdecimal-launches-first-multi-touch-attribution-solution-for-foot-traffic-measurement.html https://lciadminconsole.inmarket.com/ | supported with qualification | Code the current InMarket visit-lift offering as matched_or_synthetic, cite the dated public method explanation, and update the LCI product-name mapping explicitly. Keep Visit Stream's exposed conversion feed separate from visit-lift reporting. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-11 | GroundTruth uses last-touch attribution to credit visits; [Projected Visits:] panel made up of impressed devices | Count one exposed_only row in the fixed cohort-A design table. For the registered visit-lift-claim reading, report separately that these pages provide no lift claim; show the result with this row excluded rather than silently equating cohort membership with a claim of causal lift. | Reviewer opened: https://help.groundtruth.com/hc/en-us/articles/360000695267-Observed-Visits https://help.groundtruth.com/hc/en-us/articles/12770162930707-Projected-Visits | supported with qualification | Count one exposed_only row in the fixed cohort-A design table. For the registered visit-lift-claim reading, report separately that these pages provide no lift claim; show the result with this row excluded rather than silently equating cohort membership with a claim of causal lift. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-12 | Smart Holdout Groups uses machine learning technology to create statistically balanced target & control groups of zip codes | Use randomised_or_holdout with the explicit attribute 'geographic holdout; randomization not documented'. Withdraw the earlier default objection for this specifically named product, while retaining it for any inference about all Blis campaigns or a different legacy product. | Reviewer opened: https://blis.com/omnichannel-measurement/ | supported with qualification | Use randomised_or_holdout with the explicit attribute 'geographic holdout; randomization not documented'. Withdraw the earlier default objection for this specifically named product, while retaining it for any inference about all Blis campaigns or a different legacy product. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-13 | A synthetic control group is constructed as a counterfactual in this quasi-experimental setup. | Code matched_or_synthetic for the Net Lift OOH measurement service delivered through Mira and the other service modes. Do not require the word 'default' when the general workflow already specifies the calculation. | Reviewer opened: https://revealmobile.com/solutions/measurement | supported with qualification | Code matched_or_synthetic for the Net Lift OOH measurement service delivered through Mira and the other service modes. Do not require the word 'default' when the general workflow already specifies the calculation. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-14 | Randomized PSA-based control: This is the default approach | Code randomized PSA as the documented default and synthetic control as available alternatively. Do not recode the default merely because the methodology permits another design. | Reviewer opened: https://business.azira.com/azira-location-measurement-and-attribution-methodology | supported as stated | None: the reviewer re-opened the source and agrees with the finding (and, where applicable, the internal verifier). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-15 | At the core of the new dashboard is Adsquare’s proprietary Control Condition methodology | Count matched_or_synthetic among method-disclosed rows if admitted to the US cohort. Retain a separate US-availability flag and describe the category as vendor-documented standard methodology rather than an independently verified implementation. | Reviewer opened: https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://adsquare.com/resources/newsroom/adsquare-redefines-campaign-measurement-with-unified-attribution-dashboard | supported with qualification | Count matched_or_synthetic among method-disclosed rows if admitted to the US cohort. Retain a separate US-availability flag and describe the category as vendor-documented standard methodology rather than an independently verified implementation. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-16 | Our incrementality methodology uses causal machine learning to simulate a randomized control trial, isolating true campaign lift through behaviorally matched control groups. | Count matched_or_synthetic. Label product-wide vendor method evidence separately from case-study evidence; do not code randomized_or_holdout from the words simulate a randomized control trial. | Reviewer opened: https://cuebiq.com/measurement/ https://cuebiq.com/blog/advanced-measurement-targeting-platform/ https://cuebiq.com/blog/ctv-ltv-attribution/ | supported with qualification | Count matched_or_synthetic. Label product-wide vendor method evidence separately from case-study evidence; do not code randomized_or_holdout from the words simulate a randomized control trial. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-17 | Vistar supports foot traffic studies through trusted partners including FourSquare, InMarket and Cuebiq | Keep the fixed Vistar row as not_documented for a single default. Record provider and metric per study; avoid counting Vistar as a separate independent method implementation when the same supplier is already a row. | Reviewer opened: https://www.vistarmedia.com/blog/dooh-measurement-solutions https://www.vistarmedia.com/partner-inmarket https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://cuebiq.com/blog/advanced-measurement-targeting-platform/ | supported with qualification | Keep the fixed Vistar row as not_documented for a single default. Record provider and metric per study; avoid counting Vistar as a separate independent method implementation when the same supplier is already a row. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-18 | Synthetic control groups are modeled using a privacy-first, third-party identity graph | Admit a named Kochava DOOH outcomes-measurement service conditionally on US availability and purchased-task fit; retain not_documented for the default, with MediaLift matched_or_synthetic recorded as a selectable design. Replace any broad no-method statement with this specific method disclosure. | Reviewer opened: https://www.kochava.com/ https://www.kochava.com/product/strategic-services/ooh-dooh-advertising-measurement/ | supported with qualification | Admit a named Kochava DOOH outcomes-measurement service conditionally on US availability and purchased-task fit; retain not_documented for the default, with MediaLift matched_or_synthetic recorded as a selectable design. Replace any broad no-method statement with this specific method disclosure. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-19 | “Compare exposed audiences to matched control groups”; “Determine incremental visit and spend from your campaigns”; “We track both groups to your destination.” | Count matched_or_synthetic. Keep visit and spend measurement bases separate; the cardholder control description does not establish that every arrival count is a transaction count. | Reviewer opened: https://www.arrivalist.com/media-insights | supported with qualification | Count matched_or_synthetic. Keep visit and spend measurement bases separate; the cardholder control description does not establish that every arrival count is a transaction count. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-20 | PlaceIQ proactively holds out 10% of its users from ever receiving an ad. | Preserve the fixed row with current default not_documented. Record the historical holdout separately, including unknown assignment mechanism and current availability. Do not count it as a current randomized or holdout default. | Reviewer opened: https://www.precisely.com/about-us/placeiq-is-now-part-of-precisely/ https://www.precisely.com/product/precisely-audiences/ https://www.precisely.com/product/precisely-placeiq/ https://mmaglobal.com/files/a-b_control_case_study_final.pdf | supported with qualification | Preserve the fixed row with current default not_documented. Record the historical holdout separately, including unknown assignment mechanism and current availability. Do not count it as a current randomized or holdout default. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-21 | Build better audiences, measure offline marketing attribution, and study consumer visits. | Keep the requested fixed row as not_documented and mark cohort-A admission unconfirmed. Use a cohort-C data row for the presently documented offer unless a named current measurement service and reporting method are established. | Reviewer opened: https://www.unacast.com/ https://www.unacast.com/industries/adtech-martech https://www.unacast.com/methodology https://www.unacast.com/post/real-time-location-data https://www.unacast.com/post/store-visit-intelligence | supported with qualification | Keep the requested fixed row as not_documented and mark cohort-A admission unconfirmed. Use a cohort-C data row for the presently documented offer unless a named current measurement service and reporting method are established. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-22 | The control group comes from Lifesight’s lookalike algorithm. | Preserve the fixed row and list current-cutoff default as not_documented; record historical matched_or_synthetic separately. If current continuity and US availability are confirmed, reclassify the default to matched_or_synthetic and show the resulting denominator sensitivity. Do not equate omission from the current menu with a dated exit. | Reviewer opened: https://lifesight.io/ https://docs.lifesight.io/docs/what-is-lifesight https://support.lifesight.io/en/articles/3026093-overview https://support.lifesight.io/en/articles/3003595-measurement-report https://lifesight.io/blog/visit-lift-measurement/ | supported with qualification | Preserve the fixed row and list current-cutoff default as not_documented; record historical matched_or_synthetic separately. If current continuity and US availability are confirmed, reclassify the default to matched_or_synthetic and show the resulting denominator sensitivity. Do not equate omission from the current menu with a dated exit. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-E-F28; data/hypotheses.json H3 |
| R2B-23 | H3-recount | Publish the fixed-set capability count and a separate registered lift-claim count. GroundTruth’s two named metrics do not claim lift in their opened definitions, so the stricter lift-claim reading excludes that row: six matched/synthetic versus two holdouts among eight documented defaults (75%), still contradicted. If the current InMarket family cannot be mapped to the exact legacy LCI row, mark that row unknown: fixed-set six of eight nonholdout, and strict lift-claim five of seven matched, both still contradicted. Lifesight’s historical lookalike method is available evidence but not a current-cutoff vote without continuity/admission evidence. The change from the prior unresolved reading follows better sources and corrected method coding, not the four-product/tie clarification. Keep the five unknown current defaults visible rather than coding them as matched. Adsquare’s US admission is… | Reviewer opened: https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ https://inmarket.com/solutions/measurement/ https://go.inmarket.com/hubfs/sellsheets/NinthDecimal-LCI%20Powered%20by%20InMarket.pdf https://www.globenewswire.com/news-release/2019/07/11/1881759/0/en/ninthdecimal-launches-first-multi-touch-attribution-solution-for-foot-traffic-measurement.html https://help.groundtruth.com/hc/en-us/articles/360000695267-Observed-Visits https://help.groundtruth.com/hc/en-us/articles/12770162930707-Projected-Visits https://blis.com/omnichannel-measurement/ https://revealmobile.com/solutions/measurement https://business.azira.com/azira-location-measurement-and-attribution-methodology https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://adsquare.com/… | contradicted | H3 reads contradicted on the recount (7 of 9; strict 6 of 8), confirmed product by product in G7. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/hypotheses.json H3; LOC-E-F28 |
| R2B-24 | Gravy Analytics was informed about the incident on January 4, 2025. ... misappropriated access key ... obtained some files | Replace inaccessible document ID 25485205 with working mirror ID 25485945. Attribute discovery, suspected vector and limited file acquisition to the company. Keep occurrence date, duration, contents and submission date unknown. Record intruder assertions and researcher observations separately with independent sources. | Reviewer opened: https://www.documentcloud.org/documents/25485945-datatilsynet-unacast-security-incident-notification-redacted/ https://s3.documentcloud.org/documents/25485945/datatilsynet-unacast-security-incident-notification-redacted.pdf?t=1736802348944 | supported with qualification | The internal verifier's wording of LOC-H-F08 stands: Gravy learned of the incident from the intruder on 4 January 2025, suspected a misappropriated AWS key, the intruder obtained some files, and processing was restored by 9 January; the notice is undated and the intruder's volume claims are unverified. The reviewer was wrong that document 25485205 is inaccessible: both DocumentCloud ids open with the same file; 25485205 (NRK's upload) stays the primary citation and 25485945 is recorded as a mirror. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-H-F08 |
| R2B-25 | Crowley, October 20, 2026: Almost exactly a month ago, Foursquare was acquired. Webb, September 24, 2026: Foursquare sold. To an AdTech company called Infillion. | Register Infillion ownership as completed according to dated participant statements by October 20, 2026; record exact closing date unverified. Keep September 18 as announcement date only. The September 24 shareholder account can be a supplementary earlier completion report, not an invented statutory closing date. If the register permits only a closing release/filing as settlement evidence, retain completion reported by participants as a qualified status instead of asserting pending. | Reviewer opened: https://teendrama.substack.com/p/some-final-words-on-foursquare https://buttondown.com/rickwebb/archive/good-morning-hello-how-are-you-1820/ https://infillion.com/blog/infillion-to-acquire-foursquare-closing-the-loop-between-advertising-and-real-world-results/ | supported with qualification | Completion reported by participants by 20 October 2026; legal closing date unverified (see LOC-U-F02 final wording). The board seat Crowley mentions was his pre-sale seat, not one kept after the deal. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-U-F02; data/product-register.json |
| R2B-26 | Details are available to the owner or operator. | Replace the blanket access blocker with: “The public registry list was opened and contains 34 registered agents on 24 October 2026. Geo/DOOH test results, per-agent spec versions and badge verification modes remain independently unverified from the public sources successfully opened.” Retain unresolved for those detailed claims. To settle them, preserve dated raw per-agent compliance responses or an accessible public conformance export; do not infer zeros from restricted detail views. | Reviewer opened: https://agenticadvertising.org/api/registry/agents?limit=500 https://agenticadvertising.org/registry/?tab=agents https://agenticadvertising.org/registry/?tab=agents#agent/sales/gotom-sandbox | supported with qualification | The list API returns 34 agents. The reviewer was wrong that per-agent compliance detail is owner-only: the public compliance endpoint reproduces LOC-J-F07 (11 badges, all spec mode; no DOOH specialism; geo storyboards untested on two agents; three agents on 3.2.6). Live data: re-read at the 30 October freeze. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-J-F07 |
| R2B-27 | minimum duration of 30 days and 1 million impressions per month | Keep the exact LCI minimum-sample and significance claims unresolved pending an accessible dated method document or a reproducible public glossary copy. Record the Choozle integration minimum only under its own product/metric scope. Do not promote search snippets of the glossary into opened evidence, or treat these numerical rules as universal across InMarket outputs. | Reviewer opened: https://lciadminconsole.inmarket.com/ https://inmarket.com/solutions/measurement/ https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ https://go.inmarket.com/hubfs/sellsheets/NinthDecimal-LCI%20Powered%20by%20InMarket.pdf https://go.inmarket.com/lcihij23 https://go.inmarket.com/insights https://help.choozle.com/offline-attribution https://choozle.com/wp-content/uploads/Choozle_offline-attribution.pdf | supported with qualification | InMarket's minimum-sample and significance thresholds sit in a glossary behind a sign-in and are marked unverified from public sources; Choozle documents a different minimum (30 days and 1 million impressions a month) for its InMarket Visit Stream integration only (see LOC-C-F24 final wording). | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-C-F24 |
| R2B-28 | February Digital GroundTruth Revoke accreditation | MRC placed GroundTruth's Location, Place and Visit dataset on a six-month hiatus in September 2023. In its July 26, 2024 communication, MRC reported revocation following withdrawal from accreditation. The row is labelled February despite appearing in the April–June report; the effective month remains unconfirmed. Do not substitute July as the action date or assert a corrected month without another MRC record. | Reviewer opened: https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/101623_MRC_Accreditation_Status_Communication.pdf https://mediaratingcouncil.org/sites/default/files/News/MRC%20Accreditation%20Status%20Quarterly%20Update_Q2%2024%207.26.pdf | supported with qualification | MRC placed GroundTruth's Location, Place and Visit dataset on a six-month hiatus in September 2023. In its July 26, 2024 communication, MRC reported revocation following withdrawal from accreditation. The row is labelled February despite appearing in the April–June report; the effective month remains unconfirmed. Do not substitute July as the action date or assert a corrected month without another MRC record. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | R2-04 |
| R2B-29 | a statement about POIs rather than individuals | Keep the historical SafeGraph POI-coverage finding. Do not present its turnout-estimation RMSE as a vendor-reported projection error, an individual-level demographic result, or a current-product failure. Separate that correction from the hypothesis adjudication. | Reviewer opened: https://arxiv.org/pdf/2011.07194 | supported with qualification | Keep the historical SafeGraph POI-coverage finding. Do not present its turnout-estimation RMSE as a vendor-reported projection error, an individual-level demographic result, or a current-product failure. Separate that correction from the hypothesis adjudication. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | R2-12 |
| R2B-30 | R2-17 | Remove the default-status objection for Smart Holdout Groups and code geographic holdout under the registered inclusive category. Keep the cross-stream source-scope reconciliation and randomization qualification. | Reviewer opened: https://blis.com/omnichannel-measurement/ | supported with qualification | Remove the default-status objection for Smart Holdout Groups and code geographic holdout under the registered inclusive category. Keep the cross-stream source-scope reconciliation and randomization qualification. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | R2-17 |
| R2B-31 | R2-30 | Downgrade the prior blanket contradiction to a qualification resolved through the product table. Preserve the distinction between two holdouts and two randomized methods, the prohibition on changing the registered category, and the separate GroundTruth lift-claim-scope issue. Use the completed B recount for the final denominator. | Reviewer opened: https://blis.com/omnichannel-measurement/ https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://revealmobile.com/solutions/measurement | supported with qualification | Downgrade the prior blanket contradiction to a qualification resolved through the product table. Preserve the distinction between two holdouts and two randomized methods, the prohibition on changing the registered category, and the separate GroundTruth lift-claim-scope issue. Use the completed B recount for the final denominator. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | R2-30 |
| R2B-32 | R2-47 | Use supported_with_qualification and C1’s working source and wording; forensic scope remains unknown. | Reviewer opened: https://s3.documentcloud.org/documents/25485945/datatilsynet-unacast-security-incident-notification-redacted.pdf?t=1736802348944 | supported with qualification | The internal verifier's wording of LOC-H-F08 stands: Gravy learned of the incident from the intruder on 4 January 2025, suspected a misappropriated AWS key, the intruder obtained some files, and processing was restored by 9 January; the notice is undated and the intruder's volume claims are unverified. The reviewer was wrong that document 25485205 is inaccessible: both DocumentCloud ids open with the same file; 25485205 (NRK's upload) stays the primary citation and 25485945 is recorded as a mirror. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-H-F08 |
| R2B-33 | "count":34 | Upgrade only the 34-agent list count from unresolved to verified. Leave the compliance/test/badge components unresolved and leave H7 unresolved. No downgrade or withdrawal is warranted for my other I/J findings on the newly opened registry record; the postal-area distinction remains a design clarification rather than evidence of an observed failed workflow. | Reviewer opened: https://agenticadvertising.org/api/registry/agents?limit=500 | supported with qualification | The list API returns 34 agents. The reviewer was wrong that per-agent compliance detail is owner-only: the public compliance endpoint reproduces LOC-J-F07 (11 badges, all spec mode; no DOOH specialism; geo storyboards untested on two agents; three agents on 3.2.6). Live data: re-read at the 30 October freeze. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-J-F07 |
| R2B-34 | R2-63 | Replace the undifferentiated unresolved completion status with completion reported by dated participants by October 20, 2026, exact legal closing date unverified. Keep the level and provenance of evidence explicit. | Reviewer opened: https://teendrama.substack.com/p/some-final-words-on-foursquare https://buttondown.com/rickwebb/archive/good-morning-hello-how-are-you-1820/ | supported with qualification | Completion reported by participants by 20 October 2026; legal closing date unverified (see LOC-U-F02 final wording). The board seat Crowley mentions was his pre-sale seat, not one kept after the deal. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | LOC-U-F02; data/product-register.json |
| R2B-35 | R2-78 | Report historical SafeGraph polling-place coverage as adverse evidence. Make its formal H2 contradiction conditional on those checks. Current retail-visit agreement stays unresolved where matched validation is absent; do not assign a single verdict across products and vintages. | Reviewer opened: https://arxiv.org/pdf/2011.07194 | supported with qualification | Report historical SafeGraph polling-place coverage as adverse evidence. Make its formal H2 contradiction conditional on those checks. Current retail-visit agreement stays unresolved where matched validation is absent; do not assign a single verdict across products and vintages. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | R2-78 |
| R2B-36 | R2-79 | Replace the previous default-evidence objection with the adjudicated B method table and count. Retain uncertainty only for genuinely uncodable product methods or scope mismatches, not because a general workflow lacks a default label. The completed fixed-set recount is six matched/synthetic, two holdouts, one exposed-only and five unknown; the stricter lift-claim count is six of eight matched, hence contradicted. | Reviewer opened: https://docs.foursquare.com/campaign-products/docs/what-is-attribution https://revealmobile.com/solutions/measurement https://blis.com/omnichannel-measurement/ https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard https://cuebiq.com/blog/advanced-measurement-targeting-platform/ https://www.arrivalist.com/media-insights | supported with qualification | Replace the previous default-evidence objection with the adjudicated B method table and count. Retain uncertainty only for genuinely uncodable product methods or scope mismatches, not because a general workflow lacks a default label. The completed fixed-set recount is six matched/synthetic, two holdouts, one exposed-only and five unknown; the stricter lift-claim count is six of eight matched, hence contradicted. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | R2-79 |
| R2B-37 | and A includes visitation reporting alongside lift. | Narrow R2-89 to the retained C task-mask mismatch and any genuinely inapplicable output dimensions. Accept A's common causal-capability dimension under the A1 wording above, with honest descriptive output distinct from a false causal claim and unknown design distinct from no control. Preserve the separate H3 population/default-coding qualification; do not infer that every A product makes a visit-lift claim. | Reviewer opened: | supported as stated | None beyond R2B-01. | Shared rules: every external finding is checked against its primary source before it changes anything; bound absence claims; evidence class follows funding and design. | data/scoring-rules.json |
| R3-01 | Chapter 11 capability matrix; data/capability-matrix.json, row kochava-dooh-advertising-measurement, privacy | The privacy risk cap is applied without establishing that the event involved this measurement product's location data. The cell expressly substitutes the defendant's identity for the product-data link required by the frozen rule. | The privacy definition in data/scoring-rules.json requires an event 'involving the product’s location data'; the Kochava cell says 'the cap rests on the operating entity being a defendant' and that 'this product’s location source is undisclosed'. Its scale cell separately refuses to inherit the former data-broker business's volume because that business is now Collective Data Solutions. The recorded privacy value is 1 versus recorded current controls 3; keeping 3 and removing this cap alone would change Kochava's recomputed GQ from 33.5611 to 37.3611 and Least to Partly documented, although the current-controls score also needs its product scope checked. | contradicted | Kochava DOOH measurement: risk cap removed (no established link between the order and this product's data); the order kept as an event flag; current controls re-anchored to 2 (company-level). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 11 capability matrix; data/capability-matrix.json, row kochava-dooh-advertising-measurement, privacy |
| R3-02 | Chapter 11 capability matrix; data/capability-matrix.json and source/data/scores.json, GroundTruth visit measurement, coverage | Coverage 3 is awarded on an assumed US share of a combined US–Canada place count. No cited figure establishes the required 100,000 US places, so the numeric threshold is not supported by the recorded evidence. | The measurement coverage anchor 3 requires '100,000 or more US places with polygons or geofences documented'. The cell cites 4.8 million Blueprinted locations across the United States and Canada, acknowledges 'The US share is not stated', then reasons that 100,000 US polygons follows from 'any US share of at least about 2%'. The cell supplies no evidence for that minimum share. | contradicted | GroundTruth visit measurement coverage 3 -> n/e: no US-only place count is documented. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 11 capability matrix; data/capability-matrix.json and source/data/scores.json, GroundTruth visit measurement, coverage |
| R3-03 | Chapter 11 matrix privacy cells: azira-measurement and foursquare; InMarket measurement uncapped controls; data/capability-matrix.json and source/data/scores.json | Several current-controls scores of 3 are supported only by company-level policies although anchor 3 requires at least one product-specific consent route or suppression control. The rationales themselves say the product-specific condition is unmet, so the scores exceed their frozen anchor. | The Azira measurement cell assigns 3 but says 'neither the consent route nor the suppression is specific to Azira Measurement'; Foursquare Attribution assigns 3 but says neither is documented specifically for Attribution. InMarket measurement records uncapped current controls 3 while saying 'None of these is documented specifically for the measurement product'. Privacy anchor 2 is the company-policy tier and anchor 3 requires a product-specific control. These are applications of the final rule, not challenges to the closed R2/R2B policy decision. | contradicted | Privacy current controls 3 -> 2 for Foursquare Attribution, Azira Measurement and InMarket visit-lift measurement (company-level controls are anchor 2); InMarket keeps its capped value of 1. A scan of every privacy cell found no other such case. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 11 matrix privacy cells: azira-measurement and foursquare; InMarket measurement uncapped controls; data/capability-matrix.json and source/data/scores.json |
| R3-04 | Chapter 11 matrix; Placer.ai documented production volume; data/capability-matrix.json and source/data/scores.json | Placer's volume score 3 bypasses the rule for a figure without a definition or stated observation period. The rationale admits that the only monthly figure has an undefined active-device basis, yet awards the 20-million tier before the marketing cap. | The cell uses a vendor claim relayed by the February 2026 city fact sheet, 'over 20 million monthly active users', expressly says 'active' is undefined, and concludes that no definition or period is stated. LOC-PLACER-S25 also records period 'not stated'. The volume definition requires a stated observation period, and anchor 1 covers a figure with no period or definition; grade D's cap of 3 cannot raise the anchor. Changing this cell 3→1 alone recomputes Placer's GQ from 59.5786 to 57.7214 (60→58); it remains Most documented. | contradicted | Placer.ai documented production volume 3 -> 1: the only monthly figure has no definition or period. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 11 matrix; Placer.ai documented production volume; data/capability-matrix.json and source/data/scores.json |
| R3-05 | Chapter 11 matrix; method_disclosure for placer, quadrant and veraset-movement; data/capability-matrix.json | The disclosure element count credits update cadence as an entire element even where uncertainty reporting is expressly absent. That changes three-element records into four-element records and raises them from anchor 2 to anchor 3. | The frozen method-disclosure definition makes element (7) 'update cadence and uncertainty reporting'. Placer's rationale counts four elements, the fourth being cadence, and states there is no per-estimate uncertainty; Quadrant also counts four including cadence and history while stating there is no uncertainty reporting. Veraset likewise counts four, with its final element only daily delivery/history. Excluding the incomplete element leaves three elements in each case, for which anchor 2 applies. | contradicted | Method disclosure 3 -> 2 for Placer.ai, Quadrant and Veraset Movement: element 7 needs uncertainty reporting as well as cadence. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 11 matrix; method_disclosure for placer, quadrant and veraset-movement; data/capability-matrix.json |
| R3-06 | Chapter 7 standfirst, inshort and 'Four states now ban the sale'; chapter 0 inshort; figure-state-laws / data/figure-data.json states; data/rules.json; plain/index.html paragraph beginning 'Four states now ban' | The national total of four states with precise-geolocation sale bans omits New Jersey's enacted and immediately effective 30 June 2026 ban. The paper itself records GroundTruth switching off New Jersey features, but never resolves the underlying statute. | Chapter 7 names only Maryland, Oregon, Virginia and Connecticut, and figure-data.states.sale_ban_states is 4. Opened primary enacted text https://pub.njleg.state.nj.us/Bills/2026/PL26/25_.HTM : P.L.2026 c.25 §1 amends C.56:8-166.12(a)(6) to prohibit sensitive-data sales regardless of consumer-count threshold; §§2–3 include 1,750-foot precise geolocation and prohibit data brokers/collectors selling or licensing sensitive data, subject to listed exemptions; §8 makes the act effective immediately apart from the registry provision; approved 30 June 2026. | contradicted | New Jersey P.L.2026, c.25 (approved and effective 30 June 2026) added: a ban on sale of sensitive data including precise geolocation (1,750 ft) regardless of consumer count, and on sale or licensing by data brokers and collectors. State sale bans: four -> five in the abstract, chapters 7 and 12, the rules table, the figure and the plain edition. New claim LOC-R3-F01, source LOC-NFA-S02. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 7 standfirst, inshort and 'Four states now ban the sale'; chapter 0 inshort; figure-state-laws / data/figure-data.json states; data/rules.json; plain/index.html paragraph beginning 'Four states now ban' |
| R3-07 | data/profiles.json InMarket differentiation[2], capabilities.causal_design and capabilities.method_disclosure; source/evidence/swot/inmarket.json strengths[1]; data/source-register.json LOC-C-S02 (alias LOC-INMARKET-S11); rendered InMarket profile and SWOT | The accepted public-access qualification for the LCI glossary has not reached the deep profile and SWOT: they still credit publicly published 10/25-million-impression minimums, standard error and a confidence rule. The source register still labels the glossary full-access, directly conflicting with the paper's current unverified, sign-in-gated treatment. | The InMarket differentiation says LCI is 'one of two that publish numeric minimum samples and a confidence rule (10 MM / 25 MM impressions; >95%)'; its method_disclosure counts projection and uncertainty as disclosed from S11, and the SWOT calls the glossary a strength. Chapter 5/claim LOC-C-F24 and closed correction R2B-27 explicitly mark the InMarket thresholds unverified from public sources behind sign-in. LOC-C-S02 retains access_limits/access_note='full' and aliases S11. | supported with qualification | InMarket profile, SWOT and the source record LOC-C-S02: the LCI glossary thresholds are described as read once and not verifiable from a public route (sign-in); access set to partial. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/profiles.json InMarket differentiation[2], capabilities.causal_design and capabilities.method_disclosure; source/evidence/swot/inmarket.json strengths[1]; data/source-register.json LOC-C-S02 (alias LOC-INMARKET-S11); rendered InMarket profile and SWOT |
| R3-08 | data/profiles.json Placer.ai headline; source/data/profile-overlay.json placer.headline; source/evidence/swot/placer.json weaknesses[0]; rendered Placer profile/SWOT | The Placer headline and SWOT say every visit figure it shows is a projection, although the same profile and SWOT explicitly document raw panel visits displayed alongside projected totals. This turns a documented metric distinction into an absolute product claim. | Placer headline: 'every visit it shows is a projection'; SWOT weakness: 'Every visit figure is a projection'. The profile's differentiation[0] says 'Exposes raw panel visits beside projected visits in its API and dashboard', and SWOT strengths[1] repeats the same fact with LOC-PLACER-S01/S08. | supported with qualification | Placer.ai headline and SWOT: population visit estimates are projections from raw panel visits it also shows. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/profiles.json Placer.ai headline; source/data/profile-overlay.json placer.headline; source/evidence/swot/placer.json weaknesses[0]; rendered Placer profile/SWOT |
| R3-09 | data/profiles.json GroundTruth headline; source/data/profile-overlay.json groundtruth.headline; source/evidence/swot/groundtruth.json weaknesses[1]; rendered GroundTruth profile/SWOT | GroundTruth's headline expands the no-control/no-lift finding for Observed Visits and Projected Visits into a claim about all its visit measurement. The profile itself records a separate DOOH assertion that more than half of 42,000 visits were incremental, whose control design is unknown rather than established absent. | Headline: 'Visit counts with no control and no lift claim'; SWOT: 'Visit measurement is exposed-only with no control'. The same profile's performance entry for Place Exchange says 'over 42,000 in-store visits from exposed consumers', 'more than half' incremental, with design='not_documented'; chapter 9 also reports the unsupported incremental assertion. R2B-11 limits the no-lift qualification to the two visit-count help pages. | supported with qualification | GroundTruth headline and SWOT name Observed and Projected Visits; the separate DOOH incrementality assertion is described as documenting no control design. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/profiles.json GroundTruth headline; source/data/profile-overlay.json groundtruth.headline; source/evidence/swot/groundtruth.json weaknesses[1]; rendered GroundTruth profile/SWOT |
| R3-10 | Chapter 6 'Sensitive places' InMarket-order paragraph; data/profiles.json Vistar risks[1]/timeline 2024-05-01; source/evidence/swot/vistar.json threats[2]; rendered editions | Several surfaces say InMarket is categorically barred from sharing any precise location data, while chapter 7 correctly states the narrower sale-or-licensing ban. The order's broader use/transfer/share prohibition attaches to products categorising or targeting consumers by sensitive locations, so the summary overstates legal scope. | Chapter 6 says the order 'prohibits it from selling, sharing or licensing any precise location data'; Vistar's risk/SWOT repeats a blanket sharing ban. Opened order https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf , p.5 provisions II–III: II prohibits sale or licensing of Location Data for valuable consideration; III separately prohibits use/sale/license/transfer/share of sensitive-location categorisation/targeting products. Chapter 7's table uses this narrower distinction. | supported with qualification | Chapter 6 and the Vistar profile and SWOT now match the InMarket order: II bans sale or licensing of location data; III separately bans sensitive-location categorisation or targeting products. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 6 'Sensitive places' InMarket-order paragraph; data/profiles.json Vistar risks[1]/timeline 2024-05-01; source/evidence/swot/vistar.json threats[2]; rendered editions |
| R3-11 | Chapter 13, evidence-class rule; Chapters 3–5; Chapter 4 fig-ground-truth and study-evidence table; Appendix/claim-ledger LOC-B-F12/F13/F17/F24/F25 and LOC-PLACER-F09/F10; related source-register and study-evidence records | The published independence classifications do not follow the paper's own rule: studies whose source records explicitly say funding is unknown remain independent in the ledger, study table and ground-truth figure, while the Facebook and Yahoo research retains independent-measurement labels despite disclosed platform employment. This concerns whether independence was established, not whether those studies' results are wrong or whether unknown funding proves a conflict. | Chapter 13 says unknown-funded studies are not treated as independent and platform-employed coauthors are labelled as such. Representative package conflicts: LOC-B-S06 (Minnesota) funding unknown, yet LOC-B-F12 and LOC-PLACER-F09/F10 are independent_measurement and the Placer/StreetLight figure cells count as independent; LOC-B-S07 (Advan preprint) funding unknown, yet LOC-B-F13 and its study/figure row are independent; LOC-B-S24 funding unknown and LOC-B-F17 review_note expressly says independence is unknown, yet its class and Chapter 3 table/paragraph say independent. Newly opened primary https://arxiv.org/pdf/2201.07055, PDF p.44 Funding and Competing Interests, confirms Gordon and Zettelmeyer were part-time Facebook employees for data access and Moakler was a Meta employee and stockholder, while LOC-B-F24 remains independent_measurement. LOC-B-S25 records Yahoo employment during the | supported with qualification | Evidence class recoded from independent measurement to synthesis, with the reason, for claims resting on studies with unknown funding (LOC-B-F09, F12, F13, F14, F17, LOC-PLACER-F09, F10) and on platform-coauthored studies (LOC-B-F23 to F26); chapter 4 and 11 wording, the study table tags and the ground-truth figure updated. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 13, evidence-class rule; Chapters 3–5; Chapter 4 fig-ground-truth and study-evidence table; Appendix/claim-ledger LOC-B-F12/F13/F17/F24/F25 and LOC-PLACER-F09/F10; related source-register and study-evidence records |
| R3-12 | source/paper.config.json hero_summary and site.llms_txt; location-intelligence-2026.html hero; location-intelligence-2026.md opening italic subtitle; source/plain.md §What location data is final paragraph and §1 final sentence; plain/index.html #what and #guesses | The prominent claim that visit counts are mostly projections is stronger than the coded evidence: an undisclosed basis is being treated as if it were probably projected. The data establish that all four products which disclose a visit basis offer projections, not that projections are a majority of sold visit counts. | The hero says "visit counts are mostly projections"; plain says "Most visit counts you see are the second kind" and "A chart that says visits is most often a scaled-up guess." data/figure-data.json number_kinds records visit_products=13, visit_basis_disclosed=4, all four projected, and nine not_disclosed products; fig-number-kinds expressly warns "A number coded not disclosed may be projected; the vendor does not say." No volume-weighted or representative market denominator is supplied. | supported with qualification | Hero, llms.txt and plain edition: visit counts are projected or carry no stated basis; every firm that states the basis sells a projection. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/paper.config.json hero_summary and site.llms_txt; location-intelligence-2026.html hero; location-intelligence-2026.md opening italic subtitle; source/plain.md §What location data is final paragraph and §1 final sentence; plain/index.html #what and #guesses |
| R3-13 | fig-pipeline Verified box, bottom SVG note and caption; location-intelligence-2026.html and location-intelligence-2026.md; source/paper.config.json site.research_papers.summary | The verification figure mixes the 493 independently checked research claims with outcome totals for all 501 ledger entries, then asserts that every ledger claim was checked. Its caption also describes a residual unresolved sample even though the final ledger has none. | The Verified box shows "493 claims re-checked", followed by "321 as stated", "166 with qualification", "14 contradicted" (sum 501) and "0 not re-checked"; its note says "Every claim in the ledger was checked by a verifier or a review." In data/claim-ledger.json, checked_independently=true for 493 entries, whose outcomes are 313 supported_as_stated, 166 supported_with_qualification and 14 contradicted; the seven LOC-HYP-H* syntheses and LOC-GQ-F01 have checked_independently=false and account for the other eight supported entries. Chapter 13 correctly distinguishes them. The figure caption says "the rest are labelled unresolved" while data/stats.json claims_unresolved=0. The site-index summary also calls all 501 entries verified claims rather than distinguishing the 493 checked research claims. | supported with qualification | Pipeline figure and listing summary distinguish 493 checked research claims (313 / 166 / 14) from 8 author syntheses; the unresolved-residual caption removed. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | fig-pipeline Verified box, bottom SVG note and caption; location-intelligence-2026.html and location-intelligence-2026.md; source/paper.config.json site.research_papers.summary |
| R3-14 | source/plain.md §3 No one has checked the counts against real store traffic, second paragraph; plain/index.html #check | The plain edition turns the Placer trail comparison’s difference into a measured overcount by calling its count "30.5% too high". The full paper explicitly says the comparison is not like for like, so the percentage cannot be presented as an error against an equivalent ground-truth population. | Plain: "One trail check found a firm's count was 30.5% too high." Chapter 4 Parks, trails and one Placer.ai check says Placer estimated 13,233 against 10,138 counter visits but "the comparison is not like for like: polygons differ, counters cover only the cross-country trails, and events are handled differently"; it also reports no statistically significant difference. This is a failure to carry the accepted R2-10 qualification into the plain edition, not a challenge to that disposition. | supported with qualification | Plain edition: 30.5% above the trail counters, with the different areas and activities stated. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/plain.md §3 No one has checked the counts against real store traffic, second paragraph; plain/index.html #check |
| R3-15 | source/plain.md §2 heading and first paragraph; plain/index.html #panels | The plain edition broadens a missing comparable historical US panel series into the claim that nobody discloses panel size and that no firm reports it on the same basis twice. Those are much wider absence claims than the full paper’s finding. | Plain heading: "No one shows how big the phone panels are"; paragraph: "no firm shows the size of its phone group the same way twice." Chapter 2 and LOC-D-F41.final_wording bound the result to sources opened and to one-definition US panel figures spanning pre-2021 and post-2023. fig-dooh-sources and data/figure-data.json dooh.rows[0] explicitly record Geopath's 7–10 million qualified devices weekly from over 100 million US devices monthly (documentation-dated 2023). Accepted R2-24 requires the comparable-series and screened-source limits. | supported with qualification | Plain edition: no comparable history of US panel size before 2021 and after 2023 in the sources checked. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/plain.md §2 heading and first paragraph; plain/index.html #panels |
| R3-16 | source/plain.md §9 heading; plain/index.html #agents; source/chapters/00-abstract.md What it found item 10 and corresponding web/Markdown headline | The headline "can't yet buy by it" presents nonexistence of working geographic buying as established, although H7 is unresolved and the figure distinguishes unverified production claims from the absence of a public test. An implemented connector and unverified claims do not establish that buying cannot occur. | Plain heading: "AI buyers can describe a place but can't yet buy by it"; full abstract: "Agents can describe a place, not yet buy one." fig-protocol-states instead says "No location signal has a public interoperability test" and "H7 is unresolved"; data/figure-data.json protocols rows administrative geo and venue identifiers contain production-use code "vc" (vendor claim), while administrative geo has connector="conn". Chapter 10 explains that the campaign releases omit the protocol version and carried field, so the supported production branch is not established. | supported with qualification | Abstract, chapter 12 and plain edition: production buying by place is not established in the sources screened; H7 stays unresolved. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/plain.md §9 heading; plain/index.html #agents; source/chapters/00-abstract.md What it found item 10 and corresponding web/Markdown headline |
| R3-17 | data/source-register.json and source/data/sources-extra.json, LOC-SCD-S11/LOC-SCD-S12; matrix privacy cells for gimbal and bluedot | The source-access records contradict the scored privacy evidence: both policies remain recorded as unreadable while the matrix quotes detailed policy text and calls the scores verifier-qualified. At least Rezolve's current policy supports the described score, making this a stale provenance record rather than proof of a false capability claim. | LOC-SCD-S11 says its policy text could not be retrieved and LOC-SCD-S12 says 'No content read', both with access_limits 'access error'; the two matrix cells describe detailed policies and say 'verifier-scored from n/e'. On opening https://rezolve.com/privacy-policy/ for this review, sections 1.2 and 2.2 contain the cited device-location, consent and purpose content and the controller list includes Bluedot. Opening https://infillion.com/privacy-policy/ returned a shell, so the package does not currently supply a reproducible route for that detailed read. | supported with qualification | Source records LOC-SCD-S11 and S12 updated: both policies read in full in a rendered browser on 24 October 2026. The reviewer was wrong that Infillion's policy cannot be read; the cell keeps its score. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/source-register.json and source/data/sources-extra.json, LOC-SCD-S11/LOC-SCD-S12; matrix privacy cells for gimbal and bluedot |
| R3-18 | data/claim-ledger.json and source/data/claim-ledger.json LOC-J-F12 final_wording/review_note; data/agentic-map.json and CSV/rendered agentic register | The final claim ledger still says no Placer MCP server was found and represents Radar only by its documentation-search endpoint, despite the corrected chapter and deep profiles documenting both data interfaces. The agentic map likewise retains only the Radar documentation row and has no Radar data-pilot or Placer MCP row. | LOC-J-F12.final_wording names Placer.ai among vendors with no MCP server and describes Radar's endpoint as documentation search only; its review_note says 'no server has been released' for Placer. Chapter 10 'Vendors' agent interfaces sit outside buying' documents Radar's authenticated enterprise pilot announced 14 October 2026 (LOC-RADAR-F16) and Pi for Placer MCP documented by 22 October (LOC-PLACER-F19); source-register LOC-RADAR-S27 and LOC-PLACER-S14 identify the data-server documentation. | supported with qualification | LOC-J-F12 final wording and the agentic register now include Radar's authenticated enterprise MCP pilot and Placer.ai's enterprise MCP server, both outside buying. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/claim-ledger.json and source/data/claim-ledger.json LOC-J-F12 final_wording/review_note; data/agentic-map.json and CSV/rendered agentic register |
| R3-19 | Chapter 0 'What it found' item 2 and inshort; chapter 10 standfirst, 'One connector', 'No public test passes', 'The baseline is further along', and H7 summary | Prominent summaries drop the searched-evidence boundary and turn non-discovery into market-wide absence or superiority: no vendor discloses supply, no location signal has a public interoperability test, and the baseline is further along than any agentic protocol. The supporting claims establish results only for screened sources/protocols and a 34-agent registry that is not all deployments. | Chapter 0 item 2 says 'No vendor discloses a US panel series'; its inshort says 'no location signal has a public interoperability test'. Chapter 10 says 'The highest state any location signal reaches in open-protocol agent-driven buying is an implemented connector' and 'Non-agentic buying is further along than any agentic protocol'. LOC-J-S13 explicitly says the registry lists registered agents, not all AdCP deployments; LOC-J-F13 is an inference across screened evidence, and LOC-G-F20 logs unscreened vendor blogs/private vendors. R2-60 and R2-61 already prescribe a bounded comparison; these are remaining propagation failures. | supported with qualification | Abstract and chapter 10 summaries bounded to the sources, protocols and registry screened on 24 October 2026; the baseline comparison bounded to the opened Google documentation. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 0 'What it found' item 2 and inshort; chapter 10 standfirst, 'One connector', 'No public test passes', 'The baseline is further along', and H7 summary |
| R3-20 | data/profiles.json Foursquare differentiation[0] and timeline merger entry; rendered Foursquare profile; data/deep-profile-records.csv corresponding rows | Two Foursquare profile dates use announcement/publication dates where the package's settled event dates differ: GroundTruth's accreditation is dated August rather than July 2021, and the Factual merger announcement is dated 7 rather than 6 April 2020. | Foursquare differentiation[0] says GroundTruth was accredited 'in August 2021', while GroundTruth's own timeline and LOC-A-S17 record the MRC grant in July (vendor announcement 3 August). Foursquare timeline dates the Factual merger announcement 2020-04-07, while chapter 2 uses 6 April and source-register LOC-D-S14 explicitly distinguishes pub_date=2020-04-07 from event_date=2020-04-06. | supported with qualification | Foursquare profile: GroundTruth accreditation granted July 2021 (announced August); Factual merger announced 6 April 2020 (release dated 7 April). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/profiles.json Foursquare differentiation[0] and timeline merger entry; rendered Foursquare profile; data/deep-profile-records.csv corresponding rows |
| R3-21 | data/agentic-map.json first AdCP 3.2.6 administrative-geography row, note; source/data/agentic-map.json; CSV export and rendered agentic register | The agentic map assigns 'AdCP 2.5.0' to PubMatic's December 2025 campaign while the chapter and campaign claim explicitly say no protocol version was recorded. An inferred contemporaneous version must not appear as the campaign's documented version. | The first map row note reads 'PubMatic (December 2025, AdCP 2.5.0) ... describe geo- and venue-targeted agentic campaigns without naming a field, version or delivery metric'. Chapter 10 and LOC-J-F08 say the vendor release records neither the signal nor the protocol version. | supported with qualification | Agentic register: PubMatic campaign marked as recording no protocol version (the inferred 2.5.0 removed). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/agentic-map.json first AdCP 3.2.6 administrative-geography row, note; source/data/agentic-map.json; CSV export and rendered agentic register |
| R3-22 | Chapter 4, "Third-party and production uses" paragraph; Appendix/claim-ledger LOC-PLACER-F26 (claim and study_design); source-register LOC-PLACER-S40 period | The six-chain Placer comparison is described as extending over 12 quarters, but its denominator is 12 retailer-quarter comparisons, with several retailers measured in the same quarter. This overstates the elapsed coverage of this low-confidence vendor analysis. | The package LOC-PLACER-F26 says "over 12 quarters" and dates the coverage from Q1 2025 to Target Q2 FY26. Newly opened author's primary analysis https://www.growthfactor.ai/resources/blog/retail-foot-traffic-data, table under the SEC-filings comparison: 12 rows comprise Target (2), Walmart (1), Home Depot (3), Lowe's (3), Chipotle (2), and Dollar General (1); repeated FY25 Q1/Q2/Q3 rows are separate chain observations. This confirms the denominator and unit; it does not independently verify the underlying SEC/Placer figures or elevate the blog to independent validation. | supported with qualification | Placer.ai SEC comparison: 12 retailer-quarter comparisons, not 12 quarters (chapter 4, LOC-PLACER-F26). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 4, "Third-party and production uses" paragraph; Appendix/claim-ledger LOC-PLACER-F26 (claim and study_design); source-register LOC-PLACER-S40 period |
| R3-23 | Appendix/claim-ledger LOC-B-F25 claim and final_wording (coefficient-of-variation clause) | The claim attaches the coefficient of variation of about 10 to per-capita advertising cost, giving the statistic the wrong denominator. The coefficient is the standard deviation of individual sales divided by mean individual sales. | LOC-B-F25 final_wording says individual sales showed a coefficient of variation of about 10 "relative to per-capita ad cost"; its own review_note says that cost comparison qualifies volatility, not the CV. Opened primary paper https://gwern.net/doc/economics/advertising/2015-lewis.pdf, printed p.1942/PDF p.2, explains that the standard deviation of individual sales is typically ten times their mean; printed p.1955/PDF p.15 identifies the ratio as standard deviation to mean and reports a retail median of 9.9. Confidence is high because the denominator is explicit in the paper. | contradicted | LOC-B-F25: individual sales have a standard deviation about ten times their mean (the coefficient of variation is not relative to ad cost). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Appendix/claim-ledger LOC-B-F25 claim and final_wording (coefficient-of-variation clause) |
| R3-24 | Abstract, 'What it found', item 5 (source/chapters/00-abstract.md:29); chapter 12, 'What the evidence says', first bullet (source/chapters/12-next.md:15); data/claim-ledger.json, LOC-HYP-H3.claim and final_wording; chapter 1 In short; fig-control-design takeaway/title in web and Markdown | The summary calls all nine coded products products that document a lift method, although one is GroundTruth's exposed-only visit reporting, explicitly documented as making no lift claim. The final H3 clarification and chapter 5 correctly distinguish the nine-product visitation diagnostic from the eight-product lift-claim count, but these summary passages collapse them again. | Abstract item 5: 'Seven of the nine cohort A products that document a lift method'; chapter 12: 'nine document a general lift method'. Chapter 5's table codes GroundTruth 'Exposed-only, no lift claim' and its next paragraph states: 'In the strict lift-claim count, which drops GroundTruth because it makes no lift claim, six of eight are matched or synthetic'. data/hypotheses.json H3.clarifications[0] explicitly requires a separate strict lift-claim count for a product whose documented metric makes no lift claim. This finding concerns the final summaries' wording, not the already-adjudicated nine-product coding or H3 verdict. The control figure title also calls the nine records default lift methods while its GroundTruth row explicitly says no lift claim. | supported with qualification | Lift summaries use six of eight products with documented lift methods; seven of nine is kept only for documented visit-measurement methods, with the exposed-only product named. H3 stays contradicted. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Abstract, 'What it found', item 5 (source/chapters/00-abstract.md:29); chapter 12, 'What the evidence says', first bullet (source/chapters/12-next.md:15); data/claim-ledger.json, LOC-HYP-H3.claim and final_wording; chapter 1 In short; fig-control-design takeaway/title in web and Markdown |
| R3-25 | Chapter 4, 'What study would settle it', opening sentence (source/chapters/04-validation.md:83) | The sentence 'H2 needs one design, run by a party with no stake in the panel' adds a necessary independence condition absent from the registered rule. H2 also accepts a vendor-run comparison with a held-out design and published method, as the chapter correctly explains earlier. | data/hypotheses.json H2.rule_supported accepts 'a comparison with a published method by a party without a financial interest, or a vendor comparison with a held-out design and published method'. Chapter 4's opening H2 explanation repeats both routes, while the 'What study would settle it' paragraph says a party with no stake is required and narrows the example to projected retail visits across hundreds of places. | supported with qualification | Chapter 4: the settling study is framed as one sufficient design; the registered vendor-held-out route is stated. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 4, 'What study would settle it', opening sentence (source/chapters/04-validation.md:83) |
| R3-26 | Chapter 6, 'Where the location comes from: H4', Broadsign row (source/chapters/06-audiences.md:41); data/claim-ledger.json LOC-F-F17.claim/final_wording/limitations; source/evidence/research/F.json h4_source_coding_cohort_b row for Broadsign | The final H4 table still says Broadsign Ads is not in the register, although the applied register now contains a distinct cohort-B Broadsign Ads audiences record. Its exclusion can still rest on unconfirmed US availability, but the published explanation and discovery claim describe an obsolete register state. | Chapter 6 labels it 'Broadsign Ads (found in discovery, not in the register)' and codes it 'SDK via a named partner; outside the count'. data/product-register.json contains product 'Broadsign Ads audiences', cohort 'targeting', with inclusion_reason 'Discovery (stream F, round 2 R2-109)' and us_available 'unconfirmed (no opened page establishes US availability, check G2)'. The corrections ledger R2-109 records the accepted addition to the register. No change to the H4 verdict is implied. | supported with qualification | Chapter 6: Broadsign Ads is registered, outside the H4 count because its US availability is unconfirmed. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 6, 'Where the location comes from: H4', Broadsign row (source/chapters/06-audiences.md:41); data/claim-ledger.json LOC-F-F17.claim/final_wording/limitations; source/evidence/research/F.json h4_source_coding_cohort_b row for Broadsign |
| R3-27 | source/chapters/00-abstract.md What it found item 1; source/chapters/01-defining.md In short; source/chapters/05-measurement.md In short; source/plain.md §1; corresponding web and Markdown paragraphs | The summaries still report eight products with undisclosed visit bases, while the corrected figure and data count nine. Arrivalist arrivals remain an additional undisclosed visit number after its card-based spend was separated. | Abstract: "eight state no basis at all"; chapter 1: "For eight of the fourteen ... the basis ... is not disclosed"; chapter 5: "for eight products the basis is not disclosed"; plain: "Eight do not say what kind of count it is." data/figure-data.json number_kinds.names.not_disclosed lists Blis, Reveal Mobile, Adsquare, Cuebiq, Vistar, Kochava, Arrivalist, Lifesight, Precisely PlaceIQ (nine). LOC-E-F29.final_wording separately says Arrivalist arrival basis is not stated, then lists the other eight; fig-number-kinds correctly reports four disclosed of 13 visit sellers. | supported with qualification | Nine undisclosed visit bases (Arrivalist's arrival counts included), not eight, in the abstract, chapters 1 and 5 and the plain edition. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/chapters/00-abstract.md What it found item 1; source/chapters/01-defining.md In short; source/chapters/05-measurement.md In short; source/plain.md §1; corresponding web and Markdown paragraphs |
| R3-28 | fig-quotient caption caveat in location-intelligence-2026.html and location-intelligence-2026.md; fig-pipeline SVG Scored box and explanatory note | The figures retain a six-row blind second scoring pass, although the final method and published second-pass data contain sixteen rows. | fig-quotient caveat: "six rows nearest a band line re-scored blind"; fig-pipeline: "blind 2nd pass on 6 rows" and "a second AI scorer on six rows". source/chapters/13-method.md How the products were scored says "sixteen rows and 166 cells"; data/figure-data.json second_pass.rows=16, cells=166, exact_pct=81.9, within_one_pct=96.9; data/second-pass.json contains 16 row IDs. | contradicted | Figures now say sixteen rows for the blind second pass. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | fig-quotient caption caveat in location-intelligence-2026.html and location-intelligence-2026.md; fig-pipeline SVG Scored box and explanatory note |
| R3-29 | fig-pipeline SVG Scored box in location-intelligence-2026.html (and figures/fig-pipeline.svg referenced by Markdown) | The pipeline calls the scoring population 48 vendors, but the unit is a named product and several companies appear in multiple cohorts. | The figure prints "48 vendors". data/capability-matrix.json has 48 product rows but 32 distinct vendor labels; chapter 13 says "The scoring unit is a named product" and plain #scores correctly says "48 products". | contradicted | Pipeline figure: 48 products, not vendors. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | fig-pipeline SVG Scored box in location-intelligence-2026.html (and figures/fig-pipeline.svg referenced by Markdown) |
| R3-30 | source/plain.md How the firms scored; plain/index.html #scores | The score explanation omits the author’s privacy risk adjustment, a material qualification to the claim that a higher score simply means a firm shows more of its work. That adjustment changes two published bands and is explained prominently in the full scoring chapter. | Plain says "Each got 0 to 5, based only on what the firm shows in public" and "A high score means a firm shows more of its work." Chapter 11 says privacy is capped at 1 for qualifying events in the three years before the cutoff, separate from published current-control scores. LOC-GQ-F01.final_wording records eight capped products, 3–5 GQ points of impact, and two band changes without the cap (InMarket measurement and Kochava measurement). | supported with qualification | Plain edition explains the privacy risk adjustment and that both scores are shown. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/plain.md How the firms scored; plain/index.html #scores |
| R3-31 | location-intelligence-2026.md chapter 1 table Rule / Where precise ends / Evidence; more generally prose claim references throughout the Markdown edition | The Markdown edition drops the claim references that connect the web prose and tables to evidence; the chapter 1 table has an entirely empty Evidence column. Although a claim ledger and reference list remain later in the file, the reader loses the direct mapping for the actual paragraph or table row. | source/chapters/01-defining.md has claim markers in every Evidence cell, for example Maryland/Oregon/Washington/Connecticut/Virginia cites LOC-H-F24 and LOC-H-F10/F11/F12/F13. The generated location-intelligence-2026.md table prints an empty final cell for all nine rows, while the web edition retains claim chips. The same stripping removes the LOC-* claim markers attached to ordinary prose. | supported with qualification | Markdown edition: claim markers rendered as claim ids instead of being dropped. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | location-intelligence-2026.md chapter 1 table Rule / Where precise ends / Evidence; more generally prose claim references throughout the Markdown edition |
| R3-32 | fig-mrc-history caption caveat in location-intelligence-2026.html and location-intelligence-2026.md | The caption incorrectly says the 2018 MRC presentation has no source record even though that record was added and the figure itself cites it. | Caveat: "the 2018 presentation, which has no source record in the evidence files (corrections ledger R2-114)." data/source-register.json contains LOC-NFA-S01, title "MRC presentation to the ACA Executive Forums (26-27 September 2018)", full access, slide 9 passage "Location – 6 Applicants; 1 Accredited", origin data/sources-extra.json; data/figure-data.json mrc.rows[0].src is ["LOC-NFA-S01"], and the figure source list begins with that record. | contradicted | MRC history caption: the 2018 presentation now has a source record (LOC-NFA-S01); the clause saying it has none removed. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | fig-mrc-history caption caveat in location-intelligence-2026.html and location-intelligence-2026.md |
| R3-33 | source/chapters/13-method.md In short, first two sentences; corresponding chapter 13 web and Markdown summary | The method summary counts the enforcement stream twice by adding it to a workstream total which already includes it. | The summary reports "11 research workstreams, 1 vendor-universe sweep ... plus one stream of enforcement records added after the second outside review." data/stats.json has streams=12, workstreams=11 and universe_sweeps=1, and source/evidence/research has exactly A–J, U and X: ten original thematic streams, universe U, and enforcement X. Thus eleven workstreams plus U already includes X; the current wording implies thirteen source streams. | supported with qualification | Method summary: eleven workstreams = ten from the brief plus the enforcement stream; no double count. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | source/chapters/13-method.md In short, first two sentences; corresponding chapter 13 web and Markdown summary |
| R3-34 | Chapter 1, The definitions, cohort bullet; source/chapters/01-defining.md; corresponding web/Markdown prose | The opening definitions still describe four cohorts and treat C as one broad data/intelligence group, while the final scoring rules and matrix use five cohorts with separate C1 and C2 tasks. The developer definition also retains the broader location-API wording after places/maps APIs were moved to the listed table. | Chapter 1 says 'A is visit measurement ... C is location data and intelligence platforms. D is location APIs and SDKs'. data/scoring-rules.json/cohorts and chapter 13 instead define A, B, C1 foot-traffic/visit analytics, C2 raw movement data, and D geofencing/visit-detection SDKs/APIs. | contradicted | Chapter 1 cohort definitions now list the five final cohorts and the narrowed developer task. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 1, The definitions, cohort bullet; source/chapters/01-defining.md; corresponding web/Markdown prose |
| R3-35 | Chapter 13, Interests, Verve contribution sentence; source/chapters/13-method.md | The disclosure attributes the 2017 playbook credit to 'Verve Group', whereas the historical source credits 'Verve'. Naming the contributor as printed avoids implying that the current group name was the credited entity in 2017. | Chapter 13 says 'Verve Group is a listed contributor to IAB’s 2017 location-marketing playbook'. The primary PDF opened for this review, https://www.iab.com/wp-content/uploads/2017/04/IAB_Location-Playbook-for-Retail-Marketers_FINAL.pdf , PDF page 2, lists 'Verve' among the contributing member companies. | supported with qualification | Chapter 13: Verve (as printed) is credited as a contributor to the 2017 IAB playbook. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Chapter 13, Interests, Verve contribution sentence; source/chapters/13-method.md |
| R3-36 | Abstract How it was done; chapter 13 How the evidence was gathered; hero source-count label; data/source-register.json access-status fields | The headline count is a count of registered source URLs, not a count of sources whose contents were all opened. The method also calls all 69 non-full records partly readable, although that set explicitly includes failed reads and a search snippet from an unopened page. | data/source-register.json has 789 unique URL records: 720 marked full, 50 partial extract, 12 abstract only, four access error with text/metadata via another route, and three access error. LOC-V-VISTAR-S05 expressly says 'search-result summary only (page not opened)'. The abstract says 789 sources 'were opened and logged', and chapter 13 says 69 were 'only partly readable'. This finding does not assert that a claim was supported solely by the unused snippet record. | supported with qualification | Abstract, method chapter and hero label: source records registered, with the access-limited records described as such. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | Abstract How it was done; chapter 13 How the evidence was gathered; hero source-count label; data/source-register.json access-status fields |
| R3-37 | data/product-register.json and its matrix register copies / universe exports, us_available fields | Product-specific US evidence reached during scoring has not been propagated back to several register records. Some register notes remain unconfirmed even where the matrix says the new evidence closes that status; other rows need a narrower update rather than a blanket available label. | The Azira Measurement register says “unconfirmed product-level”, but its US-availability cell awards 4 and says the US case and records “settle it”. Mobilewalla’s register remains “unconfirmed”, while its cell says the opened records “close that”. Adsquare measurement’s old note says no opened page shows US attribution; its matrix now documents a named Visit Vermont US campaign but explicitly keeps general US availability unstated. The latter supports recording US production evidence, not a general-availability conclusion. | supported with qualification | Register US-availability fields updated for Azira Measurement and Mobilewalla audiences; Adsquare records named US production use without general availability. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything. | data/product-register.json and its matrix register copies / universe exports, us_available fields |
| FINAL-01 | Live web edition: inline claim diamonds and numbered citation buttons; screen CSS .claim and .cite | The inline evidence controls have no vertical click area. Visible claim diamonds and superscript citations paint on screen, but pointer clicks do not activate their evidence panels. This undermines the paper's primary traceability interaction, even though matrix-cell panels work. | Direct browser inspection on 24 October found all 1,538 button.claim elements had getBoundingClientRect().height = 0; sampled visible numbered citation buttons also had zero height. The published rules set line-height:0, border:0 and zero vertical padding on both .claim and .cite. A normal semantic click on a visible claim control timed out; clicking the painted diamond left #evpanel data-open='false', and elementFromPoint returned the surrounding paragraph. A normal click on the Placer ground-truth matrix cell opened the evidence panel successfully, separating the hit-area defect from the panel logic. Reproduced in the desktop browser at 1363 by 936; touch hardware was not tested. | contradicted | Inline claim and citation buttons now have a real hit area (inline-block, line-height 1, padding), raised by offset so the line box is unchanged; verified by pointer click in the built page. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Live web edition: inline claim diamonds and numbered citation buttons; screen CSS .claim and .cite |
| FINAL-02 | Live web edition: DOM citation controls with data-src aliases; source lookup initialization and showSource() | A separate source-lookup defect silently breaks 155 citation actions. The canonical source objects contain valid aliases, but the browser lookup indexes only canonical IDs. Controls carrying an alias therefore return without opening a source. The sources themselves are present. | Independent parsing of the captured live HTML found 2,287 data-src reference tokens. Of these, 155 tokens, representing 100 distinct IDs, are absent from the canonical-ID lookup; every one maps to a declared alias of an existing source. Example: the figure's [3] button carries data-src='LOC-C-S04', while canonical source LOC-B-S33 has aliases LOC-C-S04 and LOC-PLACER-S01 and number 3. The script initializes SRC[s.id]=s only, then showSource(id) returns immediately when SRC[id] is absent. Matrix scores and claim-ledger source_ids are canonical and have no unresolved references; this defect is confined to the affected DOM citation controls. It will persist after the click-target CSS is fixed. | contradicted | Citation ids are canonicalised at render (figures and instruments) and the page lookup also indexes source aliases; a new build check fails if any data-src id is not a canonical source id (155 unresolved tokens before, 0 after). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Live web edition: DOM citation controls with data-src aliases; source lookup initialization and showSource() |
| FINAL-03 | Chapter 11 matrix, Placer.ai / Ground-truth validation; data-backed Most documented claims in the abstract, chapter 11 and plain edition | The accepted funding correction has not reached Placer.ai's validation score. The score remains 4 because the Minnesota case is called independent, although the live study, source and claim records explicitly say its funding is unknown and independence is not established. That leaves the paper's only Most documented classification unsupported by its frozen rule. | Live matrix row placer, ground_truth: v=4, grade B, 'one independent comparison ... University of Minnesota Extension ... Grade rests on S24 alone'; the same rationale says the test is not held out. Live studies.json, Choi 2024 sponsor: 'no funder or data arrangement stated ... independence not established, round 3 R3-11'; source LOC-B-S06 (alias LOC-PLACER-S24) funding='unknown'; LOC-B-F12 / LOC-PLACER-F09 have already been recoded from independent measurement. The dimension says unknown funding is not independent, and anchor 4 requires an independent comparison or a vendor held-out design. Independent recomputation: replacing validation 4 with 3 as an illustrative conservative bound changes Placer GQ 56.2357 to 54.0071 and the C1 field 43.075 to 42.775; Most begins at 54.775, so Placer becomes Well documented. Removing the unsupported validation cell also gives Well documented (53.6615 | contradicted | Placer.ai ground-truth validation 4 -> 3: the Minnesota deck states no funder, sponsor or data arrangement, so it is a third-party comparison with a published method and no held-out design (anchor 3). With every final-sweep correction applied together, the C1 field mean falls (four other C1 rows lost points) and Placer.ai stays Most documented, by 0.23 points of offset; it is Well documented under three of the four reweightings, which the abstract and chapter 11 now say. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 matrix, Placer.ai / Ground-truth validation; data-backed Most documented claims in the abstract, chapter 11 and plain edition |
| FINAL-04 | Chapter 11 matrix, method_disclosure cells for Mobilewalla audiences, Foursquare Visits feeds and Zartico; same fixed compound-element rule as accepted R3-05 | Half-published compound elements are still being counted as full elements outside the three rows corrected in round 3. The frozen seven-element count requires panel size AND consent route, and cadence AND uncertainty. At least three further recorded scores exceed the count stated by their own rationales. | Mobilewalla v=2 counts (1) sources plus only 'part of (3) panel size (2.2 billion global devices, no consent route)' and says all other elements absent: one complete element is anchor 1, not 2. Foursquare Visits v=3 counts (1) sources, (2) visit definition, (4) normalisation, plus (3) company consent 'but no panel size with a period' and (7) cadence with 'no uncertainty': only the first three are complete, giving anchor 2. Zartico v=2 counts (3) a global device figure 'no consent route' and (7) cadence, with normalisation only asserted: neither compound element is complete, so this remains a marketing-only anchor 1. These are the same AND conditions already applied to Placer, Quadrant and Veraset. Mobilewalla's 2-to-1 correction alone changes its independently recomputed GQ 37.0532 to 35.5865 and Partly to Least documented; Foursquare Visits 3-to-2 alone gives 50.5917 (still Well). This | contradicted | Complete-element rule applied to all 48 method-disclosure cells: Foursquare Attribution 3 -> 2, Foursquare Visits feeds 3 -> 2, Mobilewalla audiences 2 -> 1, Zartico 2 -> 1, and two further cells found by the same audit, Huq Industries 2 -> 1 and Precisely Visits 2 -> 1. Nine other rationales (Azira Measurement, Cuebiq Audiences, Advan, Mapbox, Outlogic, InMarket, Placer, Quadrant, Veraset) rewritten to the correct count with unchanged scores. Mobilewalla moves from Partly to Least documented. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 matrix, method_disclosure cells for Mobilewalla audiences, Foursquare Visits feeds and Zartico; same fixed compound-element rule as accepted R3-05 |
| FINAL-05 | Chapter 11 matrix, Broadsign Ads audiences row; product register admission condition; corrections ledger R2-109 | Broadsign's Native Touch audience candidate is scored and banded even though its explicit US admission condition remains unmet. The live register, matrix note and already-accepted corrections-ledger ruling all say this product must not be scored before US eligibility is established. | The row note says 'US availability is not established, so the register’s admission condition is unmet'; register.inclusion_reason says 'Admitted only if US availability is established before scoring'. Its US-availability cell is n/e. The final wording of accepted R2-109 says 'US availability, US panel size and period remain unresolved, so the product is not scored until US eligibility and the numeric-cell minimum are established.' Despite this, the row contributes six scores, GQ 39.8244, a Partly documented band and a place in the 48-product count. Its sources are https://broadsign.com/audiences/ and https://broadsign.com/outmoove/ ; no new US eligibility evidence is recorded. Removing this row alone changes the cohort-B field from 44.0611 to 44.5655; no other band changes in that one-row calculation. | contradicted | Broadsign Ads audiences moved to the listed table: its admission condition (US availability established before scoring) is unmet. 47 products are scored; cohort B has 14 rows; chapter 13 records the removal. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 matrix, Broadsign Ads audiences row; product register admission condition; corrections ledger R2-109 |
| FINAL-06 | Chapter 12 standfirst; chapter 12 decision tool, first row / embedded decision[0].test; PDF pp. 115–116 | The buyer-facing conclusion still turns an unknown visit basis into a majority projection claim. The revised abstract and hero correctly distinguish disclosed projections from an undisclosed basis, but chapter 12 says "Visit counts are mostly projections" and tells buyers a dashboard "visits" number is "usually showing a projection". Neither frequency claim follows from the coded sample. The same standfirst says "nobody independently checks either", broadening the bounded absence of multi-site projected-retail validation to all independent checks of counts or lift. | Live chapter 12 standfirst: "Visit counts are mostly projections". decision[0].test: "A dashboard that labels a number "visits" is usually showing a projection." Chapter 1 fig-number-kinds and LOC-E-F29 instead establish that four of thirteen products selling a visit number document a projected number, three also report observed numbers, and nine visit numbers state no basis. The figure expressly warns that a number coded not disclosed may be projected but the vendor does not say. The corrected live abstract already uses "projections or unlabelled". This is remaining propagation of R3-12, not new evidence about product performance. Chapter 12 immediately qualifies its own "unchecked" bullet to no independent multi-site projected-retail comparison found; chapter 4 retains historical SafeGraph park and polling-place comparisons. Thus the absolute standfirst also outruns its qualified suppo | contradicted | Chapter 12 standfirst: visit counts are documented projections or have no stated basis, and no independent multi-site validation of projected retail visits was found in the sources screened. Decision-tool row 1: a dashboard label alone does not establish whether visits are observed or projected. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 12 standfirst; chapter 12 decision tool, first row / embedded decision[0].test; PDF pp. 115–116 |
| FINAL-07 | Chapter 1 In short; Chapter 11 Placer.ai profile Limits; live claim ledger LOC-PLACER-F01 final wording | The correction distinguishing projected Placer population totals from raw panel counts is still absent from these passages. They say every Placer visit figure is projected, even though the same live paper explicitly documents raw, unextrapolated panelVisits and panelVisitors. | Chapter 1: “Placer.ai describes every visit count it sells as an extrapolation from a device panel.” Placer Limits: “Every Placer visit, visitor and retail-visit figure is a projection ... not an observed count.” LOC-PLACER-F01 repeats that absolute in its final wording. In the same current ledger, LOC-PLACER-F06 says the API separates raw panel counts from projections and defines panelVisits/panelVisitors as raw, unextrapolated data; Chapter 1 body and the corrected profile headline also preserve that distinction. This is a remaining implementation miss of accepted R3-08, not a new challenge to how population estimates are coded. Published source route: https://docs.placer.ai/reference/glossary.md (not independently accessible in this sweep); the contradiction is directly present within the live paper. | contradicted | Chapter 1, chapter 11, the Placer.ai profile limits, the plain edition and LOC-PLACER-F01 now say that Placer's population estimates are projections and that it also exposes raw panel counts. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 1 In short; Chapter 11 Placer.ai profile Limits; live claim ledger LOC-PLACER-F01 final wording |
| FINAL-08 | Chapter 11 matrix, Arrivalist MediaInsights / Documented production volume (scale) | Arrivalist's production-volume rationale does not reconcile its undated 'Annual Campaigns' claim with the rule requiring a stated 12-month observation period. 'Annual' establishes a duration, but the page supplies no dated observation interval; the current score needs an explicit interpretation of that requirement. | The live cell assigns scale=3: 'Anchor 4 met in substance, capped at 3 by grade D ... Annual Campaigns 1000+ ... The 12-month period is not dated.' The published volume definition requires a stated observation period and geography, and cohort A may use measured campaigns 'in a stated 12-month period'; anchor 1 explicitly covers a figure with no period or definition. The primary vendor page https://www.arrivalist.com/partners was reopened in this sweep: its metrics read Annual Impressions 12B+, Annual Campaigns 1000+, Media Partners 20+, without a measurement year or dated 12-month interval. A current copyright or access date does not date those observations. Changing scale 3 to 1 alone recomputes Arrivalist GQ 42.0794 to 40.2341 (still Partly), and also moves Adsquare's relative band to Well because the measurement field mean falls. | contradicted | Arrivalist documented production volume 3 -> 1: "Annual Campaigns 1000+" carries no year, dated period or geography (page re-opened), scored as Placer.ai's undated figure was under R3-04. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 matrix, Arrivalist MediaInsights / Documented production volume (scale) |
| FINAL-09 | Chapter 11 matrix privacy dimension and Foursquare Attribution / Movement SDK rationales; published scoring-rules download | The privacy window still uses the former 30 October cutoff even though the live paper states a 24 October 2026 research cutoff. The Foursquare rationales also use the stale window to call the 26 October 2023 dismissal outside it. | The live privacy definition says 'within the three years before the cutoff (30 October 2023 to 30 October 2026)'. Foursquare Attribution says no event falls inside '2023-10-30 to 2026-10-30', and Movement SDK says the 26 October 2023 dismissal precedes the 30 October opening. Three years before 24 October 2026 begins on 24 October 2023, making 26 October 2023 inside that window. The dismissal remains non-qualifying because it is neither an order, judgment nor finding; none of the seven listed cap events is near the boundary, so no current cap necessarily changes. | supported with qualification | Privacy window moved with the cutoff to 24 October 2023 - 24 October 2026 in the scoring rules and every rationale; the 26 October 2023 Proskin dismissal now falls inside the window and is explained as non-qualifying by event type. No cap changes; chapter 13 discloses the window. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 matrix privacy dimension and Foursquare Attribution / Movement SDK rationales; published scoring-rules download |
| FINAL-10 | Chapter 11 matrix rationale text after round-3 patches: Placer, Quadrant and Veraset method_disclosure; InMarket measurement privacy | Several corrected cells append the new ruling while retaining the old, contradictory explanation, leaving a clicked score with two incompatible counts or current-control values. | Placer disclosure now starts 'Anchor 2 ... leaving three of seven' but later still says 'Anchor 3 ... Four of seven'; Quadrant and Veraset likewise start with three complete elements at anchor 2 and then retain 'Anchor 3: four ...'. InMarket privacy opens with current_controls 2, later says 'Current controls 3', and finally says it was re-anchored to 2. The scored values themselves match the accepted corrections; this is a reader-facing traceability defect. | supported with qualification | Placer, Quadrant and Veraset disclosure rationales and InMarket's privacy rationale rewritten as one final explanation each; superseded reasoning stays in this ledger. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 matrix rationale text after round-3 patches: Placer, Quadrant and Veraset method_disclosure; InMarket measurement privacy |
| FINAL-11 | Chapter 13 blind second-pass explanation and published data/second-pass.json / second-pass.csv tables | The blind-pass files retain the pre-correction first-score snapshot but do not identify that version, so their 'first' scores and quotients no longer match the published matrix. The reliability percentages still reproduce for that original snapshot; they should be labeled as historical rather than silently read as agreement with the current corrected scores. | In second-pass.json Kochava privacy.first = 1 while the matrix is now 2, and Veraset method_disclosure.first = 3 while the matrix is now 2. The bands table has Kochava first 34/Least versus current 35/Partly, and Veraset first 42 versus current 40. The original 166-cell comparison independently reproduces 81.9277% exact, 96.8504% within one among 127 jointly numeric cells, and quadratic kappa 0.89947. Substituting raw second scores into the current matrix still moves only Simpli.fi's band, but updated cohort means yield second quotients 43 for Arrivalist and 42 for Arity versus 42/41 in the retained table. These are distinct versions of the experiment, not evidence that the original agreement statistic was arithmetically wrong. | supported with qualification | Chapter 13 and data/second-pass.json label the blind pass's first scores as the d5e1dc2e snapshot, before round 3 and the final sweep; the historical blind data are not overwritten. Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: The † now shows the historical experiment only (Simpli.fi), read from data/second-pass.json, instead of substituting old second scores into the corrected matrix (which had flagged four rows); chapter 13 says the published scores carry later corrections. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 13 blind second-pass explanation and published data/second-pass.json / second-pass.csv tables |
| FINAL-12 | Abstract, How it was done; chapter 13 In short / How it was checked; fig-pipeline bottom SVG note; embedded claims entry LOC-R3-F01; PDF p.120 | The claim-checking disclosure no longer describes its own ledger. The ninth unchecked entry is the factual New Jersey statute finding, not a hypothesis or privacy-cap synthesis. The pipeline also retains the absolute statement that every ledger claim was checked despite showing nine author syntheses. | The live embedded claims array has 502 entries: checked_independently=true for 493 (310 supported_as_stated, 169 supported_with_qualification, 14 contradicted) and false for nine. The false set is LOC-HYP-H1 through LOC-HYP-H7, LOC-GQ-F01 and LOC-R3-F01. LOC-R3-F01 states the New Jersey precise-geolocation sale ban and its effective date; its classification is synthesis, but it is a new external legal fact rather than a reading or cap-effect calculation. Abstract and method say the unchecked remainder consists of syntheses "that state the hypothesis readings and the privacy-cap effect" and that every research claim was checked. The bottom fig-pipeline SVG note still says "Every claim in the ledger was checked by a verifier or a review", contradicting the corrected 493 research claims / nine author syntheses boxes. The bottom-note absolute also appears on PDF p.120. | supported with qualification | Abstract and chapter 13 describe the nine unchecked ledger entries as eight author syntheses plus the review-added New Jersey legal finding, read in the statute by a separate check; the pipeline note now limits "checked" to research claims marked as checked. Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: The pipeline box now reads "8 syntheses + 1 legal finding" and its caption matches the abstract and method. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Abstract, How it was done; chapter 13 In short / How it was checked; fig-pipeline bottom SVG note; embedded claims entry LOC-R3-F01; PDF p.120 |
| FINAL-13 | Chapter 13 In short; References introduction in web and Markdown; PDF pp.118,132 | Two remaining process summaries still call all 790 registered sources opened, although the corrected method explicitly retains at least one unopened search-result lead and other access-limited records. | Chapter 13 In short says "The research opened and logged 790 sources". References says "The full register of 790 sources opened during the research". The same live method correctly says 68 records are access-limited, including "not opened at all (one is a search-result lead kept for the trail)". The embedded source register contains 722 full, 51 partial extract, 12 abstract only, four access error/text or metadata through another route, and one access error record. The hero already labels 790 "source records registered". This is the remaining wording from R3-36, with the corrected count of limited records otherwise consistent. The same wording appears on PDF pp.118 (method summary) and132 (references introduction). | supported with qualification | Chapter 13 and the references introduction say "source records", not "sources opened". | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 13 In short; References introduction in web and Markdown; PDF pp.118,132 |
| FINAL-14 | Plain-language edition, How the firms scored, privacy-rule sentence | The plain edition now discloses the risk cap, but states it at company level and for any FTC or court order. The frozen rule is product-specific: the event must involve that product's location data. The distinction matters because the main paper removed the Kochava measurement cap when that link was not established. | Plain says "a firm with an FTC order, court order or breach in the last three years gets at most 1 point on privacy". The current main matrix and rule require an order/finding/breach involving the product's own location data; the round-3 R3-01 correction records the Kochava measurement product's product-data link as unestablished and removes its cap while keeping the entity flag. The plain wording would reapply that removed cap, and would also cover unrelated orders. | supported with qualification | Plain edition: the cap applies when an order, finding or breach is about that product's own location data. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Plain-language edition, How the firms scored, privacy-rule sentence |
| FINAL-15 | Plain-language edition, What location data is, fourth number-kind definition | The plain definition substitutes "web address" for IP address, which can be read as a website URL. The geolocation source in the paper is the device's internet/network address, not the page URL. | Plain defines the inferred kind as "Guessed from the ad: the place came from the ad call or the web address." Chapter1 defines it as location from a bid request, an IP address or a user-declared field. A website URL does not mean the same thing as the user's IP address. | supported with qualification | Plain edition: "the phone's internet (IP) address" replaces "web address". | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Plain-language edition, What location data is, fourth number-kind definition |
| FINAL-16 | Chapter 4 H2 coded and ground-truth figure caption; LOC-B-F31 final wording; Chapter 11 Placer.ai headline/SWOT/Limits/ground-truth capability; Chapter 4 and Foursquare performance record LOC-FOURSQUARE-F26; Veraset/Cabrera source LOC-B-S08 and study metadata | The funding-based independence correction has reached the study tags and several claim classes but remains contradicted by the surrounding text and one Foursquare classification. Placer’s Minnesota comparison is repeatedly still called independent; Foursquare’s 2013 check-in study is also called independent while its funding is recorded unknown. Conversely, the Veraset matrix scorer records a full-text read with ESRC funding and no declared conflicts, but the general source/study record still says abstract-only and funding unknown. Those metadata should be updated from the full primary paper. | The live method says unknown-funded studies are not independent. LOC-B-S06 (alias LOC-PLACER-S24) records funding unknown, and the Choi study row now explicitly states independence not established. Nevertheless H2 coded says “one independent trail-site check”; the Placer headline says “one independent check found”, SWOT “has one independent check on record”, and ground-truth capability begins “Independent: one single-site ...”. LOC-B-F31 final wording likewise includes the Placer check under independent evidence. The figure caption says the newest independent comparison data end January 2023, but its January 2023 Advan study has been recoded unknown-funding; separate third-party from independent there. LOC-FOURSQUARE-S52 records funding unknown, while the Foursquare performance row and LOC-FOURSQUARE-F26 remain independent_measurement and Chapter 4 calls it independent. Primary Foursquar | supported with qualification | Minnesota and the 2013 HotNets study (no funding or acknowledgement section, re-opened) described as third-party with funding not stated in chapter 4, the ground-truth caption, LOC-B-F31, the Placer.ai profile, SWOT and headline; LOC-FOURSQUARE-F26 recoded to synthesis. Cabrera et al. (Veraset): full text read, ESRC ES/Y010787/1 funding and no conflict declared, so its source, study row and LOC-B-F14 are independent again; no Veraset score changes. Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: Placer SWOT weakness, the Placer Minnesota and Foursquare 2013 profile performance rows (now synthesis) and the LOC-FOURSQUARE-F26 confidence reason no longer call the unknown-funded studies independent. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 4 H2 coded and ground-truth figure caption; LOC-B-F31 final wording; Chapter 11 Placer.ai headline/SWOT/Limits/ground-truth capability; Chapter 4 and Foursquare performance record LOC-FOURSQUARE-F26; Veraset/Cabrera source LOC-B-S08 and study metadata |
| FINAL-17 | Chapter 1 Visitation is not lift; Chapter 8 What they leave out; LOC-E-F28 final wording; control-design figure explanatory caption; plain-language edition section 4 | The corrected six-of-eight lift count coexists with remaining descriptions of all nine records as documenting a general visit-lift method. The ninth record is GroundTruth’s exposed-only visit count, expressly coded as making no lift claim. | Chapter 1 says “nine document a general visit-lift method”; Chapter 8 says “Of the nine ... that document a general visit-lift method, seven use no holdout”; LOC-E-F28 final wording begins with the same nine-lift-method description before explaining that GroundTruth has no lift claim. The control figure caption calls the 14-row coding the general method for visit lift. By contrast the current abstract, Chapter 1 inshort, Chapter 12 and LOC-HYP-H3 correctly distinguish eight lift methods (six matched/synthetic) from nine visit-measurement methods (seven nonholdout). This is residual propagation of R3-24; the codes, sensitivity checks and H3 contradicted verdict need no change. Plain section 4 also explains lift as “Nine firms say how they do it. Seven use a look-alike group or no group at all. Two use a holdout,” without separating the no-lift ninth record. | supported with qualification | "Lift method" kept for the eight products that claim lift; nine visit-measurement methods in chapter 1, chapter 8, LOC-E-F28, the control-design caption and the plain edition. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 1 Visitation is not lift; Chapter 8 What they leave out; LOC-E-F28 final wording; control-design figure explanatory caption; plain-language edition section 4 |
| FINAL-18 | Chapter 11 Placer.ai profile ground-truth validation capability; LOC-PLACER-F26 study_design metadata | The retailer-quarter correction is complete in Chapter 4 and the final claim wording, but the Placer profile still describes directional agreement over twelve quarters, and current study-design metadata repeats twelve quarters. | Live Placer ground-truth capability: “Interested party: chain-level directional agreement 9 of 12 quarters vs SEC-reported figures (S40).” LOC-PLACER-F26 study_design ends “12 quarters; author-selected chains”. Its corrected final wording and Chapter 4 say twelve retailer-quarter comparisons, and the R3-22 correction record confirms the repeated quarter periods for six chains. Original author source: https://www.growthfactor.ai/resources/blog/retail-foot-traffic-data . This is a propagation miss of the accepted unit correction, not a new assessment of the interested blog’s accuracy. | supported with qualification | Placer.ai profile and LOC-PLACER-F26 study design: 12 retailer-quarter comparisons. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 11 Placer.ai profile ground-truth validation capability; LOC-PLACER-F26 study_design metadata |
| FINAL-19 | Chapter 5 comparator table, InMarket row; claim LOC-E-F07; InMarket deep profile capabilities.causal_design and method_disclosure; cohort-A matrix inmarket causal_design and method_disclosure | The accepted public-access qualification for the legacy LCI console is still missing from several evidence-bearing surfaces. Source metadata now says the glossary was read once but cannot be reproduced through a public route, while the comparator and claim ledger repeat its thresholds without that qualification and the profile/matrix call its details public method evidence. This is a reproducibility and verification-status inconsistency; a later sign-in wall alone does not prove that a previously captured passage is false or unusable. | Live chapter 5 table lists InMarket minimums '10m / 25m impressions; >95%+' without the caveat supplied later in the same chapter. LOC-E-F07 final_wording says 'LCI documents minimum samples and uncertainty' and review_outcome remains supported_as_stated. InMarket profile method_disclosure starts 'Seven-element count from public sources' and credits projection formula and standard error/confidence from S11. The matrix method_disclosure score 3 likewise counts those legacy-console details; its causal_design rationale repeats the legacy thresholds. Canonical source LOC-C-S02 (alias LOC-INMARKET-S11), https://lciadminconsole.inmarket.com , now explicitly records later sign-in redirects and says claims resting on it are unverified from public sources. The current dated synthetic-control summary at https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ sup | supported with qualification | InMarket LCI console details qualified as read once, undated and now behind a sign-in in the chapter 5 table, LOC-E-F07, the profile and the matrix rationales; the public synthetic-control description is distinguished; the disclosure score stays 3 and the rationale states it would be 2 without the console. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 5 comparator table, InMarket row; claim LOC-E-F07; InMarket deep profile capabilities.causal_design and method_disclosure; cohort-A matrix inmarket causal_design and method_disclosure |
| FINAL-20 | Radar deep profile performance entry for “Accuracy down to 5 meters”, limits; associated exported deep-profile records | The profile treats a configurable requested-accuracy default as contradicting a best-case achieved-accuracy claim. That inference is unsupported: a hundred-metre requested OS accuracy is neither a measured error nor a lower bound on the accuracy delivered, and Radar also documents a high-accuracy preset. | The Radar performance entry quotes 'Accuracy down to 5 meters' and says it 'conflicts with the SDK’s default requested accuracy class (hundred metres) in S13'. The reopened primary documentation https://docs.radar.com/sdk/tracking (iOS options and presets) describes desiredAccuracy as the requested accuracy, maps medium to kCLLocationAccuracyHundredMeters, allows custom parameters, and sets Continuous to high/kCLLocationAccuracyBest. The current product page https://radar.com/product/geofencing makes the best-case claim. Apple's official Energy Efficiency Guide, https://developer.apple.com/library/archive/documentation/Performance/Conceptual/EnergyGuide-iOS/LocationBestPractices.html , section 'Reduce Accuracy of Standard Location Updates Whenever Possible', explicitly explains that Core Location often provides more accurate data than requested. LOC-RADAR-F06's own limitations already di | supported with qualification | Radar profile: the five-metre best case and the SDK's requested accuracy classes are stated separately; "conflicts" removed (docs re-opened: Efficient and Responsive request .medium, Continuous .high). | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Radar deep profile performance entry for “Accuracy down to 5 meters”, limits; associated exported deep-profile records |
| FINAL-21 | Cohort-B matrix azira-audience register.us_available and us_availability rationale; corresponding product-register export | The register-to-matrix reconciliation was applied to Azira Measurement but still leaves Azira Audience marked US-unconfirmed, despite the same row documenting a named US production case and US observations. The register and scored rationale therefore continue to publish incompatible states. | The embedded azira-audience register says us_available="unconfirmed (LOC-U-S35)". Its verified US-availability cell says "Anchor 5 met on substance, capped at 3 by grade D" and cites a Visit Fort Worth campaign in Chicago, built from train-station geofences and first-party data, with 8,445 attributed Fort Worth visits (LOC-SCB1-S02). The rationale itself acknowledges the stale register status. Primary case URL: https://www.azira.com/case-studies/how-azira-and-visit-fort-worth-increased-destination-awareness-and-drove-8k-visits-during-a-competitive-travel-window . This is the same propagation problem addressed for Azira Measurement, Mobilewalla and Adsquare under R3-37; those three original rows have been updated. | supported with qualification | Azira Audience register records the named US production case (vendor claim) and keeps general availability separate. Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: The stale final sentence of the Azira Audience US-availability rationale replaced with the register's current status. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Cohort-B matrix azira-audience register.us_available and us_availability rationale; corresponding product-register export |
| FINAL-22 | Chapter 7, 'The instruments read' governance table; main five-state-law table citations; figure fig-state-laws caption / figures.json | The five-state correction is present in the headline and main table, but the underlying legal register and some pointers remain at older law states. The final governance table still describes Connecticut only by its consent-to-sell rule and Virginia only by its definition; the main sale-ban rows then link to those older sources rather than their operative bans. The state figure's explanation still enumerates only four states. | In the live Markdown, the final governance table lists Connecticut PA25-113 as 'Controller shall not sell ... without ... consent' and Virginia §59.1-575 as a definition with 'consent duty in 59.1-578 not opened'. The main table correctly names Connecticut PA26-64 effective 1 October 2026 but cites [155], the older PA25-113; its Virginia sale-ban row cites [162], §59.1-575, instead of [157], §59.1-578. The paper already registers PA26-64 as [159]. The live figure title says five states, while its caption lists Virginia and Connecticut 'joining Maryland ... and Oregon' with no New Jersey sentence. Newly opened operative Virginia text https://law.lis.virginia.gov/vacode/title59.1/chapter53/section59.1-578/ A.6 prohibits selling or offering for sale precise geolocation (web ref turn144view1); newly opened New Jersey enacted text https://pub.njleg.state.nj.us/Bills/2026/PL26/25_.HTM confirms | supported with qualification | Governance table adds Virginia 59.1-578(A)(6) and Connecticut PA 26-64 secs. 14-15 (both re-opened) and marks the PA 25-113 consent row superseded for sale from 1 October 2026; LOC-H-F16 and LOC-H-F13 now cite the operative texts; the state-law caption adds New Jersey. The Virginia 1 July 2026 date rests on the existing record (the bill page did not load in this sweep). Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: The new Virginia and Connecticut register rows lost their sources because verifier-registered ids (-V) failed the source-id pattern; the pattern now accepts them, and the rows cite LOC-H-V01 and LOC-H-V03. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 7, 'The instruments read' governance table; main five-state-law table citations; figure fig-state-laws caption / figures.json |
| FINAL-23 | Chapter 3, 'The bounded finding', sentence beginning 'The search has limits'; LOC-G-F20 and related final-wording surfaces if they repeat the sentence | The categorical statement that the five named vendors 'file nothing on EDGAR' conflicts with the paper's own Placer evidence. Lack of a periodic reporting obligation is different from having no SEC filings. | The live chapter says 'Foursquare, Placer.ai, InMarket, GroundTruth and Blis file nothing on EDGAR'. Its own numbered sources [720]–[723] are Placer Labs' EDGAR submissions record and three Form Ds, and LOC-PLACER-S30 states that the index lists nine D/D-A filings from 2019 to 2024, with no periodic reports. The current Placer profile uses those filings for its financing disclosures. Primary URLs already registered include https://data.sec.gov/submissions/CIK0001786302.json and https://www.sec.gov/Archives/edgar/data/1786302/000178630224000003/primary_doc.xml . Both direct primary opens failed in this sweep, so the finding is the demonstrable contradiction between the published chapter and its own cited SEC records, not an asserted new financing fact. | contradicted | Chapter 3 and LOC-G-F20: the named private vendors file no 10-K or 10-Q reports; Placer Labs has filed Form D notices. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 3, 'The bounded finding', sentence beginning 'The search has limits'; LOC-G-F20 and related final-wording surfaces if they repeat the sentence |
| FINAL-24 | LOC-U-F02 source_ids/support in the live claim ledger; numbered sources/source register; Foursquare source LOC-FOURSQUARE-S14 reliability note | The qualified Foursquare completion wording is now correct, but the two primary participant accounts that support it have not been registered or linked to the final claim. The profile's secondary-source record still says Crowley's original could not be located even though the published corrections ledger says both originals were reopened. | LOC-U-F02.final_wording attributes completion accounts to Crowley on 20 October and a shareholder on 24 September; source_ids remains LOC-U-S07, LOC-U-S06 and LOC-U-S08, whose support passages are the September announcement and homepage banners. No record in the live 790-source register has either participant URL. Corrections R2B-25 and R2B-34 say both posts were opened on 24 October. LOC-FOURSQUARE-S14 still says Crowley's original was not located. Freshly reopened primary https://teendrama.substack.com/p/some-final-words-on-foursquare (20 October, turn155view0) describes acquisition about a month earlier; https://buttondown.com/rickwebb/archive/good-morning-hello-how-are-you-1820/ (24 September, turn155view1) describes the sale and receipt of counsel's sale paperwork. Neither supplies a verified legal closing date. | supported with qualification | Crowley (20 Oct 2026) and Webb (24 Sep 2026) posts registered as LOC-NFA-S04 and S05 and cited on LOC-U-F02 (both re-opened; neither gives a closing date); the S14 note no longer says the original was not located. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | LOC-U-F02 source_ids/support in the live claim ledger; numbered sources/source register; Foursquare source LOC-FOURSQUARE-S14 reliability note |
| FINAL-25 | Vistar profile risks[1], capabilities.privacy, timeline 2024-05-01, SWOT threats[2] | The earlier blanket-sharing error is fixed, but the replacement still omits 'precise' when describing InMarket's sale/licensing ban. Read on its own, the profile now implies that the order also bans coarse location sales. | The live Vistar SWOT says 'Partner InMarket has been barred since May 2024 from selling or licensing location data'; the risk, privacy note and timeline use the same unqualified scope. The freshly reopened FTC order https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf (turn137view1), definition F on printed p4, defines Location Data as precise location and excludes coarse data and data used solely to generate coarse location then deleted within 48 hours. Provision II bans sale/licensing for valuable consideration of that defined Location Data. Chapter 7 already expresses the precise/coarse distinction correctly. | supported with qualification | Vistar profile, SWOT, chapters 6 and 7: InMarket's ban is on selling or licensing precise location data as the order defines it. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Vistar profile risks[1], capabilities.privacy, timeline 2024-05-01, SWOT threats[2] |
| FINAL-26 | Chapter 9, "Which sellers name their sources", paragraph after the six-product source table (live.md line 852); LOC-I-F23/register context | The earlier Broadsign register correction has not reached chapter 9. It still calls Broadsign Ads audience targeting "found in discovery and not in the register", although the live register includes it and chapter 6 now correctly says its US availability is unconfirmed. This is a remaining propagation failure from R3-26, not a change to H4. | The live paragraph reads "Broadsign Ads' audience targeting, found in discovery and not in the register, names Native Touch data...". The chapter 6 H4 row instead reads "Broadsign Ads (registered; US availability unconfirmed)". The published product register contains Broadsign Ads audiences as a listed cohort-B product with US availability unconfirmed. Those two settled facts can coexist; "not in the register" cannot. | supported with qualification | Chapter 9: Broadsign Ads audiences registered but outside the US count because US availability is unconfirmed. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 9, "Which sellers name their sources", paragraph after the six-product source table (live.md line 852); LOC-I-F23/register context |
| FINAL-27 | Chapter 8, "What they leave out" and September 2018 MRC presentation references; chapter 10, "What vendors claim" (proprietary VIOOH/Displayce paragraph); LOC-A-F07.source_ids and numbered-reference wiring | Several substantive corrections are present in the prose but their primary or actual supporting sources have not been wired into the citations. Readers following the numbers reach sources that cannot establish the corrected statement. | The chapter 8 statement about the September 2022 Outcomes and Data Quality Standards cites only [445], which is the March 2017 Location-Based Advertising Measurement Guidelines. The actual 2022 source is registered as LOC-A-V01 but has no reference number, and LOC-A-F07.source_ids still contains only LOC-A-S02. The 2022 primary PDF independently opened for this sweep confirms physical-visitation coverage in section 2.1.6 (printed p12) and incrementality/RCT discussion later: https://mediaratingcouncil.org/sites/default/files/Standards/MRC%20Outcomes%20and%20Data%20Quality%20Standards%20(Final).pdf . Likewise chapter 8's September 2018 presentation sentence cites [166,167,520,568] although the actual presentation is [164]/LOC-NFA-S01 (already correctly in the figure citation). In chapter 10 the proprietary VIOOH/Displayce statement cites [198,200,201,205], which are Prebid code, the AAO r | supported with qualification | LOC-A-F07 now cites the September 2022 MRC Outcomes and Data Quality Standards (re-opened: s. 2.1.6 p. 12, s. 2.1.9 p. 16); LOC-A-F11 cites the September 2018 presentation (LOC-NFA-S01); LOC-J-F13 cites the VIOOH and Displayce records. Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: LOC-A-F07 limitations no longer say the 2022 MRC standard was unopened. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 8, "What they leave out" and September 2018 MRC presentation references; chapter 10, "What vendors claim" (proprietary VIOOH/Displayce paragraph); LOC-A-F07.source_ids and numbered-reference wiring |
| FINAL-28 | Chapter 9, "Footfall attribution without a holdout", sentence "The channel's public guidance names matched designs, not holdouts"; chapter 8 "IAB guidance" and LOC-I-F11 access limitation | The full July 2025 IAB DOOH Measurement Guide is now publicly retrievable, and it includes geographic holdouts. The paper's broad claim that the channel's public guidance names no holdouts is therefore too strong. The product-default H3 verdict is unaffected. | A normal public web search found and this sweep opened the 31-page IAB-branded July 2025 PDF at https://assets.contentstack.io/v3/assets/bltbeaed4aed52c223a/blt494e044b82a9dcb0/68cb0a53da7537f88855f47c/iab-dooh-measurement-guide.pdf . Chapter 7, printed p24 (PDF index23), recommends synthetic-control and matched-market/store methods and says combining synthetic models with geographic holdouts can reduce exposure contamination. The public IAB landing page at https://www.iab.com/guidelines/dooh-measurement-guide/ still displays an account gate, so the earlier research access limitation was plausible, but the public PDF closes it. This is methodological guidance, not evidence that any of the six screened products uses a holdout by default. | contradicted | IAB's July 2025 DOOH Measurement Guide read from its public PDF (registered as LOC-NFA-S03): it recommends synthetic-control and matched-market or matched-store tests and describes geo holdouts as reducing contamination (pp. 23-24). Chapter 9, LOC-I-F11 and LOC-A-F23 updated; the product-default H3 coding is unchanged. Closure check of release 9633c5fb (24 October 2026) found a residue, fixed in the next release: Chapter 13 limits no longer list the IAB guide as unread; it records the later full read of the public PDF. | Shared rules: every outside finding is checked against its primary source or the paper's own data before it changes anything; scores follow the frozen anchors. | Chapter 9, "Footfall attribution without a holdout", sentence "The channel's public guidance names matched designs, not holdouts"; chapter 8 "IAB guidance" and LOC-I-F11 access limitation |



## Appendix: hypotheses, universe and downloads


### The registered hypotheses and how they read

Each hypothesis was registered on 24 October 2026, before any research agent ran, with its rules for supported, contradicted and unresolved. Four were clarified after extraction; each clarification is dated and published next to the original rule in data/hypotheses.json.

| | Hypothesis | Reading | Basis | Chapter |
|---|---|---|---|---|
| H1 | SDK supply contracted after the 2019–2021 permission changes and vendors substituted other sources | Unresolved | No comparable US panel series; no filing attributes a quantified supply change to a platform change [275, 63, 289, 427, 428, 505, 504, 88] | [2](#supply), [3](#platforms) |
| H2 | Visit estimates agree with independent ground truth within the vendor's stated error | Unresolved overall; adverse for SafeGraph's 2018-vintage panel | No independent multi-site retail comparison found [98, 99, 102] | [4](#validation) |
| H3 | Visit lift rests on randomised or holdout controls by default | Contradicted | Seven of nine documented methods are matched, synthetic or exposed-only [5, 19, 119, 17, 20, 21, 22, 25, 6, 446] | [5](#measurement) |
| H4 | Location audiences rest mainly on bidstream or IP location | Unresolved | Fewer than four products document a primary source [265, 420, 421, 422, 458, 457] | [6](#audiences) |
| H5 | FTC-ordered companies no longer sell precise or sensitive location data | Contradicted | Outlogic and Collective Data Solutions market precise-location offers; Gravy dba Unacast offers device-level data [128, 137, 142, 145, 134, 130] | [7](#regulation) |
| H6 | A US visit metric holds current MRC accreditation | Contradicted (24 October 2026) | No visit or location metric on any of MRC's eleven lists [171, 172, 173, 174, 175, 176, 177, 178, 179, 180] | [8](#standards) |
| H7 | Location signals reach production use in agent-driven buying | Unresolved | Highest state: one implemented connector [198, 200, 201, 205] | [10](#agents) |

The four coded questions are answered in the chapters: Q1, what kind of number each product sells, in [chapter 1](#defining); Q2, which mobility data sits under US DOOH audiences, in [chapter 9](#dooh); Q3, how the market consolidated, in [chapter 2](#supply); Q4, what Apple and Google changed and what effects are documented, in [chapter 3](#platforms).

### The vendor universe

Every company and product the research met, with owner, cohort, status and, for scored products, the matrix row. Listed and historical products are described, not scored.

**Vendor universe**

| Company / product | Legal entity / owner | HQ | Scope group | Formats | Buying routes | Main geographies | Status at cutoff | Src |
|---|---|---|---|---|---|---|---|---|
| Foursquare | Foursquare Labs, Inc.; acquisition by Infillion announced 2026-09-18, completion not confirmed | New York (per Infillion release dateline; HQ not stated on Foursquare page) | specialist | A attribution; B audience + Proximity geofencing; C Places POI data + Spatial suite; D Places API + Movement SDK | The Trade Desk, AWS named among '550+ integrated partners' | global; US | active; acquisition announced (pending) | [347] |
| Infillion | Infillion (private) | New York, NY (release dateline) | adtech | DSP, supply, data, creative; owns Gimbal (D) and Catalina; pending owner of Foursquare | Managed service, self-serve, 'Infillion Agent Connector' | US | active; acquirer | [602] |
| Gimbal (Infillion Gimbal) | Infillion | not stated | developer | D location SDK, geofences, iBeacon; first-party visit measurement and segmentation for app owners | REST API, CRM/analytics ingestion, engagement platforms | not stated | active (owned by Infillion; gimbal.com redirects) | [473] |
| Placer.ai | Placer Labs, Inc. (private) | not stated on pages opened | specialist | C analytics platform, enrichment, data marketplace; campaign lift use (A-like) without a named product | Dashboard, embeddable widgets, enrichment | US | active | [293] |
| InMarket | InMarket (private) | not stated on page | specialist | A Measurement (LCI lineage, 'Campaign Incrementality'); B Audiences; Activation; GeoLink self-serve | GeoLink; DSPs per workstream F | US | active | [133] |
| GroundTruth | ZeroToOne.AI (per GroundTruth's own footer) | not stated | specialist | A visit measurement; B Ads Manager, Predictive Audiences, real-world targeting | Self-serve, managed service, programmatic | US | active; acquired (ZeroToOne.AI, date not on page) | [489] |
| Blis | T-Mobile US, Inc. (completed 2025-03-03, US$180m) | not readable | adtech | B omnichannel DSP with geospatial targeting; A outcome measurement (not readable) | not readable | not readable | active; acquired by T-Mobile | [36] |
| Vistar Media | T-Mobile US, Inc. (completed 2025-02-03, US$621m) | not stated | adtech | DOOH DSP/SSP/ad server (Cortex, Adstruc); B DOOH audience targeting; A attribution not named on homepage | Own DSP and SSP | global; US | active; acquired by T-Mobile | [36] |
| Reveal Mobile | Reveal Mobile, Inc. (owner not stated) | Raleigh, NC | specialist | A OOH measurement (foot traffic, brand, sales lift); OOH impressions; B OOH audience retargeting | Mira Console, managed service, enterprise API; DSPs/SSPs | US | active | [531] |
| Azira | not stated on page (successor to Near Intelligence per workstream D) | not stated on homepage | specialist | A Azira Measurement; B Azira Audience, Azira Campaigns (DSP); C Azira Insights, Azira One | Own DSP; data feeds | global; US | active | [458] |
| Unacast (incl. Gravy Analytics) | Unacast (owner not stated) | Ashburn, VA | specialist | C location data, foot traffic; identity graphing and Data Linkages (MAID-HEM-IP) | Data delivery; Datarade | global; US | active; merged with Gravy Analytics | [125] |
| Adsquare | not stated | Berlin; offices incl. New York | specialist | A attribution dashboard ('Validation'); B audiences (names per workstream F) | DSPs per workstream F | global; US office | active | [455] |
| Lifesight | not stated | not stated | adtech | MMM, incrementality testing, attribution; no location product | n/a | global | pivot (out of scope) | [481] |
| Cuebiq | Cuebiq Group, LLC | New York, NY (1460 Broadway) | specialist | A Footfall Measurement; B Cuebiq Audiences; C Data Platform (Spectus domain redirects here) | Cuebiq API; dashboard | US | active | [479] |
| Spectus | Cuebiq Group, LLC (presumed from redirect) | n/a | specialist | C clean-room data platform (historical brand?) | n/a | n/a | unconfirmed (domain redirects to Cuebiq) | [480] |
| Precisely (PlaceIQ) | Precisely Software Incorporated | not opened | enterprise | A/B/C PlaceIQ lines (Precisely Audiences per workstream F) | not opened | US | unconfirmed by U (page 403) | [421] |
| Kochava | Kochava Inc. (private) | not stated | adtech | A store/footfall visitation via Kochava Foundry; MMM; app measurement; Collective not named | Platform | global; US | active; data business moved to Collective Data Solutions (per FTC, workstream H) | [24] |
| Arrivalist | not stated | not stated | specialist | A destination media measurement; C visitor and spend intelligence | Platform | US | active | [609] |
| Samba | Samba TV | not stated | adtech | TV data; no location product seen on consumer site | n/a | US | not assessed (business site not opened) | [526] |
| Locala | not stated | not stated | adtech | B adaptive advertising with store-visit claims; no named location product | not stated | not stated | active | [610] |
| Adsmovil | Adsmovil (founded 2011 by Alberto Pardo; 'minority-owned') | Miami, FL | adtech | US Hispanic media network; location data as a targeting signal | O&O and premium Hispanic inventory; CTV, video, display, audio, DOOH | US; Latin America | active | [611] |
| Mobilewalla | not stated | not stated | specialist | B audience segments; location-based marketing; telecom and consumer intelligence lead | DMPs/DSPs per workstream F | global | active (copyright 2025) | [141] |
| Simpli.fi | not stated on page (Blackstone/GTCR per workstream D) | not stated | adtech | B geo-fencing, addressable geo-fencing; Simpli.Hub multi-location | Own DSP | US | active | [612] |
| Outlogic | not stated on page (Digital Envoy per workstream D) | not stated | specialist | C Observation Panel and attribution data; B audiences | Snowflake; AWS S3 daily and hourly; REST API; Eyeota; Adsquare | global; US (Texas notice) | active | [128] |
| Lotame | not stated | not stated | adtech | Data and identity platform; no location product seen | n/a | global | out of scope | [524] |
| Verve Group | Verve Group, Inc. | not stated | adtech | Publisher and audience products; no location product seen | n/a | global | out of scope | [525] |
| SafeGraph | SafeGraph Inc. | not stated | specialist | C Places, Geometry, Address, Spend; no visit product | Integrations to buyer environments | global | active (visit data no longer sold; Patterns sold to Advan per workstream D) | [613] |
| Veraset | not stated | not stated | specialist | C Movement, Visits, Home & Work, Trips; MatchLink (HEM/IP); advertising audiences and attribution use | Orchestrator; data delivery | global; North America | active | [478] |
| Advan Research | not stated | not stated | specialist | C FIT, REI, REVEAL, SPENDVIEW, Patterns+ | FTP, AWS S3, Google Cloud; dashboards | US (45M devices claimed) | active | [477] |
| Huq Industries | not stated | not stated (UK-framed) | specialist | C footfall, spend, catchment platform | Platform, free trial | UK; global claimed | active | [614] |
| Arity | not stated on pages opened | not stated | enterprise | B Arity Audiences, Marketing Platform; C Geosight, Roadway Insights, Road Traffic Analytics; D Arity SDK | not stated | US | active; acquirer of INRIX assets (announced 2026-10-16) | [493] |
| INRIX | Arity (agreement announced 2026-10-16; close expected 2026) | not opened | enterprise | C transportation intelligence and mobility analytics | not opened | not opened | acquisition announced (pending) | [494] |
| Quadrant | Appen (per 'an Appen company') | not stated | specialist | C mobile location data, POI data, hashed-email data | Datarade; technical docs | global | active | [425] |
| Tamoco | not stated | not stated | specialist | C geospatial data; B audiences; A measurement; insights | Data catalog | global | active | [615] |
| Onemata | unknown | unknown | specialist | unknown | unknown | unknown | unconfirmed (domain shows Wix error) | [482] |
| PassBy | Passby Technologies Ltd | London, UK | specialist | C retail market intelligence (visits modelled); AI agents; feeds and API | Feeds & API; agents | UK; US phone listed | active | [616] |
| Zartico | Zartico, Inc. | not stated | specialist | C destination visitor intelligence; A marketing measurement for destinations | Platform | US | active | [617] |
| Radar | Radar Labs, Inc. | not stated | developer | D geofencing, maps, location verification (Protect), trips, fraud (Reveal); MCP server announced | SDKs, APIs, integrations, MCP | global; US | active | [390] |
| Bluedot | Rezolve AI plc (closed 2025-02-20) | Delaware and Australian entities | developer | D mobile location technology (geofencing) | not opened | US, Australia | acquired (bluedot.io redirects to rezolve.com) | [474] |
| Skyhook | Qualcomm Technologies (completed by May 2022) | Boston (historical) | developer | Wi-Fi positioning (now Qualcomm Terrestrial Positioning Service) | n/a | global | acquired (historical) | [476] |
| Complementics | ECONTEXT.AI LLC dba Complementics | not stated | specialist | C location feeds, POI visitation, audiences, app usage, identity (MAID-HEM-IP), consumer data | Direct delivery | global; US consumer data | active | [426] |
| Datastream Group | unknown | unknown | specialist | unknown (search lead: location feed) | unknown | unknown | unconfirmed (domain parked) | [483] |
| Plot Projects | unknown | unknown | developer | geofencing SDK (seed lead) | unknown | unknown | unconfirmed (403) | [484] |
| Esri | Esri | not stated | enterprise | ArcGIS Business Analyst (demographic, spending, census data); no mobility panel named | Desktop, web, mobile apps | global; US | listed (platform, not a location-data product) | [618] |
| HERE | HERE | not stated | enterprise | Location data platform; HERE Anonymizer; maps and mobility services | Platform | global | listed | [619] |
| Mapbox | Mapbox | not stated | developer | D maps, navigation, search, Places API; C Movement Data | SDKs, APIs | global | active | [620] |
| Irys | unknown | not stated | specialist | C datasets (audience creation, behavioural analytics) | Datarade | not stated | listed on marketplace (unconfirmed) | [485] |
| Intuizi | unknown | not stated | specialist | C device signal data ('400mm+ Devices') | Datarade | global | listed on marketplace (unconfirmed) | [485] |
| DRAKO | unknown | not stated | specialist | C geolocation data | Datarade | not stated | listed on marketplace (unconfirmed) | [485] |
| Mobvista | Mobvista Inc. (HKEX: 1860) | not stated | adtech | Mobile user acquisition (Mintegral), analytics; no location product | n/a | global | out of scope | [621] |
| Precisely PlaceIQ attribution | Precisely |  | enterprise | A visit measurement |  | US | active | [622, 27] |
| Precisely Audiences | Precisely |  | enterprise | B audiences |  | US | active | [622, 421] |
| Broadsign Ads audiences | Broadsign |  | adtech | B audiences |  | US | active | [538, 538] |
| Advan Patterns+ | not stated |  | specialist | C1 foot-traffic analytics |  | US | active | [477, 623] |
| Precisely Visits | Precisely |  | enterprise | C1 foot-traffic analytics |  | US | active | [624, 625] |
| Huq Industries | not stated |  | specialist | C1 foot-traffic analytics |  | US | active | [626, 614] |
| Unacast device-level location data | Unacast (Gravy acquired November 2023) |  | specialist | C2 movement data |  | US | active | [137, 328] |
| Collective Data Solutions data feeds | Kochava Inc. (subsidiary; FTC v. Kochava, D. Idaho, Dkt. 101 and 138) |  | specialist | C2 movement data |  | US | active | [627, 145] |
| Unacast foot-traffic data |  |  | specialist | C1 foot-traffic analytics |  | US | active | [137, 331] |

### Downloads

Every dataset behind the paper, in CSV and JSON.

- `location-intelligence-2026.pdf` — the print edition: A4, with page numbers and chapter bookmarks
- `location-intelligence-2026.md` — the reading edition in Markdown, with numbered references
- `data/source-register.csv` — every source with publication, event and access dates, type, geography, funding, access limits and reliability notes
- `data/claim-ledger.csv` — every material claim with sources, passage locator, scope, evidence class, review outcome, confidence and final wording
- `data/capability-matrix-wide.csv` — vendors × dimensions, scores only (0–5, n/e, n/a)
- `data/capability-matrix-long.csv` — one row per cell with availability, evidence grade, date, rationale and sources
- `data/capability-dimensions.csv` — the dimension definitions and anchors
- `data/gq.csv` — Ground Quotient per vendor: group means, coverage, band, the reweightings and every dimension score
- `data/gq-method.csv` — GQ groups, weights, shrinkage, field mean, band thresholds, and the reweighting, shrinkage and missing-evidence tests
- `data/gq-missing-evidence.csv` — every scored cell turned into n/e one at a time, with the GQ before and after
- `data/swot.csv` — every SWOT line for the profiled vendors with its basis (record, vendor documentation, vendor claim, inference) and sources
- `data/deep-profiles.csv` — the deep-profile summaries: what is sold, buyers, earnings, dependencies, fit, limits
- `data/deep-profile-records.csv` — every profiled capability, performance claim, financial figure and timeline event with sources
- `data/study-evidence.csv` — studies with design, sample, sponsor and limits
- `data/corrections-ledger.csv` — every change made after verification and challenge passes, with implementation and verification
- `data/vendor-universe.csv` — the reasoned vendor sample with scope group, owner, formats, routes and status
- `data/agentic-map.csv` — AI and agentic capabilities with status labels and observability
- `data/rules.csv` — rules and governance instruments by jurisdiction
- `data/decision-tool.csv` — the decision tool rows
- `data/product-register.csv` — the product register: every scored or listed product with its purchased service, operating entity, owner, cohort, inclusion reason, aliases and US availability
- `data/hypotheses.csv` — the hypotheses registered before extraction, with the rules for supported, contradicted and unresolved
- `data/figure-data.json` — the canonical dataset behind every data figure
- `data/derived-checks.json` — every derived number with its formula and inputs
- `data/stats.json` — every count the prose and figures quote about the research itself, computed from the ledgers
- `data/second-pass.csv` — the blind second scoring pass: one row per re-scored cell with both passes’ scores, grades and reasons, and the adjudication of each material disagreement
- `data/second-pass.json` — the same cells with both passes’ sources, each re-scored row’s band under both passes, and the agreement statistics’ inputs
- `data/scoring-rules.json` — dimension definitions, anchors, cell rules and the GQ definition, fixed before scoring



## References

The 734 sources cited in this paper, numbered in order of first citation. The full register of 793 source records, with access dates, access limits and reliability notes, is in data/source-register.csv.

[1] Location-based Marketing Playbook for Retailers. IAB Mobile Location Working Group, 2017-04. https://www.iab.com/wp-content/uploads/2017/04/IAB_Location-Playbook-for-Retail-Marketers_FINAL.pdf Accessed 2026-10-24.

[2] (Author's work) Measurement, on the Street's Terms. Evgeny Popov, No Fluff Advisory, 2026-08-12. https://nofluffadvisory.com/writing/measurement-on-the-streets-terms/ Accessed 2026-10-24.

[3] Location data at Placer.ai (help centre article). Placer.ai (Placer Labs Inc.), unknown. https://connect.placer.ai/use-cases/location-data-at-placer-ai Accessed 2026-10-24.

[4] Visitation Trends & Key Insights from 2019-2021 (Arrivalist presentation to Lee County VCB industry meeting). Arrivalist (hosted by Lee County Visitor & Convention Bureau, visitfortmyers.com), 2022-03 (upload path). https://www.visitfortmyers.com/sites/default/files/2022-03/LVCB%20-%20Industry%20Meeting%20-%20Arrivalist%20Distribution%20Version.pdf Accessed 2026-10-24.

[5] Azira Location Measurement and Attribution Methodology. Azira, 2025 (footer 'Azira @ 2025'; no publication date). https://business.azira.com/azira-location-measurement-and-attribution-methodology Accessed 2026-10-24.

[6] Observed Visits (GroundTruth Help Center). GroundTruth, 2024-09 (article text says 'as of September 17 2024'; last-updated not shown). https://help.groundtruth.com/hc/en-us/articles/360000695267-Observed-Visits Accessed 2026-10-24.

[7] Proving Linear TV Drives Real-World Results (Foursquare Attribution case study). Foursquare, 2026-04-16. https://foursquare.com/?p=56555 Accessed 2026-10-24.

[8] OOH Case Study: Target (Vistar Media programmatic DOOH, measured with Foursquare). Vistar Media, hosted by the Out of Home Advertising Association of America (OAAA), 2022-12 (PDF last modified 2022-12-02). https://oaaa.org/wp-content/uploads/2022/12/Target_MPA_2022_FINAL.pdf Accessed 2026-10-24.

[9] Vistar Media US Case Study: DOOH QSR, Jack in the Box. Vistar Media, hosted by OAAA, 2024-02 (upload path). https://oaaa.org/wp-content/uploads/2024/02/Vistar-Media-US-Case-Study-DOOH-QSR-Jack-in-the-Box.pdf Accessed 2026-10-24.

[10] Case Study: Multi-Location Quick-Service Restaurant Increases In-Store Visits with Simpli.fi. Simpli.fi, 2025-04 (file name 'Apr25'; uploaded 2025-07). https://simpli.fi/wp-content/uploads/2025/07/Simpli.fi_Case-Study_Multi-Location-QSR_Apr25.pdf Accessed 2026-10-24.

[11] Measure the impact of TikTok campaigns on your store visits via InMarket. TikTok for Business (with InMarket), 2024-10-14. https://ads.tiktok.com/business/en/blog/inmarket-store-visits-sales-lift-measurement Accessed 2026-10-24.

[12] Industry-First A/B Test Proves 29% Lift in Store Visits for Big Box Retailer (PlaceIQ case study). PlaceIQ (PDF hosted by MMA Global), undated (PlaceIQ-branded, so before Precisely's 2022 acquisition; a Super Bowl campaign, year not stated). https://mmaglobal.com/files/a-b_control_case_study_final.pdf Accessed 2026-10-24.

[13] Tropical Smoothie Cafe (Blis case study). Blis, undated. https://blis.com/tropical-smoothie-cafe-case-study/ Accessed 2026-10-24.

[14] Ground Truth Analysis of Unacast Foot Traffic Data. Unacast, 2021-02-25 (last updated 2024-08-05). https://www.unacast.com/post/examining-ground-truth-data Accessed 2026-10-24.

[15] Sample our Clean & Privacy Compliant Data (Mobilewalla data sample request page). Mobilewalla, undated (footer © 2025). https://www.mobilewalla.com/data-sample-request Accessed 2026-10-24.

[16] Foursquare's Attribution Methodology Takes Your Media Dollars Further. Foursquare, 2022-06-06. https://foursquare.com/?p=11557 Accessed 2026-10-24.

[17] LCI admin console glossary (Location Conversion Index). InMarket (page carries NinthDecimal branding), undated. https://lciadminconsole.inmarket.com Accessed 2026-10-24. Access: partial extract.

[18] Projected Visits. GroundTruth Help Center (Marc Fevrier), 2025-11-15. https://help.groundtruth.com/hc/en-us/articles/12770162930707-Projected-Visits Accessed 2026-10-24.

[19] Omnichannel measurement - Smart Holdout Groups. Blis, unknown. https://blis.com/omnichannel-measurement/ Accessed 2026-10-24.

[20] OOH Performance Measurement - Reveal (Net Lift Study). Reveal Mobile, Inc., unknown (copyright 2026). https://revealmobile.com/solutions/foot-traffic-attribution/ Accessed 2026-10-24.

[21] Inside adsquare's Attribution Dashboard. adsquare, 2025-05-20. https://adsquare.com/resources/the-lab/inside-adsquares-attribution-dashboard Accessed 2026-10-24.

[22] THE 20 HOURS THAT MATTER: Bridge the LTV, CTV, and IRL Attribution Gap. Cuebiq Group, LLC, unknown (copyright 2026; privacy policy updated 2026-10-22). https://cuebiq.com/?p=36373 Accessed 2026-10-24.

[23] InMarket | Vistar Media Partner. Vistar Media, unknown (copyright 2026). https://www.vistarmedia.com/partner-inmarket Accessed 2026-10-24.

[24] Kochava homepage and navigation. Kochava Inc., unknown (copyright 2008-2026). https://www.kochava.com/ Accessed 2026-10-24.

[25] MediaInsights (Arrivalist). Arrivalist, 2026-10-23 (page metadata). https://www.arrivalist.com/media-insights Accessed 2026-10-24.

[26] Overview (Lifesight Measurement module). Lifesight help centre (Intercom), 2024-04-23. https://intercom.help/lifesight-support/en/articles/3026093-overview Accessed 2026-10-24.

[27] Precisely Audiences - privacy-centric audience segment data. Precisely, unknown (2026). https://www.precisely.com/product/precisely-audiences Accessed 2026-10-24.

[28] Attribution (product page). Foursquare, unknown (copyright 2026). https://foursquare.com/products/attribution/ Accessed 2026-10-24.

[29] Snap Inc. Form 10-Q for the quarter ended September 30, 2019, Note 'Business Acquisitions and Divestitures' (XBRL report R12). Snap Inc. (SEC EDGAR), 2019-10. https://www.sec.gov/Archives/edgar/data/1564408/000156459019037449/R12.htm Accessed 2026-10-24.

[30] Foursquare buys Placed from Snap Inc. on the heels of $150M in new funding. TechCrunch (Sarah Perez), 2019-05-30. https://techcrunch.com/2019/05/30/foursquare-buys-placed-from-snap-inc-on-the-heels-of-150m-in-new-funding Accessed 2026-10-24.

[31] Telenav, Inc. Form 10-K for the fiscal year ended June 30, 2020, Note 'Sale of Ads Business' (XBRL report R18). Telenav, Inc. (SEC EDGAR), 2020-08. https://www.sec.gov/Archives/edgar/data/1474439/000147443920000029/R18.htm Accessed 2026-10-24.

[32] Telenav, Inc. Form 10-K FY2020, Note 'Summary of business and significant accounting policies' (XBRL report R8). Telenav, Inc. (SEC EDGAR), 2020-08. https://www.sec.gov/Archives/edgar/data/1474439/000147443920000029/R8.htm Accessed 2026-10-24.

[33] Clearlake & TA Associates-Backed Precisely Signs a Definitive Agreement to Acquire PlaceIQ. Precisely (press release), 2022-01-19. https://www.precisely.com/press-release/clearlake-ta-associates-backed-precisely-signs-a-definitive-agreement-to-acquire-placeiq Accessed 2026-10-24.

[34] FAQ: General - PlaceIQ. Precisely, undated (served 2026-10-24). https://www.precisely.com/faq-general-placeiq Accessed 2026-10-24.

[35] Perion Acquires Hivestack, a Leading Global Full-Stack Digital Out-Of-Home (DOOH) Platform. Perion Network Ltd. (investor press release), 2023-12-12. https://perion.com/investors/press/perion-acquires-hivestack-a-leading-global-full-stack-digital-out-of-home-dooh-platform/ Accessed 2026-10-24.

[36] T-Mobile US, Inc. Form 10-K for the fiscal year ended December 31, 2025 (Item 7, Acquisitions: 'Acquisition of Vistar Media Inc.' and 'Acquisition of Blis Holdco Limited'). T-Mobile US, Inc. (SEC EDGAR), 2026. https://www.sec.gov/Archives/edgar/data/1283699/000128369926000010/tmus-20251231.htm Accessed 2026-10-24.

[37] Lamar Advertising Company, Form 8-K Exhibit 99.1: Fourth Quarter and Year Ended December 31, 2024 Operating Results ('Recent Developments'). Lamar Advertising Company (SEC EDGAR), 2025-02-20. https://www.sec.gov/Archives/edgar/data/1090425/000162828025006550/lamr-20241231xexx991.htm Accessed 2026-10-24.

[38] Location Data Firms Factual and Foursquare to Merge. Daily Research News Online (MrWeb), 2020-04-07. https://mrweb.com/drno/news29549.htm Accessed 2026-10-24.

[39] Factual and Foursquare announce merger (client news). Gunderson Dettmer LLP, 2020-04-06. https://www.gunder.com/print/v2/content/2757/factual-and-foursquare-announce-merger.pdf?lang=en Accessed 2026-10-24.

[40] NinthDecimal Joins InMarket to Create the Definitive Leader in Real-Time, Data-Driven Marketing. InMarket Media, LLC, 2020-09 (undated page; copyright 2025; dated by LOC-D-S22). https://go.inmarket.com/ninthdecimal-joins-inmarket Accessed 2026-10-24.

[41] InMarket buys NinthDecimal to compete with Foursquare more effectively. MarTech (Greg Sterling), 2020-09-10. https://martech.org/inmarket-buys-ninthdecimal-to-compete-with-foursquare-more-effectively/ Accessed 2026-10-24.

[42] Reston's X-Mode is rebranding as Outlogic and upping its data ethics following an acquisition by Atlanta's Digital Envoy. Technical.ly DC, 2021-08-05. https://technical.ly/dc/2021/08/05/x-mode-digital-envoy/ Accessed 2026-10-24.

[43] Advan Acquires SafeGraph's Patterns Business, Expanding Leadership in Location Intelligence and Foot Traffic Analytics. Advan Research Corporation (press release), 2022-11-03. https://advanresearch.com/insights/advan-acquires-safegraphs-patterns-business-expanding-leadership-in-location-intelligence-and-foot-traffic-analytics Accessed 2026-10-24.

[44] Unacast and Gravy Analytics: The Future of Location Insights. Unacast (Thomas Walle), 2023-11-29. https://unacast.com/post/unacast-and-gravy-analytics-merge Accessed 2026-10-24.

[45] Broadsign announces acquisition of Place Exchange (release as carried by Inavate APAC). Broadsign (via Inavate APAC), 2025-12-02. https://www.inavateapac.com/news/article/broadsign-announces-acquisition-of-place-exchange Accessed 2026-10-24.

[46] Broadsign Acquires Place Exchange. AV Network, 2025-11-25. https://www.avnetwork.com/news/broadsign-acquires-place-exchange Accessed 2026-10-24. Access: partial extract.

[47] ZeroToOne.AI Acquires GroundTruth to Deploy at Scale an AI Driven Predictive Intelligence Platform for Enterprise Decision Systems. ZeroToOne.AI / GroundTruth (press release on groundtruth.com), 2026-03-25. https://www.groundtruth.com/?p=464714 Accessed 2026-10-24.

[48] GroundTruth 'About' page. GroundTruth (a ZeroToOne.AI company), 2026 (copyright year; undated). https://www.groundtruth.com/about/ Accessed 2026-10-24.

[49] Nelson Mullins Advises ZeroToOne.AI in Acquisition of GroundTruth. Nelson Mullins Riley & Scarborough LLP, 2026-04-24. https://www.nelsonmullins.com/insights/news/press_releases/nelson-mullins-advises-zerotoone-ai-in-acquisition-of-groundtruth Accessed 2026-10-24.

[50] Infillion buys Foursquare to link ads with 16 billion visits. PPC Land, 2026-09-19. https://ppc.land/infillion-buys-foursquare-to-link-ads-with-16-billion-visits/ Accessed 2026-10-24.

[51] Infillion Acquires Foursquare, Adding More Location Data To Its Ever-Growing Ad Tech Stack. AdExchanger, 2026-09-18. https://adexchanger.com/marketers/infillion-acquires-foursquare-adding-more-location-data-to-its-ever-growing-ad-tech-stack/ Accessed 2026-10-24.

[52] Infillion to Acquire Foursquare, Closing the Loop Between Advertising and Real-World Results. Infillion (press release on its blog), 2026-09-18. https://infillion.com/blog/infillion-to-acquire-foursquare-closing-the-loop-between-advertising-and-real-world-results/ Accessed 2026-10-24.

[53] Exclusive: Infillion acquires location data company Foursquare. Axios, 2026-09-18. https://www.axios.com/2026/09/18/infillion-acquires-foursquare-location-data Accessed 2026-10-24.

[54] Near Intelligence, Inc. Form 8-K (date of earliest event reported December 8, 2023): Chapter 11 petitions, DIP financing, stalking-horse asset purchase agreement, Nasdaq delisting notice. Near Intelligence, Inc. (SEC EDGAR), 2023-12-11. https://www.sec.gov/Archives/edgar/data/1826671/000121390023094534/ea189803-8k_nearintell.htm Accessed 2026-10-24.

[55] Near Intelligence, Inc. Form 8-K (date of earliest event reported March 1, 2024): Item 2.01 Completion of Acquisition or Disposition of Assets. Near Intelligence, Inc. (SEC EDGAR), 2024-03-01. https://www.sec.gov/Archives/edgar/data/1826671/000121390024019010/ea0201070-8k_nearintelli.htm Accessed 2026-10-24.

[56] xAd Unveils Major Rebranding Effort as GroundTruth to Expand Beyond Media (press release as carried by Retail Dive). xAd, Inc. / GroundTruth (via Retail Dive press-release service), 2017-06-15. https://www.retaildive.com/press-release/20170615-xad-unveils-major-rebranding-effort-as-groundtruth-to-expand-beyond-media-1/ Accessed 2026-10-24.

[57] Data Cleanroom Spectus Brings Privacy, Speed and Affordability to Geospatial Data Market (release listing). DRW Venture Capital (investor copy of the Spectus release), 2022-02-23. https://drw.com/vc/updates/data-cleanroom-spectus-brings-privacy-speed-and-affordability-to-geospatial-data-market Accessed 2026-10-24. Access: access error; text or metadata via another route.

[58] Cuebiq transition page: Workbench clients migrate to Spectus (clara-beta.cuebiq.com). Cuebiq Inc., undated (served 2026-10-24). https://clara-beta.cuebiq.com/ Accessed 2026-10-24.

[59] Azira: A New Beginning (CEO blog, Gladys Kong). Azira LLC, 2024-03-22. https://azira.com/blogs/azira-a-new-beginning Accessed 2026-10-24. Access: partial extract.

[60] Location Intel Provider Cuebiq Is Shutting Down Its SDK In The Name Of Privacy. AdExchanger (Allison Schiff), 2021-02-23. https://www.adexchanger.com/mobile/location-intel-provider-cuebiq-is-shutting-down-its-sdk-in-the-name-of-privacy/ Accessed 2026-10-24.

[61] Apple, Google tell developers to remove location tracing tech. Tribune India (Tribune Web Desk, wire copy), reporting the Wall Street Journal, 2020-12-12. https://www.tribuneindia.com/news/science-technology/apple-google-tell-developers-to-remove-location-tracing-tech-183315 Accessed 2026-10-24.

[62] Google and Apple want to ban developers that use X-Mode software to track you. Android Central, reporting the Wall Street Journal. https://www.androidcentral.com/google-and-apple-want-ban-developers-use-x-mode-software-track-you Accessed 2026-10-24.

[63] Monthly Patterns (SafeGraph documentation, legacy product notice and panel carrier breakdown). SafeGraph Inc., 2020-07 (carrier table 'as of the July-2020 release'); page updated 2025-04-29. https://docs.safegraph.com/docs/monthly-patterns Accessed 2026-10-24.

[64] City Guide Sunset. Foursquare Labs, Inc., 2025 (page undated; states events of 2024-12-15 and 2025-04-28). https://foursquare.com/city-guide-sunset/ Accessed 2026-10-24.

[65] Apple previews iOS 13. Apple Inc. (Newsroom), 2019-06-03. https://www.apple.com/newsroom/2019/06/apple-previews-ios-13/ Accessed 2026-10-24.

[66] About iOS 13 updates. Apple Inc. (Support), 2024-02-26. https://support.apple.com/en-us/HT210393 Accessed 2026-10-24.

[67] About the security content of iOS 13. Apple Inc. (Support), 2023-11-06. https://support.apple.com/en-us/HT210606 Accessed 2026-10-24.

[68] Apple reimagines the iPhone experience with iOS 14. Apple Inc. (Newsroom), 2020-06-22. https://www.apple.com/newsroom/2020/06/apple-reimagines-the-iphone-experience-with-ios-14/ Accessed 2026-10-24.

[69] iOS 14 is available today. Apple Inc. (Newsroom), 2020-09-16. https://www.apple.com/newsroom/2020/09/ios-14-is-available-today/ Accessed 2026-10-24.

[70] What's new in location (WWDC20 session 10660). Apple Inc. (Developer), 2020-06. https://developer.apple.com/videos/play/wwdc2020/10660/ Accessed 2026-10-24.

[71] iOS 15 is available today. Apple Inc. (Newsroom), 2021-09-20. https://www.apple.com/newsroom/2021/09/ios-15-is-available-today/ Accessed 2026-10-24.

[72] iCloud Private Relay Overview. Apple Inc., 2021-12. https://www.apple.com/privacy/docs/iCloud_Private_Relay_Overview_Dec2021.PDF Accessed 2026-10-24.

[73] What's new in location authorization (WWDC24 session 10212). Apple Inc. (Developer), 2024-06. https://developer.apple.com/videos/play/wwdc2024/10212/ Accessed 2026-10-24.

[74] Apple security releases. Apple Inc. (Support), 2026. https://support.apple.com/en-us/100100 Accessed 2026-10-24.

[75] Core Location updates. Apple Inc. (Developer documentation). https://developer.apple.com/documentation/updates/corelocation Accessed 2026-10-24.

[76] Details for app privacy questions now available. Apple Inc. (Developer News), 2020-09-03. https://developer.apple.com/news/?id=hx9s63c5 Accessed 2026-10-24.

[77] Upcoming AppTrackingTransparency requirements. Apple Inc. (Developer News), 2021-04-20. https://developer.apple.com/news/?id=ecvrtzt2 Accessed 2026-10-24.

[78] iOS 14.5 delivers Unlock iPhone with Apple Watch, more diverse Siri voice options, and new privacy controls. Apple Inc. (Newsroom), 2021-04-26. https://www.apple.com/newsroom/2021/04/ios-14-5-offers-unlock-iphone-with-apple-watch-diverse-siri-voices-and-more/ Accessed 2026-10-24.

[79] Privacy updates for App Store submissions. Apple Inc. (Developer News), 2024-02-29. https://developer.apple.com/news/?id=3d8a9yyh Accessed 2026-10-24.

[80] Describing use of required reason API (developer documentation). Apple Inc. (Developer Documentation), 2026. https://developer.apple.com/documentation/bundleresources/describing-use-of-required-reason-api Accessed 2026-10-24. Access: partial extract.

[81] NSPrivacyAccessedAPIType (required-reason API categories). Apple Inc. (Developer documentation). https://developer.apple.com/documentation/bundleresources/app-privacy-configuration/nsprivacyaccessedapitypes/nsprivacyaccessedapitype Accessed 2026-10-24.

[82] Welcoming Android 10!. Google (Android Developers Blog; Stephanie Cuthbertson), 2019-09-03. https://android-developers.googleblog.com/2019/09/welcoming-android-10.html Accessed 2026-10-24.

[83] Privacy changes in Android 10. Google (Android Developers), 2026-10-16. https://developer.android.com/about/versions/10/privacy/changes Accessed 2026-10-24.

[84] Turning it up to 11: Android 11 for developers. Google (Android Developers Blog; Stephanie Cuthbertson), 2020-09-08. https://android-developers.googleblog.com/2020/09/android11-final-release.html Accessed 2026-10-24.

[85] Location updates in Android 11. Google (Android Developers), 2026. https://developer.android.com/about/versions/11/privacy/location Accessed 2026-10-24.

[86] Android 12 is live in AOSP!. Google (Android Developers Blog; Dave Burke), 2021-10-04. https://android-developers.googleblog.com/2021/10/android-12-is-live-in-aosp.html Accessed 2026-10-24.

[87] Request location permissions (runtime). Google (Android Developers), 2026. https://developer.android.com/develop/sensors-and-location/location/permissions/runtime Accessed 2026-10-24.

[88] Safer and More Transparent Access to User Location. Google (Android Developers Blog; Krish Vitaldevara, Google Play Trust & Safety), 2020-02-19. https://android-developers.googleblog.com/2020/02/safer-location-access.html Accessed 2026-10-24.

[89] Get more information about your apps in Google Play. Google (The Keyword; Suzanne Frey), 2022-04-26. https://blog.google/products/google-play/data-safety/ Accessed 2026-10-24.

[90] Provide information for Google Play's Data safety section (Play Console Help). Google Play, 2026. https://support.google.com/googleplay/android-developer/answer/10787469 Accessed 2026-10-24.

[91] Update on Plans for Privacy Sandbox Technologies. Google (Privacy Sandbox blog; Anthony Chavez, VP), 2025-10-17. https://privacysandbox.google.com/blog/update-on-plans-for-privacy-sandbox-technologies Accessed 2026-10-24.

[92] Privacy Sandbox feature status. Google (Privacy Sandbox), 2026-08-14. https://privacysandbox.google.com/overview/status Accessed 2026-10-24.

[93] Preview: Permissions and APIs that Access Sensitive Information (location button). Google Play, 2026-04. https://support.google.com/googleplay/android-developer/answer/16909972 Accessed 2026-10-24.

[94] Permissions and APIs that Access Sensitive Information (Play policy). Google Play, 2026. https://support.google.com/googleplay/android-developer/answer/9888170 Accessed 2026-10-24.

[95] Understanding the bias of mobile location data across spatial scales and over time: A comprehensive analysis of SafeGraph data in the United States (PLOS ONE 19(1): e0294430). Zhenlong Li, Huan Ning, Fengrui Jing, M. Naser Lessani, 2024-01-19. https://pmc.ncbi.nlm.nih.gov/articles/PMC10798630/ Accessed 2026-10-24.

[96] Multiscale dynamic human mobility flow dataset in the U.S. during the COVID-19 epidemic (Scientific Data 7, 390). Yuhao Kang, Song Gao, Yunlei Liang, Mingxiao Li, Jinmeng Rao, Jake Kruse (University of Wisconsin-Madison), 2020-11-12. https://pmc.ncbi.nlm.nih.gov/articles/PMC7661515/ Accessed 2026-10-24.

[97] Using mobile device data to track the effects of the COVID-19 pandemic on spatiotemporal patterns of national park visitation (Sustainability 13(16), 9366; doi 10.3390/su13169366). John A. Kupfer, Zhenlong Li, Huan Ning, Xiao Huang, 2021-08-02. https://pure.psu.edu/en/publications/using-mobile-device-data-to-track-the-effects-of-the-covid-19-pan/ Accessed 2026-10-24. Access: abstract only.

[98] Assessing the validity of mobile device data for estimating visitor demographics and visitation patterns in Yellowstone National Park (Journal of Environmental Management 317, 115410). Yun Liang, Junjun Yin, Bing Pan, Michael S. Lin, Lauren Miller, B. Derrick Taff, Guangqing Chi (Pennsylvania State University; Hong Kong Polytechnic University; Yellowstone National Park), 2022-06-10. https://par.nsf.gov/servlets/purl/10451340 Accessed 2026-10-24.

[99] Leveraging Administrative Data for Bias Audits: Assessing Disparate Coverage with Mobility Data for COVID-19 Policy (arXiv 2011.07194v2; Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pp. 173-184). Amanda Coston, Neel Guha, Derek Ouyang, Lisa Lu, Alexandra Chouldechova, Daniel E. Ho (Carnegie Mellon University; Stanford University), 2021-04-16. https://arxiv.org/abs/2011.07194 Accessed 2026-10-24.

[100] Exploring Economic Sectoral Dynamics Through High-resolution Mobility Data (arXiv 2506.13985v2). Timothy F. Leslie (George Mason University), Hossein Amiri, Andreas Zuefle (Emory University), 2025-08-25. https://arxiv.org/html/2506.13985v2 Accessed 2026-10-24.

[101] Mobile Analytics vs. Traditional Surveys: A case study exploring visitation patterns and visitor demographics at an outdoor recreation destination (slide report). Ami Choi, University of Minnesota Extension Tourism Center, 2024-04. https://conservancy.umn.edu/server/api/core/bitstreams/fdf3b6a9-7824-4e64-b330-f6392f4491b4/content Accessed 2026-10-24.

[102] Visitation data accuracy (help centre article). Placer.ai (Placer Labs Inc.), unknown. https://connect.placer.ai/use-cases/visitation-data-accuracy Accessed 2026-10-24.

[103] Sampling Bias in Cuebiq Dataset in MSAs (blog post, CIS Big Data for Mobility project). University of Washington bigdata4mobility project (author listed as the project), 2022-05-30. https://sites.uw.edu/bigdata4mobility/2022/05/30/sampling-bias-in-cuebiq-dataset-in-msas Accessed 2026-10-24.

[104] The Promise and Limitations of Locational App Data for Origin-Destination Analysis: A Case Study (FHWA-HEP-20-022). Thomas Adler, Vince Bernardin, J. Dumont, Leah Flake, Hadi Sadrsadat (RSG, Inc.) for the US Federal Highway Administration, 2017-10-01. https://rosap.ntl.bts.gov/view/dot/55798 Accessed 2026-10-24. Access: abstract only.

[105] One country, multiple portraits: representativeness in GPS-based mobility data is source-specific and spatially dependent (arXiv 2606.23616v1). Carmen Cabrera, Francisco Rowe, Miguel Gonzalez-Leonardo, Juan Ignacio Vilchis-Garcia, Elisa Omodei, Maribel Hernandez-Rosales, 2026-06-22. https://arxiv.org/html/2606.23616v1 Accessed 2026-10-24.

[106] FAQ | Unacast Documentation. Unacast (docs.unacast.com), undated (footer "Copyright © 2026 Unacast"). https://docs.unacast.com/datasets/faq/ Accessed 2026-10-24.

[107] What Do Measures of Real-time Corporate Sales Tell Us about Earnings Surprises and Post-Announcement Returns? (NBER Working Paper 22366). Kenneth A. Froot, Namho Kang, Gideon Ozik, Ronnie Sadka, 2016-06. https://www.nber.org/papers/w22366 Accessed 2026-10-24. Access: abstract only.

[108] Data Bias in Human Mobility is a Universal Phenomenon but is Highly Location-specific (arXiv 2508.00149v1). Katinka den Nijs, Elisa Omodei, Vedran Sekara, 2025-07-31. https://arxiv.org/abs/2508.00149 Accessed 2026-10-24. Access: abstract only.

[109] A systematic machine learning approach to measure and assess biases in mobile phone population data (arXiv 2509.02603v1). Carmen Cabrera, Francisco Rowe, 2025-08-29. https://arxiv.org/abs/2509.02603 Accessed 2026-10-24. Access: abstract only.

[110] App Tracking (working-paper abstract hosted by Bocconi University marketing department). Authors not shown on the document (a Lehigh University seminar listing found in search attributes a talk of this title to Michelle Andrews, Emory University; not verified), 2022-04-19. https://marketing.unibocconi.eu/sites/default/files/files/media/attachments/apptracking20220419123109.pdf Accessed 2026-10-24. Access: abstract only.

[111] A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook (working paper 12 April 2018; published in Marketing Science 38(2), 2019). Brett R. Gordon, Florian Zettelmeyer (Kellogg School of Management, Northwestern University), Neha Bhargava, Dan Chapsky (Facebook), 2018-04-12. https://www.kellogg.northwestern.edu/faculty/gordon_b/files/fb_comparison.pdf Accessed 2026-10-24.

[112] Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement (arXiv 2201.07055v2; published in Marketing Science 42(4), 768-793, 2023). Brett R. Gordon, Robert Moakler, Florian Zettelmeyer, 2022-10-04. https://arxiv.org/abs/2201.07055v2 Accessed 2026-10-24. Access: abstract only.

[113] The Unfavorable Economics of Measuring the Returns to Advertising (Quarterly Journal of Economics 130(4), 1941-1973). Randall A. Lewis, Justin M. Rao, 2015-07-06. https://gwern.net/doc/economics/advertising/2015-lewis.pdf Accessed 2026-10-24.

[114] Online ads and offline sales: measuring the effect of retail advertising via a controlled experiment on Yahoo! (Quantitative Marketing and Economics 12(3), 235-266). Randall A. Lewis, David H. Reiley, 2014-09. https://link.springer.com/article/10.1007/s11129-014-9146-6 Accessed 2026-10-24. Access: abstract only.

[115] Ghost Ads: Improving the Economics of Measuring Ad Effectiveness (NBER conference working paper, 18 February 2016; published in Journal of Marketing Research 54(6), 2017). Garrett A. Johnson (University of Rochester), Randall A. Lewis (Netflix), Elmar I. Nubbemeyer (Google), 2016-02-18. https://conference.nber.org/conf_papers/f84173.pdf Accessed 2026-10-24.

[116] How accurate is footfall data?. Joe Capocci, Huq Industries, 2026-08-06. https://huq.io/blog/how-accurate-is-footfall-data Accessed 2026-10-24.

[117] Advan Research foot traffic covers 2,000 tickers of which 1,365 publicly traded (press release). Advan Research Corporation, 2019-06-11. https://advanresearch.com/advan-news/advan-research-foot-traffic-covers-2-000-tickers-of-which-1-365-publicly-traded Accessed 2026-10-24.

[118] Cuebiq Footfall Attribution Benchmarks: The Advent of Accurate Location-based Metrics for Today's Marketer. Cuebiq (PDF hosted by IAB), 2017-06 (upload path). https://www.iab.com/wp-content/uploads/2017/06/Cuebiq_Footfall-Attribution_Benchmarks_.pdf Accessed 2026-10-24.

[119] What is Attribution? (Foursquare Attribution documentation). Foursquare, 2026-02-12. https://docs.foursquare.com/campaign-products/docs/what-is-attribution Accessed 2026-10-24.

[120] Foursquare's Attribution Methodology Takes Your Media Dollars Further. Foursquare, 2022-06-06. https://foursquare.com/resources/blog/ad-tech/location-foursquares-attribution-methodology-takes-your-media-dollars-further/ Accessed 2026-10-24.

[121] Foot Traffic Measurement 101: Connecting Ad Spend to Store Visits (blog). inMarket Media, LLC, 2026-07-29. https://inmarket.com/blog-foot-traffic-measurement-101-connecting-ad-spend-to-store-visits/ Accessed 2026-10-24.

[122] Ads Manager Solutions Platform | GroundTruth. GroundTruth (xAd, Inc.), 2026. https://www.groundtruth.com/solutions/ads-manager/ Accessed 2026-10-24.

[123] Blis introduces Smart Holdout Groups, a market-leading omnichannel measurement offering (release). Blis, 2024-07-12 (page date; dateline 'New York, NY – July 11, 2024'). https://blis.com/blis-introduces-smart-holdout-groups-a-market-leading-omnichannel-measurement-offering/ Accessed 2026-10-24.

[124] Blis rolls out omnichannel measurement solution into broader market (release). Blis, 2024-10-09. https://blis.com/blis-rolls-out-omnichannel-measurement-solution-into-broader-market/ Accessed 2026-10-24.

[125] Unacast homepage and product navigation. Unacast, unknown. https://www.unacast.com/ Accessed 2026-10-24.

[126] Decision and Order, In the Matter of X-Mode Social, Inc. and Outlogic, LLC, Docket No. C-4802. Federal Trade Commission, 2024-04-11. https://www.ftc.gov/system/files/ftc_gov/pdf/X-ModeSocialDecisionandOrder.pdf Accessed 2026-10-24.

[127] FTC Finalizes Order with X-Mode and Successor Outlogic Prohibiting it from Sharing or Selling Sensitive Location Data. Federal Trade Commission, 2024-04-12. https://www.ftc.gov/news-events/news/press-releases/2024/04/ftc-finalizes-order-x-mode-successor-outlogic-prohibiting-it-sharing-or-selling-sensitive-location Accessed 2026-10-24.

[128] Outlogic home page ('World Leader of Privacy-centric Geolocation Data'). Outlogic, LLC, 2026. https://outlogic.io/ Accessed 2026-10-24.

[129] Outlogic Privacy Notice (Last Update: January 2026). Outlogic, LLC, 2026-01. https://outlogic.io/privacy-policy/ Accessed 2026-10-24.

[130] Decision and Order, In the Matter of InMarket Media, LLC, Docket No. C-4803. Federal Trade Commission, 2024-04-29. https://www.ftc.gov/system/files/ftc_gov/pdf/InMarketMedia-DecisionandOrder.pdf Accessed 2026-10-24.

[131] In the Matter of InMarket Media, LLC (case page, FTC Matter 2023088). Federal Trade Commission, 2024-05-01. https://www.ftc.gov/legal-library/browse/cases-proceedings/2023088-inmarket-media-llc Accessed 2026-10-24.

[132] FTC Finalizes Order with InMarket Prohibiting It from Selling or Sharing Precise Location Data (press release). US Federal Trade Commission, 2024-05-01. https://www.ftc.gov/news-events/news/press-releases/2024/05/ftc-finalizes-order-inmarket-prohibiting-it-selling-or-sharing-precise-location-data Accessed 2026-10-24.

[133] InMarket | Real-Time Marketing and Measurement (homepage). inMarket Media, LLC, 2026. https://inmarket.com/ Accessed 2026-10-24.

[134] InMarket Privacy Policy (Last updated: June 30, 2026). inMarket Media, LLC, 2026-06-30. https://inmarket.com/privacy/ Accessed 2026-10-24.

[135] Final consent package (Complaint; Decision and Order), In the Matter of Gravy Analytics, Inc. and Venntel, Inc., Docket No. C-4810. Federal Trade Commission, 2025-01-13. https://www.ftc.gov/system/files/ftc_gov/pdf/212_3035_-_gravy_analytics_final_consent_package_without_signatures.pdf Accessed 2026-10-24.

[136] FTC Finalizes Order Prohibiting Gravy Analytics, Venntel from Selling Sensitive Location Data. Federal Trade Commission, 2025-01-14. https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-finalizes-order-prohibiting-gravy-analytics-venntel-selling-sensitive-location-data Accessed 2026-10-24.

[137] Unacast Privacy Policy and Retention Schedule (Last Updated: September 2026). Gravy Analytics, Inc. dba Unacast, 2026-09. https://www.unacast.com/privacy-policy Accessed 2026-10-24.

[138] Venntel home page. Venntel, Inc., 2026. https://venntel.com/ Accessed 2026-10-24.

[139] Decision and Order, In the Matter of Mobilewalla, Inc., Docket No. C-4811. Federal Trade Commission, 2025-01-13. https://www.ftc.gov/system/files/ftc_gov/pdf/Mobilewalla-D&O.pdf Accessed 2026-10-24.

[140] FTC Finalizes Order Banning Mobilewalla from Selling Sensitive Location Data (press release). US Federal Trade Commission, 2025-01-14. https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-finalizes-order-banning-mobilewalla-selling-sensitive-location-data Accessed 2026-10-24.

[141] Vertical AI - Consumer Intelligence Platform | Mobilewalla (homepage). Mobilewalla, Inc., 2025. https://www.mobilewalla.com/ Accessed 2026-10-24.

[142] Mobilewalla Privacy Center: Business Services Privacy Notice (Effective June 30, 2023, updated January 9, 2026). Mobilewalla, Inc., 2026-01-09. https://www.mobilewalla.com/business-services-privacy-notice-2023 Accessed 2026-10-24.

[143] Stipulated Order for Injunction and Other Relief, FTC v. Kochava Inc. and Collective Data Solutions, LLC, Case No. 2:22-cv-00377-BLW, Dkt. 138. US District Court for the District of Idaho (Judge B. Lynn Winmill), 2026-06-25. https://www.ftc.gov/system/files/ftc_gov/pdf/Kochava-Order.pdf Accessed 2026-10-24.

[144] FTC to Ban Kochava and Subsidiary from Selling Sensitive Location Data to Settle Charges They Sold Location Data Linked to Millions of Mobile Devices. Federal Trade Commission (press release), 2026-05-04. https://www.ftc.gov/news-events/news/press-releases/2026/05/ftc-ban-kochava-subsidiary-selling-sensitive-location-data-settle-charges-they-sold-location-data Accessed 2026-10-24.

[145] The Collective Data Solutions home page ('Powering Your Growth With Insights & Innovation'). The Collective Data Solutions, LLC, 2026. https://collectivedata.io/ Accessed 2026-10-24.

[146] Privacy Block (The Collective Data Solutions) and Kochava Privacy Block page. The Collective Data Solutions, LLC; Kochava Inc., 2026. https://collectivedata.io/privacy-block/ Accessed 2026-10-24.

[147] Kochava /collective/ URL (serves 'App Measurement - Mobile App Analytics Tools, Data Management | Kochava'). Kochava Inc., 2026. https://www.kochava.com/collective/ Accessed 2026-10-24.

[148] Memorandum Decision and Order (motion to dismiss Second Amended Complaint denied), FTC v. Kochava Inc., Dkt. 101. US District Court for the District of Idaho, 2025-02-03. https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-3.pdf Accessed 2026-10-24.

[149] FTC Finalizes Order Settling Allegations that GM and OnStar Collected and Sold Geolocation Data Without Consumers' Informed Consent. Federal Trade Commission, 2026-01-14. https://www.ftc.gov/news-events/news/press-releases/2026/01/ftc-finalizes-order-settling-allegations-gm-onstar-collected-sold-geolocation-data-without-consumers Accessed 2026-10-24.

[150] In the Matter of General Motors LLC, General Motors Holdings LLC and OnStar, LLC: Decision and Order (Docket C-4828). Federal Trade Commission, 2026-01-14. https://www.ftc.gov/system/files/ftc_gov/pdf/2423052c4828gmlexisfinalorder.pdf Accessed 2026-10-24.

[151] Chapter 191, Laws of 2023 (Engrossed Substitute House Bill 1155), Consumer Health Data (My Health My Data Act), Washington. Washington State Legislature, 2023-04-27. https://lawfilesext.leg.wa.gov/biennium/2023-24/Pdf/Bills/Session%20Laws/House/1155-S.SL.pdf Accessed 2026-10-24.

[152] Chapter 455 (Senate Bill 541), Maryland Online Data Privacy Act of 2024, 2024 Laws of Maryland. Maryland General Assembly, 2024-05-09. https://mgaleg.maryland.gov/2024RS/Chapters_noln/CH_455_sb0541e.pdf Accessed 2026-10-24.

[153] Oregon Laws 2025, Chapter 251 (HB 2008), Relating to consumer protection; amending ORS 646A.578. Oregon Legislative Assembly, 2025-06-05. https://www.oregonlegislature.gov/bills_laws/lawsstatutes/2025orlaw0251.pdf Accessed 2026-10-24.

[154] Enrolled House Bill 2008 (HB 2008-B), 83rd Oregon Legislative Assembly, 2025 Regular Session. Oregon Legislative Assembly (OLIS), 2025-05-27. https://olis.oregonlegislature.gov/liz/2025R1/Downloads/MeasureDocument/HB2008/Enrolled Accessed 2026-10-24.

[155] Public Act No. 25-113 (Substitute Senate Bill No. 1295), An Act Concerning Broadband Internet, Gaming, Social Media, Online Services and Consumer Contracts, Connecticut. Connecticut General Assembly, 2025. https://www.cga.ct.gov/2025/ACT/PA/PDF/2025PA-00113-R00SB-01295-PA.PDF Accessed 2026-10-24.

[156] P.L.2026, c.25 (Assembly Bill 5328, first reprint), An Act concerning personal data, data brokers, data collectors, and amending P.L.2023, c.266, New Jersey. New Jersey Legislature, 2026-06-30. https://pub.njleg.state.nj.us/Bills/2026/PL26/25_.HTM Accessed 2026-10-24.

[157] Code of Virginia 59.1-578, Data controller responsibilities; transparency. Virginia Law Library (law.lis.virginia.gov), 2026-10-24. https://law.lis.virginia.gov/vacode/title59.1/chapter53/section59.1-578/ Accessed 2026-10-24.

[158] Virginia Becomes Third State to Ban the Sale of Consumers' Geolocation Data. Willkie Farr & Gallagher (Compliance Concourse), 2026-04-15. https://complianceconcourse.willkie.com/articles/virginia-becomes-third-state-to-ban-the-sale-of-consumers-geolocation-data/ Accessed 2026-10-24.

[159] Public Act No. 26-64 (Substitute Senate Bill No. 4), An Act Concerning Consumer Privacy and Protection. Connecticut General Assembly, 2026-05-27. https://www.cga.ct.gov/2026/ACT/PA/PDF/2026PA-00064-R00SB-00004-PA.PDF Accessed 2026-10-24.

[160] SB 4 (2026) bill status page. Connecticut General Assembly, 2026. https://www.cga.ct.gov/asp/cgabillstatus/cgabillstatus.asp?selBillType=Bill&which_year=2026&bill_num=4 Accessed 2026-10-24.

[161] California Consumer Privacy Act of 2018 (Civil Code 1798.100 et seq.) as amended, text as posted by the California Privacy Protection Agency. California Privacy Protection Agency (statute text), 2024. https://cppa.ca.gov/regulations/pdf/cppa_act.pdf Accessed 2026-10-24.

[162] Code of Virginia 59.1-575, Definitions (Consumer Data Protection Act). Virginia Law Library (law.lis.virginia.gov), 2025. https://law.lis.virginia.gov/vacode/title59.1/chapter53/section59.1-575/ Accessed 2026-10-24.

[163] California Data Broker Registry (page). California Privacy Protection Agency (CalPrivacy), 2026-07-29. https://cppa.ca.gov/data_broker_registry/ Accessed 2026-10-24.

[164] MRC presentation to the ACA Executive Forums (26-27 September 2018). Media Rating Council, 2018-09. https://acaweb.ca/en/wp-content/uploads/sites/2/2018/10/ACA-Executive-Forums-MRC.pdf Accessed 2026-10-24.

[165] Media Rating Council Accreditation Updates, January 2020 through March 2020. Media Rating Council, 2020-04-13. https://www.mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/041320%20MRC%20Accreditation%20Status%20Communication.pdf Accessed 2026-10-24.

[166] Media Rating Council Accreditation Updates, July 2020 through September 2020. Media Rating Council, 2020-10-13. https://www.mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/101320%20MRC%20Accreditation%20Status%20Communication.pdf Accessed 2026-10-24.

[167] Media Rating Council Accreditation Updates, October 2022 through December 2022. Media Rating Council, 2023-01-24. https://www.mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/012423%20MRC%20Accreditation%20Status%20Communication.pdf Accessed 2026-10-24.

[168] Media Rating Council Accreditation Updates, July 2021 through September 2021. Media Rating Council, 2021-10-20. https://www.mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/102021%20MRC%20Accreditation%20Status%20Communication.pdf Accessed 2026-10-24.

[169] Media Rating Council Accreditation Updates, July 2023 through September 2023. Media Rating Council, 2023-10-16. https://www.mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/101623_MRC_Accreditation_Status_Communication.pdf Accessed 2026-10-24.

[170] Media Rating Council Accreditation Updates, April 2024 through June 2024. Media Rating Council, 2024-07-26. https://www.mediaratingcouncil.org/sites/default/files/News/MRC%20Accreditation%20Status%20Quarterly%20Update_Q2%2024%207.26.pdf Accessed 2026-10-24.

[171] Accredited Services: Digital. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/digital Accessed 2026-10-24.

[172] Accredited Services: Multimedia. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/multimedia Accessed 2026-10-24.

[173] Accredited Services: Out-of-Home. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/out-of-home Accessed 2026-10-24.

[174] Accredited Services: Television. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/television Accessed 2026-10-24.

[175] Accredited Services: Radio. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/radio Accessed 2026-10-24.

[176] Accredited Services: Outside the U.S.. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/outside-US Accessed 2026-10-24.

[177] Services Under Review: Digital. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/services-under-review/digital Accessed 2026-10-24.

[178] Services Under Review: Out-of-Home. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/services-under-review/out-of-home Accessed 2026-10-24.

[179] Services Under Review: Television. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/services-under-review/television Accessed 2026-10-24.

[180] Services Under Review: Radio. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/services-under-review/radio Accessed 2026-10-24.

[181] Services Under Review: Services Outside the U.S.. Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation/services-under-review/services-outside-US Accessed 2026-10-24.

[182] Geopath High-Level Media Measurement Process. Geopath, 2023-05-31. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323447855/Geopath+High-Level+Media+Measurement+Process Accessed 2026-10-24.

[183] Geopath Mobile Device Panel. Geopath, 2023-05-31. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449407 Accessed 2026-10-24.

[184] Reach & Frequency: What Has Changed and Why it Matters. Geopath (Brian Schopper), 2022-11-15. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2128379915 Accessed 2026-10-24.

[185] Geopath Mobile Device Location & Movement Patterns (Confluence). Geopath, 2023-05-31. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449264/Geopath+Mobile+Device+Location+Movement+Patterns Accessed 2026-10-24. Access: partial extract.

[186] 2026 Planning Dataset to Become Default on November 12. Geopath, 2026-10-19. https://blog.geopath.org/2026/10/19/2026-planning-dataset-to-become-default-on-november-12/ Accessed 2026-10-24. Access: partial extract.

[187] Vistar Media Adds Foursquare as Latest Digital Out-of-Home Impact Measurement Partner. Vistar Media, 2020-02-11. https://www.vistarmedia.com/news/foursquare-dooh-measurement-partner Accessed 2026-10-24.

[188] Vistar Media announces new partnership with Adsquare to drive innovation in programmatic OOH. ExchangeWire (release-based report), 2023-06-07. https://www.exchangewire.com/blog/2023/06/07/vistar-media-announces-new-partnership-with-adsquare-to-drive-innovation-in-programmatic-ooh/ Accessed 2026-10-24.

[189] Partner Directory. Vistar Media, undated. https://www.vistarmedia.com/partner-directory Accessed 2026-10-24.

[190] Vistar Media homepage. Vistar Media, undated. https://www.vistarmedia.com/ Accessed 2026-10-24.

[191] Partner: GroundTruth (Vistar Media partner page). Vistar Media, undated. https://www.vistarmedia.com/partner-groundtruth Accessed 2026-10-24.

[192] PX General Sales Presentation 2026 (PDF). Place Exchange (part of Broadsign), 2026. https://www.placeexchange.com/assets/marketing/PX_General_Sales_Presentation_2026.pdf Accessed 2026-10-24.

[193] PerView Campaign Reach and Frequency Measurement. Place Exchange, undated. https://www.placeexchange.com/perview Accessed 2026-10-24. Access: partial extract.

[194] Measurement & Attribution. Broadsign, undated. https://broadsign.com/measurement-attribution/ Accessed 2026-10-24.

[195] DOOH (channel page). Perion, undated. https://perion.com/channels/dooh/ Accessed 2026-10-24.

[196] Footfall analysis: Zoom Media x Gillette Lift Study (Hivestack; 'KCG x Zoom Footfall Study'). Hivestack (hosted as IAB UK member content), 2023-08. https://www.iabuk.com/sites/default/files/member-content/2023-08/KCG%20x%20Zoom%20Footfall%20Study.pdf Accessed 2026-10-24.

[197] AdCP targeting overlay schema (core/targeting.json) at tag v3.2.6. AgenticAdvertising.org / adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/static/schemas/source/core/targeting.json Accessed 2026-10-24.

[198] Prebid Sales Agent v2.0.0: src/adapters/gam/managers/targeting.py. Prebid.org (prebid/salesagent, GitHub), 2026-07-01. https://raw.githubusercontent.com/prebid/salesagent/v2.0.0/src/adapters/gam/managers/targeting.py Accessed 2026-10-24.

[199] Prebid Sales Agent v2.0.0 pyproject.toml (adcp SDK pin). Prebid.org (prebid/salesagent), 2026-07-01. https://raw.githubusercontent.com/prebid/salesagent/v2.0.0/pyproject.toml Accessed 2026-10-24.

[200] AAO agent registry: agent list and per-agent compliance records. AgenticAdvertising.org (AAO), 2026-10-24. https://agenticadvertising.org/api/registry/agents?limit=500 Accessed 2026-10-24.

[201] PubMatic and Butler/Till Launch AdCP-Enabled Agentic AI Campaign Across CTV (Business Wire release, syndicated copy). PubMatic and Butler/Till (Business Wire), syndicated by FinancialContent, 2025-12-15. https://markets.financialcontent.com/dowtheoryletters/article/bizwire-2025-12-15-pubmatic-and-butlertill-launch-adcp-enabled-agentic-ai-campaign-across-ctv Accessed 2026-10-24. Access: access error; text or metadata via another route.

[202] AdCP geographic place schemas (core/geo-place-area.json, geo-place-system.json, geo-place-type.json) at tag v3.2.6. AgenticAdvertising.org / adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/static/schemas/source/core/geo-place-area.json Accessed 2026-10-24.

[203] PR #6093 feat(media-buy): add identifier-based place targeting. adcontextprotocol/adcp (GitHub), 2026-08-15. https://github.com/adcontextprotocol/adcp/pull/6093 Accessed 2026-10-24.

[204] AdCP release v3.2.0-beta.0 and CHANGELOG section 3.2.0-beta.0. adcontextprotocol (GitHub), 2026-08-17. https://github.com/adcontextprotocol/adcp/releases/tag/v3.2.0-beta.0 Accessed 2026-10-24.

[205] (Author's work) Issue #5969 RFC: geo_cells / geo_cells_exclude targeting - hierarchical cell-grid systems as a declared geographic unit. EvgenyAndroid (the paper's author) on adcontextprotocol/adcp; triage comments by bokelley; comments by nwoodman-spec, rickcentralcontrolcom, newbiechris741, 2026-07-21. https://github.com/adcontextprotocol/adcp/issues/5969 Accessed 2026-10-24.

[206] (Author's work) Issues #6013 (RFC: Cell Interoperability and Execution Conversion) and #6014 (RFC: Cell-grain delivery reporting), companions to #5969. EvgenyAndroid on adcontextprotocol/adcp, 2026-07-24. https://github.com/adcontextprotocol/adcp/issues/6014 Accessed 2026-10-24.

[207] AdCP targeting overlay schema at tag v3.3.0-beta.0. AgenticAdvertising.org / adcontextprotocol (GitHub), 2026-10-22. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.3.0-beta.0/static/schemas/source/core/targeting.json Accessed 2026-10-24.

[208] AdCP placement schema (core/placement.json) and identifier-types enum at tag v3.2.6. AgenticAdvertising.org / adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/static/schemas/source/core/placement.json Accessed 2026-10-24.

[209] AdCP compliance specialism sales_dooh (index.yaml v1.0.0) at tag v3.2.6. adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/static/compliance/source/specialisms/sales-dooh/index.yaml Accessed 2026-10-24.

[210] Broadsign's sell-side AI agent joins with Global Netherlands and Draft Digital to deliver first-ever end-to-end agentic AI-powered OOH campaign. Broadsign, 2026-05-27. https://broadsign.com/blog/broadsigns-sell-side-ai-agent-joins-with-global-netherlands-and-draft-digital-to-deliver-first-ever-end-to-end-agentic-ai-powered-ooh-campaign/ Accessed 2026-10-24. Access: partial extract.

[211] AdCP delivery metrics schema (core/delivery-metrics.json) at tag v3.2.6. AgenticAdvertising.org / adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/static/schemas/source/core/delivery-metrics.json Accessed 2026-10-24.

[212] AdCP attribution-methodology enum and outcome-measurement schema at tag v3.2.6. AgenticAdvertising.org / adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/static/schemas/source/enums/attribution-methodology.json Accessed 2026-10-24.

[213] OpenProposal 3.0 (AAMP 3.0) Version 3.0 draft 1, spec/openproposal-3.0.md. IAB Tech Lab (IABTechLab/OpenProposal), 2026-09-22. https://github.com/IABTechLab/OpenProposal/blob/main/spec/openproposal-3.0.md Accessed 2026-10-24.

[214] iab-agentic-primitives 0.5.1 (pre-release): Package.json, QuoteRequest.json, primitives/inventory.py. IAB Tech Lab (IABTechLab/iab-agentic-primitives), 2026-08-10. https://github.com/IABTechLab/iab-agentic-primitives Accessed 2026-10-24.

[215] IAB Tech Lab seller-agent 2.4.2 (AAMP reference seller agent): media_kit model, GAM model, adapters. IAB Tech Lab (IABTechLab/seller-agent), 2026-08-17. https://github.com/IABTechLab/seller-agent Accessed 2026-10-24.

[216] Agentic Real Time Framework (ARTF) Version 1.0 FINAL and protobuf definitions (agenticrtbframework.proto). IAB Tech Lab (IABTechLab/agentic-real-time-framework), 2025-11-12. https://github.com/IABTechLab/agentic-real-time-framework Accessed 2026-10-24.

[217] foursquare/foursquare-places-mcp: MCP Server enabling access to the Foursquare API. Foursquare (GitHub), 2025-01-22. https://github.com/foursquare/foursquare-places-mcp Accessed 2026-10-24.

[218] mapbox/mcp-server: Mapbox Model Context Protocol (MCP) server. Mapbox (GitHub), 2026-10-15. https://github.com/mapbox/mcp-server Accessed 2026-10-24.

[219] OpenRTB 2.6 specification (2.6.md, main branch), section 3.2.19 Object: Geo. IAB Tech Lab, 2022-10 (base 2.6-202210; rolling non-breaking updates since 2.6-202211). https://github.com/InteractiveAdvertisingBureau/openrtb2.x/blob/main/2.6.md Accessed 2026-10-24.

[220] Google Authorized Buyers: OpenRTB guide (BidRequest Geo object). Google, 2026-10-15. https://developers.google.com/authorized-buyers/rtb/openrtb-guide Accessed 2026-10-24.

[221] adcontextprotocol/adcp GitHub releases (latest and recent tags). adcontextprotocol (GitHub), 2026-10-23. https://github.com/adcontextprotocol/adcp/releases Accessed 2026-10-24.

[222] Radar documentation MCP server (docs.radar.com/mcp). Radar, 2026-10-24. https://docs.radar.com/mcp Accessed 2026-10-24.

[223] MCP server (docs, integrations). Radar Labs, Inc., undated (live docs). https://docs.radar.com/integrations/mcp.md Accessed 2026-10-24.

[224] Pi for Placer MCP (Placer documentation). Placer.ai (Placer Labs Inc.), 2026-10-22 (updatedAt in page front matter). https://docs.placer.ai/page/placer-mcp.md Accessed 2026-10-24.

[225] Companies House: BLIS HOLDCO LIMITED (13974159) company overview, previous names, filing history and persons with significant control. Companies House (UK registrar), live register. https://find-and-update.company-information.service.gov.uk/company/13974159/persons-with-significant-control Accessed 2026-10-24.

[226] Blis Holdco Limited, Group of companies' accounts made up to 30 June 2024 (strategic report, group statement of comprehensive income, notes 1.6 and 3). Blis Holdco Limited (filed at Companies House 15 January 2025; auditor Moore Kingston Smith LLP), 2025-01-15 (filed). https://find-and-update.company-information.service.gov.uk/company/13974159/filing-history/MzQ1MTI2NTU0OWFkaXF6a2N4/document?format=pdf&download=0 Accessed 2026-10-24.

[227] Now live: Announcing T-Mobile data in the Blis DSP (release). Blis, 2025-09-10. https://blis.com/now-live-announcing-t-mobile-data-in-the-blis-dsp/ Accessed 2026-10-24.

[228] Blis homepage and 'Our Technology' page (titles only). Blis (T-Mobile Advertising Solutions), 2026. https://www.blis.com/ Accessed 2026-10-24. Access: partial extract.

[229] Cookieless DSP Technology | Geospatial AI | Blis ('Our Technology': Blis Platform, Dynamic Audience Targeting, Smart Holdout Groups, Geo Data, Geospatial AI, Telco Data; including collapsed 'Read more' text). Blis, 2024-05-30 (datePublished); modified 2026-05-15. https://blis.com/our-technology/ Accessed 2026-10-24.

[230] Blis Press Kit 2021. Blis, 2021-08. https://blis.com/wp-content/uploads/2021/08/Blis_Press_Kit_2021.pdf Accessed 2026-10-24.

[231] Location Sciences Group PLC: Final Results for the year ended 31 December 2019 (RNS 1714G, 'Annual Financial Report'). Location Sciences Group PLC (AIM: LSAI), via Investegate, 2020-03-16. https://www.investegate.co.uk/announcement/rns/location-sciences-group--lsai/annual-financial-report/5984407 Accessed 2026-10-24. Access: partial extract.

[232] Location Sciences Group PLC: Final Results for the year ended 31 December 2020 (RNS 6337S). Location Sciences Group PLC (AIM: LSAI), as reproduced on Stockopedia (Sorted Group news archive), 2021-03-18. https://www.stockopedia.com/share-prices/sorted-group-LON:SORT/news/location-sciences-final-results-for-the-year-ended-31-december-2020-019e6c70-5421-726e-9922-071c2e28453f/ Accessed 2026-10-24.

[233] Mobile Ad Firms Location Sciences and Blis Locked in Defamation Dispute. Adweek (Andrew Blustein), 2020-12-15. https://www.adweek.com/programmatic/mobile-ad-firms-location-sciences-blis-defamation-dispute/ Accessed 2026-10-24. Access: partial extract.

[234] T-Mobile Enhances Marketing Capabilities and Expands Advertising Solutions with Close of Vistar Media and Announcement of Blis Acquisitions (joint release, as hosted on blis.com). T-Mobile US, Inc. and Blis, 2025-03-05. https://blis.com/t-mobile-enhances-marketing-capabilities-and-expands-advertising-solutions-with-close-of-vistar-media-and-announcement-of-blis-acquisitions/ Accessed 2026-10-24.

[235] Blis Privacy Policy For Online Advertising and Related Uses. Blis Global Ltd, 2025-11-17 (stated effective date of the latest update; page modified 2025-11-17). https://blis.com/blis-privacy-policy-for-online-advertising-and-related-uses/ Accessed 2026-10-24.

[236] How Foursquare Built The Most Powerful Attribution Solution. Foursquare, 2024-09-30. https://foursquare.com/resources/blog/products/how-foursquare-built-the-most-powerful-attribution-solution Accessed 2026-10-24.

[237] Eligible Platforms. Foursquare, undated (accessed 2026-10-24). https://docs.foursquare.com/campaign-products/docs/eligible-platforms Accessed 2026-10-24.

[238] Self-Serve Requirements (minimums, cost, ready-to-use audiences). Foursquare, undated (accessed 2026-10-24). https://docs.foursquare.com/campaign-products/docs/requirements Accessed 2026-10-24.

[239] Research Visits Feed. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/research-visits-feed Accessed 2026-10-24.

[240] Foursquare Data Normalization. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/foursquare-data-normalization Accessed 2026-10-24.

[241] How Does Visits Work?. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/how-does-visits-work Accessed 2026-10-24.

[242] Visits Delivery Options. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/visits-delivery-options Accessed 2026-10-24.

[243] NAI Precise Location Information Solution Provider Voluntary Enhanced Standards, Updated and Effective October 2024. Network Advertising Initiative, 2024-10 (document text); PDF metadata created 2025-03-21. https://thenai.org/wp-content/uploads/2025/03/NAI-Precise-Location-Information-Solution-Provider-Voluntary-Enhanced-Standards.pdf Accessed 2026-10-24.

[244] Foursquare Privacy Policy for Consumer Services. Foursquare, 2026-02-20 (Last Updated); effective 2026-03-22. https://foursquare.com/legal/privacy-center/foursquare-privacy-policy-consumer-services/ Accessed 2026-10-24.

[245] MRC grants Foursquare Visits first-ever accreditation. Foursquare, 2020-08-13. https://foursquare.com/resources/blog/news/foursquare-visits-receives-first-media-rating-coun/ Accessed 2026-10-24.

[246] Attribution FAQ. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/campaign-products/docs/attribution-faq Accessed 2026-10-24.

[247] Calculating Feasibility (Campaign Feasibility Calculator). Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/campaign-products/docs/calculating-feasibility Accessed 2026-10-24.

[248] Relative Performance Metrics. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/campaign-products/docs/relative-performance-metrics Accessed 2026-10-24.

[249] Research Visits Feed FAQs. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/research-visits-feed-faqs Accessed 2026-10-24.

[250] Consumer Health Data Privacy Notice. Foursquare, 2024-03-29 (Last Updated). https://foursquare.com/legal/privacy-center/consumer-health-data-privacy-notice/ Accessed 2026-10-24.

[251] California Data Broker Registry, registration information download (registry.csv). California Privacy Protection Agency (CalPrivacy), 2026-07-29. https://cppa.ca.gov/data_broker_registry/registry.csv Accessed 2026-10-24.

[252] Blueprints 101 (GroundTruth Help Center). GroundTruth, 2026-02-25. https://help.groundtruth.com/hc/en-us/articles/360051394534-Blueprints-101 Accessed 2026-10-24.

[253] How do you determine the boundaries of a store location or POI? (GroundTruth Help Center). GroundTruth, 2024-04-09. https://help.groundtruth.com/hc/en-us/articles/360000694907-How-do-you-determine-the-boundaries-of-a-store-location-or-POI Accessed 2026-10-24.

[254] GroundTruth Blueprints: An inside look at the tech that powers our Platform. GroundTruth, 2026-03-03. https://www.groundtruth.com/insight/groundtruth-blueprints-an-inside-look-at-the-tech-that-powers-our-platform/ Accessed 2026-10-24.

[255] Glossary (GroundTruth Help Center). GroundTruth, 2026-01-06. https://help.groundtruth.com/hc/en-us/articles/4412649349523-Glossary Accessed 2026-10-24.

[256] GroundTruth Privacy & Location Targeting Guidelines (GroundTruth Help Center). GroundTruth, 2026-09-28. https://help.groundtruth.com/hc/en-us/articles/360001084668-GroundTruth-Privacy-Location-Targeting-Guidelines Accessed 2026-10-24.

[257] Ground Truth Partners with Place Exchange for Foot Traffic Attribution. Billboard Insider (press release copy; release datelined New York, October 9, 2025), 2025-10-10. https://billboardinsider.com/ground-truth-partners-with-place-exchange-for-foot-traffic-attribution/ Accessed 2026-10-24.

[258] Why GroundTruth? Key Competitive Differentiators (GroundTruth Help Center). GroundTruth, 2026-08-13. https://help.groundtruth.com/hc/en-us/articles/360000720328-Why-GroundTruth-Key-Competitive-Differentiators Accessed 2026-10-24.

[259] Our Latest Accuracy Audit Produces the Best Results Yet. GroundTruth, 2019-04-16. https://www.groundtruth.com/insight/latest-accuracy-audit-produces-best-results-yet/ Accessed 2026-10-24.

[260] How does GroundTruth verify location accuracy? (GroundTruth Help Center). GroundTruth, 2024-04-09. https://help.groundtruth.com/hc/en-us/articles/360000704788-How-does-GroundTruth-verify-location-accuracy Accessed 2026-10-24.

[261] How Does Location Work: Sources of location data (GPS, Wifi, Cell Tower Triangulation) (GroundTruth Help Center). GroundTruth, 2024-04-09. https://help.groundtruth.com/hc/en-us/articles/360000709047-How-Does-Location-Work-Sources-of-location-data-GPS-Wifi-Cell-Tower-Triangulation Accessed 2026-10-24.

[262] A Letter from Our Founder & CEO: xAd Acquires WeatherBug. xAd, Inc. (founder and CEO), 2016-11-15. https://www.groundtruth.com/insight/a-letter-from-our-founder-ceo-xad-acquires-weatherbug/ Accessed 2026-10-24.

[263] Location & Audience Targeting (GroundTruth Ads Manager help centre article 53523927168659). GroundTruth, 2026-09-23. https://help.groundtruth.com/hc/en-us/articles/53523927168659-Location-Audience-Targeting Accessed 2026-10-24.

[264] Predictive AI Audiences (from ZeroToOne.AI) (GroundTruth Help Center). GroundTruth, 2026-10-15. https://help.groundtruth.com/hc/en-us/articles/51864249180307-Predictive-AI-Audiences-from-ZeroToOne-AI Accessed 2026-10-24.

[265] GroundTruth – Website & Ad Platform Privacy Policy. GroundTruth, 2026-04-02. https://www.groundtruth.com/privacy-policy/ Accessed 2026-10-24.

[266] CPV and CPM: Why Both Deserve a Place in Your Marketing Mix. GroundTruth, 2023-06-05. https://www.groundtruth.com/?p=253206 Accessed 2026-10-24.

[267] Intent, Purchase, and Location Data At Scale (Data page). inMarket Media, LLC, undated (footer 2026). https://inmarket.com/data/ Accessed 2026-10-24.

[268] Audience Targeting Solutions | InMarket (Audiences page). inMarket Media, LLC, 2026. https://inmarket.com/solutions/audiences/ Accessed 2026-10-24.

[269] Reddit and InMarket Announce Partnership for Real-World Advertising Measurement with InMarket’s Lift Conversion Index (LCI). inMarket Media, LLC, 2024-11-12. https://inmarket.com/reddit-and-inmarket-announce-partnership-for-real-world-advertising-measurement-with-inmarkets-lift-conversion-index-lci/ Accessed 2026-10-24.

[270] InMarket Announces Partnership for Real-Time, Visit-Based Optimization Solution with Yahoo DSP (press release). inMarket Media, LLC via Business Wire, 2026-04-23. https://www.businesswire.com/news/home/20260423154739/en/InMarket-Announces-Partnership-for-Real-Time-Visit-Based-Optimization-Solution-with-Yahoo-DSP Accessed 2026-10-24.

[271] InMarket and Basis Expand Strategic Partnership to Bring RMN Audiences, Lift Optimized Audiences, and Outcomes-Focused Measurement Natively into Basis Platform (press release). inMarket Media, LLC and Basis via Business Wire, 2026-05-27. https://www.businesswire.com/news/home/20260527338831/en/InMarket-and-Basis-Expand-Strategic-Partnership-to-Bring-RMN-Audiences-Lift-Optimized-Audiences-and-Outcomes-Focused-Measurement-Natively-into-Basis-Platform Accessed 2026-10-24.

[272] Know Where Your Incrementality Came From: Results Now Broken Out by Retailer Type (blog). inMarket Media, LLC, 2026-10-02. https://inmarket.com/blog-new-retailer-type-reporting-cpg-incrementality/ Accessed 2026-10-24.

[273] InMarket Launches Maestrø: The First Agentic Media Optimization Platform to Guarantee Real-World Outcomes Across Platforms (press release). inMarket Media, LLC via Business Wire, 2026-06-18. https://www.businesswire.com/news/home/20260618075727/en/InMarket-Launches-Maestr-The-First-Agentic-Media-Optimization-Platform-to-Guarantee-Real-World-Outcomes-Across-Platforms Accessed 2026-10-24.

[274] NinthDecimal LCI Powered by InMarket (sell sheet, PDF). InMarket / NinthDecimal, unknown. https://go.inmarket.com/hubfs/sellsheets/NinthDecimal-LCI%20Powered%20by%20InMarket.pdf Accessed 2026-10-24.

[275] Complaint, In the Matter of InMarket Media, LLC (FTC Matter concerning collection and use of precise location data through the InMarket SDK). Federal Trade Commission, 2024-01. https://www.ftc.gov/system/files/ftc_gov/pdf/Complaint-InMarketMediaLLC.pdf Accessed 2026-10-24.

[276] GeoLink: InMarket’s self-serve activation platform. inMarket Media, LLC, undated (footer 2026). https://inmarket.com/solutions/geolink/ Accessed 2026-10-24.

[277] Your CPG Measurement Partner Is Changing. Here’s What to Do Next. (blog). inMarket Media, LLC, 2026-03-30. https://inmarket.com/your-cpg-measurement-partner-is-changing-heres-what-to-do-next/ Accessed 2026-10-24.

[278] Texas Secretary of State, Registry of Registered Data Brokers (search portal). Texas Secretary of State, live registry (456 items listed on 2026-10-24). https://texas-sos.appianportalsgov.com/data-broker-registry Accessed 2026-10-24.

[279] Why Marketing Measurement Is Broken — and What Outcome Intelligence Does Instead (blog). inMarket Media, LLC, 2026-06-15. https://inmarket.com/why-marketing-measurement-is-broken-and-what-outcome-intelligence-does-instead/ Accessed 2026-10-24.

[280] Welcome to PAPI (Placer API reference, v1.2.1) and the docs index llms.txt. Placer.ai (Placer Labs Inc.), undated (pages "Updated 11 months ago" at access). https://docs.placer.ai/reference/welcome-to-papi Accessed 2026-10-24.

[281] Welcome to Placer Feeds! (Placer data-feed documentation). Placer.ai (Placer Labs Inc.), undated ("Updated 11 months ago" at access). https://docs.placer.ai/docs/welcome-to-placers-feeds Accessed 2026-10-24.

[282] Access feeds via Snowflake (Placer data-feed documentation). Placer.ai (Placer Labs Inc.), 2026-01-05 (updatedAt in page front matter). https://docs.placer.ai/docs/access-feeds-via-snowflake.md Accessed 2026-10-24.

[283] Glossary (Placer API documentation). Placer.ai (Placer Labs Inc.), 2025-11-17 (updatedAt in page front matter). https://docs.placer.ai/reference/glossary.md Accessed 2026-10-24.

[284] POI data accuracy (Placer Academy help-centre article). Placer.ai (Placer Labs Inc.), undated (site build stamp 2025-09-16). https://connect.placer.ai/use-cases/poi-data-accuracy Accessed 2026-10-24.

[285] An Introduction to Placer Data Version 2.0 (Placer Academy help-centre article). Placer.ai (Placer Labs Inc.), undated (describes a launch "on March 31st"; the API changelog dates it 2024-03-31). https://connect.placer.ai/use-cases/introducing-placer-data-version-2-0 Accessed 2026-10-24.

[286] Sensitive Locations Handling (Placer Academy help-centre article). Placer.ai (Placer Labs Inc.), undated (site build stamp 2025-09-16). https://connect.placer.ai/use-cases/sensitive-locations-handling Accessed 2026-10-24.

[287] What is a POI? (Placer Academy help-centre article). Placer.ai (Placer Labs Inc.), undated. https://connect.placer.ai/use-cases/what-is-a-poi Accessed 2026-10-24.

[288] Data version 2.1 - API enhancements (Placer API changelog). Placer.ai (Placer Labs Inc.), 2024-09 (rollout "expected to take place on September 28th, 2024"). https://docs.placer.ai/changelog/data-version-21-releasing-to-the-api.md Accessed 2026-10-24.

[289] Greater Salt Lake: Placer.ai Analysis (September 2024 presentation deck filed with the Utah Public Notice website). Placer.ai, 2024-09. https://www.utah.gov/pmn/files/1194509.pdf Accessed 2026-10-24.

[290] Placer.ai Location Intelligence Platform: vendor and system fact sheet. City of Santa Cruz, California (IT department; PDF author field "Michael Murray (IT)"), 2026-02-10 (PDF creation date in metadata). https://www.santacruzca.gov/files/assets/city/v/1/it/documents/placer-ai-fact-sheet.pdf Accessed 2026-10-24.

[291] Placer.ai: AM/FM Radio Campaign Drives Significant Increase In Holiday Foot Traffic For A Major Retailer. Pierre Bouvard, Cumulus Media | Westwood One Audio Active Group (blog), 2026-09-08. https://www.westwoodone.com/?p=29356 Accessed 2026-10-24. Access: partial extract.

[292] STAR Registry Entries for Placer Labs Inc.. Cloud Security Alliance, listed since 2026-07-21. https://cloudsecurityalliance.org/star/registry/placer-labs-inc Accessed 2026-10-24.

[293] Location Intelligence & Foot Traffic Data Software – Placer.ai (homepage). Placer Labs, Inc., 2026. https://www.placer.ai/ Accessed 2026-10-24.

[294] Location Analytics for Advertising: Audience Insights (Placer.ai solution page). Placer Labs, Inc., 2026. https://www.placer.ai/solutions/advertising Accessed 2026-10-24.

[295] Letter from 14 US Senators to Placer.ai CEO on abortion clinic data (17 May 2022). Senators Elizabeth Warren, Tammy Baldwin, Patty Murray, Tina Smith, Bernard Sanders, Edward J. Markey, Richard Blumenthal, Cory A. Booker, Amy Klobuchar, Christopher S. Murphy, Ron Wyden, Tammy Duckworth, Alex Padilla, Ben Ray Luján, 2022-05-17. https://www.warren.senate.gov/imo/media/doc/2022.05.17%20Letters%20to%20Placer.ai%20on%20Abortion%20Clinic%20Data.pdf Accessed 2026-10-24.

[296] Placer.ai response to Senator Warren (13 June 2022). Noam Ben-Zvi, Co-Founder and CEO, Placer.ai, 2022-06-13. https://www.warren.senate.gov/imo/media/doc/Placer.ai%20Response%20to%20Senator%20Warren.pdf Accessed 2026-10-24.

[297] Placer AI Data Breach Investigation. Almeida Law Group, 2026-10 (page shows "Date of data breach: October 12, 2026"; no separate publication date). https://www.almeidalawgroup.com/data-breach-news/placer-ai-data-breach-investigation/ Accessed 2026-10-24. Access: partial extract.

[298] Geofences (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/geofences.md Accessed 2026-10-24.

[299] Places (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/places.md Accessed 2026-10-24.

[300] Tracking options reference (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/tracking.md Accessed 2026-10-24.

[301] SDK Overview (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/sdk.md Accessed 2026-10-24.

[302] API Reference (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/api.md Accessed 2026-10-24.

[303] GitHub API: radarlabs SDK repositories and latest releases (radar-sdk-ios, radar-sdk-android, react-native-radar, radar-sdk-js, flutter-radar). GitHub (repository metadata published by Radar Labs), state at access. https://api.github.com/repos/radarlabs/radar-sdk-ios Accessed 2026-10-24.

[304] Privacy Notice. Radar Labs, Inc., 2025-05-13 (last updated). https://radar.com/privacy Accessed 2026-10-24.

[305] Data Processing Addendum. Radar Labs, Inc., 2023-10-06 (last updated). https://radar.com/dpa Accessed 2026-10-24.

[306] FAQs (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/faqs.md Accessed 2026-10-24.

[307] How accurate is geofencing? The truth about real-world precision (blog). Radar Team, Radar Labs, Inc., 2025-11-12. https://radar.com/blog/how-accurate-is-geofencing Accessed 2026-10-24.

[308] Radar Geofencing (product page). Radar Labs, Inc., undated (footer © 2026). https://radar.com/product/geofencing Accessed 2026-10-24.

[309] About | Radar. Radar Labs, Inc., undated (footer © 2026). https://radar.com/about Accessed 2026-10-24.

[310] Radar for gaming (industry page). Radar Labs, Inc., undated (footer © 2026). https://radar.com/industries/gaming Accessed 2026-10-24.

[311] Place categories (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/places/categories.md Accessed 2026-10-24.

[312] Radar Trust Center. Radar Labs, Inc. (hosted by Vanta), state at access. https://trust.radar.com/ Accessed 2026-10-24. Access: partial extract.

[313] Our commitment to privacy (blog). Nick Patrick, Radar Labs, Inc., 2018-10-24. https://radar.com/blog/our-commitment-to-privacy Accessed 2026-10-24.

[314] Kentucky Horse Racing and Gaming Corporation licence details: Radar Labs, Inc. (licence 384420). Kentucky Horse Racing and Gaming Corporation, state at access. https://khrc.ky.gov/Search/Lic_Details.aspx?id=384420 Accessed 2026-10-24.

[315] Radar 2025 recap and 2026 preview (blog). Nick Patrick, Radar Labs, Inc., 2026-01-02. https://radar.com/blog/radar-2025-recap-and-2026-preview Accessed 2026-10-24.

[316] Campaign scheduling and user insights (blog). Nick Patrick, Radar Labs, Inc., 2026-07-23. https://radar.com/blog/campaign-scheduling-and-user-insights Accessed 2026-10-24.

[317] Radar Engage (solution page). Radar Labs, Inc., undated (footer © 2026). https://radar.com/solutions/engage Accessed 2026-10-24.

[318] User insights (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/user-insights.md Accessed 2026-10-24.

[319] TriplePoint Venture Growth BDC Corp. Form 10-Q for the quarter ended June 30, 2026 (schedule of investments). TriplePoint Venture Growth BDC Corp. (SEC filing), 2026-08-05 (filed). https://www.sec.gov/Archives/edgar/data/1580345/000158034526000025/tpvg-20260630.htm Accessed 2026-10-24.

[320] TriplePoint Private Venture Credit Inc. Form 10-Q for the quarter ended June 30, 2026 (schedule of investments). TriplePoint Private Venture Credit Inc. (SEC filing), 2026-08-14 (filed). https://www.sec.gov/Archives/edgar/data/1792509/000179250926000024/tpvc-20260630.htm Accessed 2026-10-24.

[321] Terms of Service. Radar Labs, Inc., 2023-05-24 (last updated). https://radar.com/terms Accessed 2026-10-24.

[322] Pricing | Radar. Radar Labs, Inc., undated (footer © 2026). https://radar.com/pricing Accessed 2026-10-24.

[323] Android SDK (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/android.md Accessed 2026-10-24.

[324] Visit Length (Metrics) | Unacast Documentation. Unacast (docs.unacast.com), undated. https://docs.unacast.com/datasets/metrics/visit_length Accessed 2026-10-24.

[325] Foot Traffic (Metrics) | Unacast Documentation. Unacast (docs.unacast.com; banner says the documentation "has moved to Pine59"), undated (footer "Copyright © 2026 Unacast"; mentions a Q4 2025 release). https://docs.unacast.com/datasets/metrics/foot_traffic Accessed 2026-10-24.

[326] Welcome to Pine59. Pine59 (co-founders Tor Åge Ballo, Thomas Walle, Kjartan Slette), 2026-09-01 (datePublished in page metadata). https://pine59.com/blog/welcome-to-pine59 Accessed 2026-10-24.

[327] Our Unique Methodology | Unacast. Unacast, undated (live page). https://www.unacast.com/methodology Accessed 2026-10-24.

[328] Location Data | Unacast (product page). Unacast, undated (live page). https://www.unacast.com/products/location-data Accessed 2026-10-24.

[329] From Raw to Processed Location Data. Unacast (Sierra Daggett, Product Marketing Manager), 2024-07-29. https://www.unacast.com/post/raw-to-processed-location-data Accessed 2026-10-24.

[330] Unacast Mobility Data | Global Mobile Location Data - 70B Daily Signals (Datarade listing). Datarade (provider-supplied listing), undated (live listing). https://datarade.ai/data-products/unacast-mobility-data-global-mobile-location-data-current-unacast Accessed 2026-10-24. Access: partial extract.

[331] Foot Traffic Data | Unacast (product page: Location Intelligence APIs and DaaS). Unacast, 2026-10-24 (HTTP Last-Modified header 24 Oct 2026 12:35 GMT). https://www.unacast.com/products/foot-traffic-data Accessed 2026-10-24.

[332] Pricing | Unacast. Unacast, undated (live page). https://www.unacast.com/pricing Accessed 2026-10-24.

[333] Confidential notification of a personal data breach to the Norwegian Data Protection Authority (Datatilsynet), Gravy Analytics / Unacast (redacted). Gravy Analytics / Unacast, submitted by BakerHostetler; published by NRK, undated notice (DocumentCloud upload 2025-01-13; an identical copy, same file hash, was uploaded by NRK on 2025-01-12 as document 25485205). https://www.documentcloud.org/documents/25485945-datatilsynet-unacast-security-incident-notification-redacted/ Accessed 2026-10-24.

[334] Vistar Media brings T-Mobile audiences to digital out-of-home, exclusively through its demand-side platform. Vistar Media (byline Allison Cantwell), 2026-10-05. https://www.vistarmedia.com/news/vistar-media-t-mobile-audiences Accessed 2026-10-24.

[335] T-Mobile Brings Its Mobile Data Exclusively To Vistar To Scale Up DOOH Targeting. Anthony Vargas, AdExchanger, 2026-10-05. https://adexchanger.com/digital-out-of-home/t-mobile-brings-its-mobile-data-exclusively-to-vistar-to-scale-up-dooh-targeting Accessed 2026-10-24.

[336] T-Mobile to Acquire Vistar Media, Fueling Better Ad Experiences for Consumers and More Effective Products for Advertisers. Vistar Media (republished T-Mobile announcement; byline Leslie Lee), 2025-01-13. https://www.vistarmedia.com/news/t-mobile-to-acquire-vistar-media-fueling-better-ad-experiences-for-consumers-and-more-effective-products-for-advertisers Accessed 2026-10-24.

[337] Vistar Media + The Trade Desk (partner page and activation guide). Vistar Media, undated (page footer (c) 2026). https://www.vistarmedia.com/vistar-the-trade-desk-partnership Accessed 2026-10-24.

[338] The definitive guide to DOOH audience & targeting (Vistar Media blog). Aliana Heffernan, Vistar Media, 2021-01-01. https://www.vistarmedia.com/blog/vistar-media-dooh-audience-targeting Accessed 2026-10-24.

[339] Services Privacy Notice (Last Update: July 1, 2026). Vistar Media, Inc., 2026-07-01. https://www.vistarmedia.com/privacy Accessed 2026-10-24.

[340] In the Matter of T-Mobile USA, Inc., Forfeiture Order, FCC 24-43 (File No. EB-TCD-18-00027702). Federal Communications Commission, 2024-04-29 (released; adopted 2024-04-17). https://docs.fcc.gov/public/attachments/FCC-24-43A1.pdf Accessed 2026-10-24.

[341] Supreme Court docket No. 25-1422, Sprint Corporation, nka Sprint LLC, et al. v. Federal Communications Commission, et al.. Supreme Court of the United States, docketed 2026-06-26. https://www.supremecourt.gov/docket/docketfiles/html/public/25-1422.html Accessed 2026-10-24.

[342] Blis Privacy Centre (Privacy FAQs; 'Blis third parties and partners' table). Blis Global Ltd, undated (live page). https://blis.com/privacy-centre/ Accessed 2026-10-24.

[343] Blis is the first global location intelligence company to offer independent location data verification by Location Sciences (release). Blis and Location Sciences (PR Newswire UK), 2019-09-04. https://prnewswire.co.uk/news-releases/blis-is-the-first-global-location-intelligence-company-to-offer-independent-location-data-verification-by-location-sciences-871735475.html Accessed 2026-10-24.

[344] Blis featured in WARC: Omnichannel campaigns deliver greater sales impact (post). Blis, 2026-07-27. https://blis.com/blis-featured-in-warc-omnichannel-campaigns-deliver-greater-sales-impact/ Accessed 2026-10-24.

[345] NAI Members. Network Advertising Initiative, undated (live page). https://thenai.org/members/ Accessed 2026-10-24.

[346] 67% of Marketers Say Audience Data Is Siloed. Carrier data is the fix. (blog, page title 'How Blis Uses Carrier Data to Improve Audience Targeting'). Blis (Avery Rudman), 2026-10-01. https://blis.com/67-of-marketers-say-audience-data-is-siloed-carrier-data-is-the-fix/ Accessed 2026-10-24.

[347] Location Intelligence Platform | Foursquare (homepage). Foursquare Labs, Inc., 2026. https://foursquare.com/ Accessed 2026-10-24.

[348] Places Data Overview and Supported Countries (Foursquare Data Products documentation). Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/places-data-overview Accessed 2026-10-24.

[349] Places Pro and Premium Data Schemas. Foursquare, undated (accessed 2026-10-24). https://docs.foursquare.com/data-products/docs/places-pro-and-premium Accessed 2026-10-24.

[350] Measurement API (MAPI) overview. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/developer/reference/mapi-overview Accessed 2026-10-24.

[351] Proving Linear TV Drives Real-World Results (Foursquare Attribution case study). Foursquare, 2026-04-16. https://foursquare.com/resources/case-studies/qsr/proving-linear-tv-drives-real-world-results/ Accessed 2026-10-24.

[352] Privacy Center. Foursquare, undated (live page). https://foursquare.com/legal/privacy-center/ Accessed 2026-10-24.

[353] Proskin v. Foursquare Labs, Inc., No. 3:23-cv-30078 (D. Mass.), docket metadata via CourtListener RECAP search API. CourtListener (Free Law Project) / US District Court, D. Massachusetts, 2023-07-24 (filed). https://www.courtlistener.com/docket/67629544/proskin-v-foursquare-labs-inc/ Accessed 2026-10-24. Access: partial extract.

[354] Movement SDK FAQs. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/developer/docs/movement-sdk-faqs Accessed 2026-10-24.

[355] On the Validity of Geosocial Mobility Traces (HotNets 2013). Zhang, Zhou, Zhao, Wang, Su, Metzger, Zheng, Zhao (UC Santa Barbara), 2013-11. https://sites.cs.ucsb.edu/~ravenben/publications/pdf/foursquare-hotnets13.pdf Accessed 2026-10-24.

[356] GroundTruth Earns MRC Accreditation for Visitation Data. GroundTruth, 2021-08-03. https://www.groundtruth.com/?p=424154 Accessed 2026-10-24.

[357] How does GroundTruth verify its location data? (GroundTruth Help Center). GroundTruth, 2024-04-09. https://help.groundtruth.com/hc/en-us/articles/360000709067-How-does-GroundTruth-verify-its-location-data Accessed 2026-10-24.

[358] Geofencing companies comparison. GroundTruth, 2023-11-09 (article:modified_time). https://www.groundtruth.com/geofencing-companies-comparison/ Accessed 2026-10-24.

[359] What is a standard match rate for visitation? (GroundTruth Help Center). GroundTruth, 2025-03-31. https://help.groundtruth.com/hc/en-us/articles/39887097051411-What-is-a-standard-match-rate-for-visitation Accessed 2026-10-24.

[360] Tracking & Measurement (GroundTruth Help Center). GroundTruth, 2026-09-01. https://help.groundtruth.com/hc/en-us/articles/53501393669907-Tracking-Measurement Accessed 2026-10-24.

[361] What is Foot Traffic Attribution?. GroundTruth (glossary), unknown (copyright 2026). https://www.groundtruth.com/glossary_term/foot-traffic-attribution/ Accessed 2026-10-24.

[362] Reporting Overview (GroundTruth Help Center). GroundTruth, 2026-08-21. https://help.groundtruth.com/hc/en-us/articles/54568508162195-Reporting-Overview Accessed 2026-10-24.

[363] Can I set up external reporting API? (GroundTruth Help Center). GroundTruth, 2026-08-18. https://help.groundtruth.com/hc/en-us/articles/4402393255315-Can-I-set-up-external-reporting-API Accessed 2026-10-24.

[364] Reporting API documentation (reporting.groundtruth.com/api). GroundTruth, undated. https://reporting.groundtruth.com/api Accessed 2026-10-24. Access: partial extract.

[365] xAd Raises $42.5M Series E, Acquires WeatherBug; Accelerates Location Technology to Predict Where Consumers Will Go Next. xAd, Inc., 2016-11-15. https://www.groundtruth.com/article/xad-raises-42-5m-series-e-acquires-weatherbug-accelerates-location-technology-to-predict-where-consumers-will-go-next/ Accessed 2026-10-24.

[366] Which countries can I target within Ads Manager? (GroundTruth Help Center). GroundTruth, 2025-11-15. https://help.groundtruth.com/hc/en-us/articles/4402387540883-Which-countries-can-I-target-within-Ads-Manager Accessed 2026-10-24.

[367] Toyota Drives 1.2k Visits with GroundTruth's Cost Per Visit Model. GroundTruth, 2024-11-15. https://www.groundtruth.com/insight/toyota/ Accessed 2026-10-24.

[368] Media Rating Council Accreditation Updates, October 2023 through December 2023. Media Rating Council, 2024-01-08. https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/010824_MRC_Accreditation_Status_Communication.pdf Accessed 2026-10-24.

[369] Media Rating Council Accreditation Updates, January 2024 through March 2024. Media Rating Council, 2024-04-24. https://mediaratingcouncil.org/sites/default/files/News/050624_MRC_Accreditation_Status_Communication_FINAL.pdf Accessed 2026-10-24.

[370] Can we target a list of medical locations or facilities? (GroundTruth Help Center). GroundTruth, 2025-03-31. https://help.groundtruth.com/hc/en-us/articles/4402388584595-Can-we-target-a-list-of-medical-locations-or-facilities Accessed 2026-10-24.

[371] Can I blueprint physical therapy locations? (GroundTruth Help Center). GroundTruth, 2025-11-15. https://help.groundtruth.com/hc/en-us/articles/39886197033107-Can-I-blueprint-physical-therapy-locations Accessed 2026-10-24.

[372] ZeroToOne.AI Acquires GroundTruth To Scale Predictive Intelligence For Enterprise Decision Systems. Pulse 2.0 (Amit Chowdhry), 2026-03-28. https://pulse2.com/zerotoone-ai-acquires-groundtruth-to-scale-predictive-intelligence-for-enterprise-decision-systems Accessed 2026-10-24.

[373] Location Sciences Recognizes InMarket’s Stellar Location Precision and Accuracy in Initial Audiences Audits (republished InMarket release). InMarket release republished by SalesTechStar, 2020-09-01. https://salestechstar.com/?p=92942 Accessed 2026-10-24.

[374] The Total Economic Impact of InMarket’s LCI Attribution Solution (Forrester TEI landing page). inMarket Media, LLC (landing page for a Forrester Consulting study), undated page (study dated July 2021 per S10). https://inmarket.com/resources/the-total-economic-impact-of-inmarkets-lci-attribution-solution/ Accessed 2026-10-24. Access: partial extract.

[375] Measurement (InMarket Measurement: Unified Measurement, Campaign Incrementality, Visit Stream). inMarket Media, LLC, undated (footer 2026). https://inmarket.com/solutions/measurement/ Accessed 2026-10-24.

[376] Ethics Principles. inMarket Media, LLC, undated (footer 2026). https://inmarket.com/ethics-principles/ Accessed 2026-10-24.

[377] LCI AI InSights (InMarket product page; /lci/ redirects here). inMarket Media, LLC, undated (footer 2026). https://inmarket.com/solutions/lci-ai-insights/ Accessed 2026-10-24.

[378] Location Sciences Group PLC: Contract with InMarket (RNS announcement). Location Sciences Group PLC (via Investegate / RNS), 2020-09-01. https://www.investegate.co.uk/announcement/rns/location-sciences-group--lsai/contract/6401655 Accessed 2026-10-24.

[379] Privacy FAQ. Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/company/privacy-faq Accessed 2026-10-24.

[380] Placer.ai boosts valuation to $1.5B after quietly raising another $75M. TechCrunch (Ingrid Lunden), 2024-08-05. https://techcrunch.com/2024/08/05/placer-ai-boosts-valuation-to-1-5b-after-quietly-raising-another-75m/ Accessed 2026-10-24.

[381] v1.2.0 - Visit duration segmentation (Placer API changelog). Placer.ai (Placer Labs Inc.), 2024-07 ("As of July 2024"). https://docs.placer.ai/changelog/v12-visit-duration-segmentation.md Accessed 2026-10-24.

[382] Data version 2.0 - API updates (Placer API changelog). Placer.ai (Placer Labs Inc.), 2024-03. https://docs.placer.ai/changelog/data-version-20-api-updates.md Accessed 2026-10-24.

[383] Retail Foot Traffic Data: What It Costs and Where It's Wrong (2026). Clyde Christian Anderson, Founder and CEO, GrowthFactor, 2025-12-18 (updated 2026-10-09). https://www.growthfactor.ai/resources/blog/retail-foot-traffic-data Accessed 2026-10-24. Access: partial extract.

[384] Access Placer Data Programmatically With the Placer API (product page). Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/products/api Accessed 2026-10-24.

[385] Location Data Firm Provides Heat Maps of Where Abortion Clinic Visitors Live. Joseph Cox, Motherboard (Vice), 2022-05-05. https://www.vice.com/en/article/location-data-firm-heat-maps-planned-parenthood-abortion-clinics-placer-ai/ Accessed 2026-10-24.

[386] Trust Center: Our Commitment to Stay a Privacy-First Company. Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/company/trust-center Accessed 2026-10-24.

[387] Information Risk & Compliance. Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/company/information-risk-compliance Accessed 2026-10-24.

[388] Pricing. Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/pricing Accessed 2026-10-24.

[389] Protecting data privacy (Placer Academy help-centre article). Placer.ai (Placer Labs Inc.), undated (site build stamp 2025-09-16). https://connect.placer.ai/use-cases/protecting-data-privacy Accessed 2026-10-24.

[390] Radar | Location Infrastructure and Intelligence (homepage). Radar Labs, Inc., 2026. https://radar.com/ Accessed 2026-10-24.

[391] Beacons (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/beacons.md Accessed 2026-10-24.

[392] Web SDK (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/web.md Accessed 2026-10-24.

[393] Trips (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/trips.md Accessed 2026-10-24.

[394] Fraud (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/fraud.md Accessed 2026-10-24.

[395] Campaigns (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/geofencing/campaigns.md Accessed 2026-10-24.

[396] iOS SDK (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/ios.md Accessed 2026-10-24.

[397] React Native SDK (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/react-native.md Accessed 2026-10-24.

[398] Flutter package (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/flutter.md Accessed 2026-10-24.

[399] Capacitor plugin (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/capacitor.md Accessed 2026-10-24.

[400] Cordova plugin (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/cordova.md Accessed 2026-10-24.

[401] MAUI SDK (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/sdk/maui.md Accessed 2026-10-24.

[402] Geo-compliance updates: Product, press, licensing, conferences, and more (blog). Radar Labs, Inc., 2024-04-02. https://radar.com/blog/geo-compliance-updates-april-2024 Accessed 2026-10-24.

[403] Integrations (docs). Radar Labs, Inc., undated (live docs). https://docs.radar.com/integrations/integrations.md Accessed 2026-10-24.

[404] Introducing Radar’s MCP server and integrations with ChatGPT and Claude (blog). Nick Patrick, Radar Labs, Inc., 2026-10-14. https://radar.com/blog/introducing-radars-mcp-server-and-integrations-with-chatgpt-and-claude Accessed 2026-10-24.

[405] Introducing our fraud detection and geo-compliance solution (blog). Radar Team, Radar Labs, Inc., 2023-10-03. https://radar.com/blog/introducing-fraud-detection-geo-compliance-solution Accessed 2026-10-24.

[406] radar-sdk-ios: RadarSDK/PrivacyInfo.xcprivacy (Apple privacy manifest). Radar Labs, Inc. (GitHub radarlabs/radar-sdk-ios), 2024-06-01 (last commit to the file, per GitHub API). https://raw.githubusercontent.com/radarlabs/radar-sdk-ios/master/RadarSDK/PrivacyInfo.xcprivacy Accessed 2026-10-24.

[407] FTC Legal Library: Cases and Proceedings search for "Radar". U.S. Federal Trade Commission, state at access. https://www.ftc.gov/legal-library/browse/cases-proceedings?search=Radar Accessed 2026-10-24. Access: partial extract.

[408] Places (Location types) | Unacast Documentation. Unacast (docs.unacast.com), undated. https://docs.unacast.com/datasets/locations/points_of_interest Accessed 2026-10-24.

[409] Custom Locations | Unacast Documentation. Unacast (docs.unacast.com), undated. https://docs.unacast.com/datasets/locations/custom_boundaries Accessed 2026-10-24.

[410] Understanding and Comparing Mobility Data (4 February 2021). Consumer Data Research Centre (Nick Bearman), 2021-02-04. https://www.cdrc.ac.uk/understanding-and-comparing-mobility-data-4th-feb-2021/ Accessed 2026-10-24. Access: partial extract.

[411] Delivery and File Delivery Options | Unacast Documentation. Unacast (docs.unacast.com), undated. https://docs.unacast.com/datasets/delivery/daas/file-delivery-options Accessed 2026-10-24.

[412] Aggregated Foot Traffic Datasets | Unacast. Unacast, undated (live page). https://www.unacast.com/products/aggregated-foot-traffic Accessed 2026-10-24.

[413] About Pine59. Pine59, undated (footer © 2026). https://pine59.com/about Accessed 2026-10-24.

[414] Unacast (NAI member page). Network Advertising Initiative, 2021-07-24. https://thenai.org/members/gravy-analytics/ Accessed 2026-10-24.

[415] Foot traffic schema for Places | Unacast Documentation. Unacast (docs.unacast.com), undated. https://docs.unacast.com/datasets/schemas/places/foot_traffic_new Accessed 2026-10-24.

[416] Foursquare (Vistar Media partner page). Vistar Media, undated (accessed 2026-10-24). https://www.vistarmedia.com/partner-foursquare Accessed 2026-10-24.

[417] Cuebiq | Vistar Media Partner. Vistar Media, undated (page footer (c) 2026). https://www.vistarmedia.com/partner-cuebiq Accessed 2026-10-24.

[418] Vistar Media Developer Portal and Get Started with the DSP API. Vistar Media, undated (footer (c) 2026). https://developers.vistarmedia.com/apis/dsp/get-started Accessed 2026-10-24. Access: partial extract.

[419] Vistar Media research finds inclusion of store locators in DOOH advertising directly impacts visit and purchase intent. Vistar Media (byline Laura Apostol), 2026-09-23. https://www.vistarmedia.com/news/vistar-media-launches-store-locator-effect-research Accessed 2026-10-24.

[420] Targeting — ID-Precise & Geo-Contextual Audiences | Adsquare. adsquare GmbH, 2026. https://adsquare.com/the-outcomes-loop/targeting Accessed 2026-10-24.

[421] Precisely Audiences - privacy-centric audience segment data (product page). Precisely Software Incorporated, 2026. https://www.precisely.com/product/placeiq-audiences Accessed 2026-10-24.

[422] Audience Segments - Consumer Behavior | Mobilewalla. Mobilewalla, Inc., 2025. https://www.mobilewalla.com/products/audience-segments Accessed 2026-10-24.

[423] LCI AI InSights. InMarket, unknown (copyright 2026). https://inmarket.com/lci Accessed 2026-10-24.

[424] Blis rolls out omnichannel measurement solution into broader market. MarTech Series (syndicated Blis release), 2024-10-10. https://martechseries.com/sales-marketing/programmatic-buying/blis-rolls-out-omnichannel-measurement-solution-into-broader-market/ Accessed 2026-10-24.

[425] Quadrant | Location Data and POI Data Solutions (homepage). Quadrant, an Appen company, 2026. https://www.quadrant.io/ Accessed 2026-10-24.

[426] Complementics | Mobile Location Data & Location Intelligence Feeds (homepage). ECONTEXT.AI LLC dba Complementics, 2026. https://complementics.com/ Accessed 2026-10-24.

[427] Near Hits the Nasdaq. Daily Research News Online (MrWeb), 2023-03-24. https://mrweb.com/drno/news34801.htm Accessed 2026-10-24.

[428] FTC Takes Action Against Gravy Analytics, Venntel for Unlawfully Selling Location Data Tracking Consumers to Sensitive Sites. Federal Trade Commission (press release), 2024-12-03. https://www.ftc.gov/node/86832 Accessed 2026-10-24.

[429] Get ready for iCloud Private Relay (WWDC21 session 10096). Apple Inc. (Developer), 2021-06. https://developer.apple.com/videos/play/wwdc2021/10096/ Accessed 2026-10-24.

[430] Prepare Your Network or Web Server for iCloud Private Relay. Apple Inc. (Developer Support), 2021. https://developer.apple.com/support/prepare-your-network-for-icloud-private-relay/ Accessed 2026-10-24.

[431] Location permissions (overview). Google (Android Developers), 2026. https://developer.android.com/develop/sensors-and-location/location/permissions Accessed 2026-10-24.

[432] New safety section in Google Play will give transparency into how apps use data. Google (Android Developers Blog; Suzanne Frey), 2021-05-06. https://android-developers.googleblog.com/2021/05/new-safety-section-in-google-play-will.html Accessed 2026-10-24.

[433] Authorized Buyers Program Guidelines. Google, 2026-05-15. https://www.google.com/doubleclick/adxbuyer/guidelines.html Accessed 2026-10-24.

[434] IP Protection (explainer README). Google Chrome (GitHub GoogleChrome/ip-protection), 2025. https://github.com/GoogleChrome/ip-protection Accessed 2026-10-24.

[435] IP Protection: IP Geolocation explainer. Google Chrome (GitHub GoogleChrome/ip-protection), 2025. https://github.com/GoogleChrome/ip-protection/blob/main/Explainer-IP-Geolocation.md Accessed 2026-10-24.

[436] IP Protection (Privacy Sandbox developer documentation). Google (privacysandbox.google.com), 2025-10-27. https://privacysandbox.google.com/protections/ip-protection Accessed 2026-10-24.

[437] Prebid Server Activity Controls (transmitPreciseGeo). Prebid.org, 2026. https://docs.prebid.org/prebid-server/features/pbs-activitycontrols.html Accessed 2026-10-24.

[438] Prebid.js Activity Controls (transmitPreciseGeo). Prebid.org, 2026. https://docs.prebid.org/dev-docs/activity-controls.html Accessed 2026-10-24.

[439] Prebid Mobile Android: Class TargetingParams (setLocationDecimalPrecision). Prebid.org, 2026. https://docs.prebid.org/prebid-mobile-android/org/prebid/mobile/TargetingParams.html Accessed 2026-10-24.

[440] Location Accuracy: Ad Exchange Data Falls Short (Data-Driven Thinking column). Marcus D. Collins, scientist at Placed, in AdExchanger, 2015-08-27. https://adexchanger.com/data-driven-thinking/location-accuracy-ad-exchange-data-falls-short Accessed 2026-10-24.

[441] Trends on Tuesday: Mobile Location Data (DigitalGov, US General Services Administration). Treci Johnson, DigitalGov (GSA), 2014-06-17. https://federalist-466b7d92-5da1-4208-974f-d61fd4348571.sites.pages.cloud.gov/preview/gsa/digitalgov.gov/bc-archive-content-3/2014/06/17/trends-on-tuesday-mobile-location-data/index.json Accessed 2026-10-24.

[442] Thinknear Rolls Out Location Score, Highlights Accuracy Issues In Location Marketing. Judith Aquino, AdExchanger, 2014-05-19. https://adexchanger.com/mobile/thinknear-rolls-out-location-score-highlights-accuracy-issues-in-location-marketing Accessed 2026-10-24.

[443] Thinknear Location Score Index Q1 2015: Mobile Advertising's Guide to Location Accuracy (slide deck). Thinknear by Telenav, 2015-05. https://www.slideshare.net/slideshow/thinknear-location-scoreindexq12015/48283399 Accessed 2026-10-24.

[444] Geographical Targeting (Authorized Buyers real-time bidding guide). Google, 2026-09-17. https://developers.google.com/authorized-buyers/rtb/geotargeting Accessed 2026-10-24.

[445] MRC Location-Based Advertising Measurement Guidelines, Final Version 1.0, March 2017. Media Rating Council, with the Interactive Advertising Bureau and the Mobile Marketing Association (sponsoring associations), 2017-03. https://mediaratingcouncil.org/sites/default/files/Standards/MRC%20Location-Based%20Advertising%20Measurement%20Guidelines%20Final%20March%202017.pdf Accessed 2026-10-24.

[446] Advanced measurement and targeting platform (Cuebiq blog). Cuebiq Marketing Team, undated (image upload dated 2025; footer © 2026). https://cuebiq.com/blog/advanced-measurement-targeting-platform/ Accessed 2026-10-24.

[447] Measure the Outcomes that Matter To Your Business (Azira Measurement). Azira, unknown (copyright 2026). https://azira.com/solutions/azira-measurement Accessed 2026-10-24.

[448] Adsquare redefines campaign measurement with unified Attribution Dashboard. Adsquare (newsroom), 2025-06-30. https://adsquare.com/resources/newsroom/adsquare-redefines-campaign-measurement-with-unified-attribution-dashboard Accessed 2026-10-24.

[449] Tourism (Arrivalist homepage). Arrivalist, 2026-10-23 (page metadata). https://www.arrivalist.com/tourism Accessed 2026-10-24.

[450] Partners (Arrivalist). Arrivalist, unknown (copyright 2026). https://www.arrivalist.com/partners Accessed 2026-10-24.

[451] TripInsights (Arrivalist). Arrivalist, 2026-10-23 (page metadata). https://www.arrivalist.com/trip-insights Accessed 2026-10-24.

[452] How do you track visits from CTV? (GroundTruth Help Center). GroundTruth, 2021-06-25. https://help.groundtruth.com/hc/en-us/articles/4402806330003-How-do-you-track-visits-from-CTV Accessed 2026-10-24.

[453] How does GroundTruth measure in-store Visits? (GroundTruth Help Center). GroundTruth, 2024-04-09. https://help.groundtruth.com/hc/en-us/articles/360000709087-How-does-GroundTruth-measure-in-store-Visits Accessed 2026-10-24.

[454] MRC Outcomes and Data Quality Standards, Final, September 2022. Media Rating Council, with ANA, 4A's and ACA (sponsoring associations), 2022-09. https://mediaratingcouncil.org/sites/default/files/Standards/MRC%20Outcomes%20and%20Data%20Quality%20Standards%20%28Final%29.pdf Accessed 2026-10-24.

[455] Real-World Data for The Outcomes Era | Adsquare (homepage). adsquare GmbH, 2026. https://www.adsquare.com/ Accessed 2026-10-24.

[456] Addressable Geo-Fencing | Target Households with Precision (Simpli.fi). Simpli.fi, 2026. https://simpli.fi/addressable/ Accessed 2026-10-24.

[457] Beyond These Walls: The Power of Addressable Geo-Fencing (Simpli.fi). Simpli.fi, 2026. https://simpli.fi/latest/beyond-these-walls-the-power-of-addressable-geo-fencing Accessed 2026-10-24.

[458] Azira website home page. Azira LLC, 2026 (copyright year; page undated). https://azira.com/ Accessed 2026-10-24.

[459] Location-Based Audience Targeting | Foursquare Audience (product page). Foursquare Labs, Inc., 2026. https://foursquare.com/products/audience/ Accessed 2026-10-24.

[460] AdCOM 1.0 FINAL: List: Location Types and List: IP Location Services. IAB Tech Lab, 2019 (AdCOM 1.0 Final; master branch as read). https://github.com/InteractiveAdvertisingBureau/AdCOM/blob/master/AdCOM%20v1.0%20FINAL.md Accessed 2026-10-24.

[461] Guidance for NAI Members: Determining Whether Location is Imprecise. Network Advertising Initiative, 2020-02. https://thenai.org/wp-content/uploads/2024/02/nai_impreciselocation2.pdf Accessed 2026-10-24.

[462] IAB Mobile Location Data Guide for Publishers (February 2016, revised). IAB Mobile Location Working Group, 2016-02. https://www.iab.com/wp-content/uploads/2016/04/IAB_Mobile-Location-Data-Guide-for-Publishers_Feb2016-Revised.pdf Accessed 2026-10-24.

[463] Vistar Media US Case Study: DOOH Retail, Target. Vistar Media, hosted by OAAA, 2024-02 (upload path). https://oaaa.org/wp-content/uploads/2024/02/Vistar-Media-US-Case-Study-DOOH-Retail-Target.pdf Accessed 2026-10-24.

[464] Unlocking Outdoor Impact: An Interview on Programmatic DOOH Attribution. Adsquare (The Lab; Serkan Uzundonekoglu and James Morgan), 2025-07-08. https://adsquare.com/resources/the-lab/unlocking-outdoor-impact-an-interview-on-programmatic-dooh-attribution Accessed 2026-10-24.

[465] Near Intelligence, Inc. Form 10-Q for the quarter ended June 30, 2023. Near Intelligence, Inc. (SEC EDGAR), 2023-08-14. https://www.sec.gov/Archives/edgar/data/1826671/000121390023067217/f10q0623_nearintell.htm Accessed 2026-10-24.

[466] EDGAR company submissions record for CIK 0001826671 (Near Intelligence, Inc.; former name KludeIn I Acquisition Corp). U.S. Securities and Exchange Commission (data.sec.gov), 2026-10-24 (as served). https://data.sec.gov/submissions/CIK0001826671.json Accessed 2026-10-24.

[467] Near Intelligence, Inc. Form 8-K (date of report March 27, 2024): confirmation order, effective date, cancellation of equity, intent to file Form 15. Near Intelligence, Inc. (SEC EDGAR), 2024-03-27. https://www.sec.gov/Archives/edgar/data/1826671/000121390024026663/ea0201919-8k_near.htm Accessed 2026-10-24.

[468] Azira Launches Azira One, Bringing AI-Powered Location Intelligence to Destination Marketers. Azira LLC (press release), 2026-09-01. https://azira.com/press-release/azira-launches-azira-one-bringing-ai-powered-location-intelligence-to-destination-marketers Accessed 2026-10-24.

[469] X-Mode Social, Inc., and Outlogic, LLC, In the Matter of (FTC case page, Matter 2123038). Federal Trade Commission, 2024-04-11 (final order date shown). https://www.ftc.gov/node/82318 Accessed 2026-10-24.

[470] Location data at your fingertips: Foursquare is the preferred location launch partner for the AWS Data Exchange. Foursquare Labs, Inc. (blog), 2019-11-13. https://foursquare.com/resources/blog/partnerships/location-data-at-your-fingertips-foursquare/ Accessed 2026-10-24.

[471] Foursquare partners with Snowflake. Foursquare Labs, Inc. (blog), 2021-02-08. https://foursquare.com/?p=7420 Accessed 2026-10-24.

[472] POI (Point of Interest) Free Sample - SafeGraph Places (AWS Marketplace / AWS Data Exchange listing). SafeGraph via Amazon Web Services Marketplace, undated (served 2026-10-24). https://aws.amazon.com/marketplace/pp/prodview-ahrgu4yltfntw Accessed 2026-10-24.

[473] Infillion Gimbal (product page; gimbal.com redirects to infillion.com). Infillion, 2026. https://infillion.com/products/infillion-gimbal/ Accessed 2026-10-24.

[474] Rezolve AI plc Form 20-F for the fiscal year ended 31 December 2025. Rezolve AI plc (SEC EDGAR, CIK 0001920294), 2026-03-30. https://www.sec.gov/Archives/edgar/data/1920294/000119312526132456/rzlv-20251231.htm Accessed 2026-10-24.

[475] bluedot.io (HTTP redirect observed to rezolve.com). Rezolve AI plc (domain destination). https://bluedot.io/ Accessed 2026-10-24.

[476] Qualcomm Completes Skyhook Acquisition. Location Business News, 2022-05-18. https://locationbusinessnews.com/qualcomm-completes-skyhook-acquisition Accessed 2026-10-24.

[477] Location & Sales Data for Finance, Retail & CRE | ADVAN (homepage). Advan Research Corporation, 2026. https://advanresearch.com/ Accessed 2026-10-24.

[478] Veraset (homepage). Veraset, 2026. https://www.veraset.com/ Accessed 2026-10-24.

[479] Cuebiq homepage: Location Intelligence for Brands, Agencies & Research. Cuebiq Group, LLC, 2026. https://cuebiq.com/ Accessed 2026-10-24.

[480] spectus.ai (HTTP redirect observed to cuebiq.com). Cuebiq Group, LLC (domain owner presumed). https://spectus.ai/ Accessed 2026-10-24.

[481] Lifesight home page (2026). Lifesight, 2026 (copyright year; undated). https://lifesight.io/ Accessed 2026-10-24.

[482] onemata.com (Wix 'ConnectYourDomain Error'). unknown. https://onemata.com/ Accessed 2026-10-24. Access: partial extract.

[483] datastreamgroup.com (parking redirect to /lander). unknown. https://datastreamgroup.com/ Accessed 2026-10-24. Access: partial extract.

[484] Plot Projects homepage (access error). Plot Projects. https://www.plotprojects.com/ Accessed 2026-10-24. Access: partial extract.

[485] Best Foot Traffic Data Providers & Companies 2026 | Datarade. Datarade (data marketplace), 2026. https://datarade.ai/data-categories/footfall-traffic-data/providers Accessed 2026-10-24.

[486] Companies House: BLIS GLOBAL LTD (06455773) company overview, previous names, filing history and persons with significant control. Companies House (UK registrar), live register. https://find-and-update.company-information.service.gov.uk/company/06455773 Accessed 2026-10-24.

[487] About Blis | Independent DSP Established in 2004. Blis, 2024-05-27 (datePublished); modified 2026-10-08. https://blis.com/about-us/ Accessed 2026-10-24.

[488] Foursquare cofounder says acquisition gave $0 to equity holders, leaving him 'heartbroken' (Business Insider, republished by DNYUZ). Business Insider (via DNYUZ), 2026-10-20. https://dnyuz.com/2026/10/20/foursquare-cofounder-says-acquisition-gave-0-to-equity-holders-leaving-him-heartbroken/ Accessed 2026-10-24.

[489] About Us: The Leader in Location Technology | GroundTruth. GroundTruth, a ZeroToOne.AI company, 2026. https://www.groundtruth.com/about-us/ Accessed 2026-10-24.

[490] About Unacast (company page). Unacast, undated (live page). https://www.unacast.com/about Accessed 2026-10-24.

[491] FTC Finalizes Order Banning Mobilewalla from Selling Sensitive Location Data. Federal Trade Commission (press release), 2025-01-14. https://www.ftc.gov/node/87190 Accessed 2026-10-24.

[492] T-Mobile | Vistar Media Partner. Vistar Media, undated (page footer (c) 2026). https://www.vistarmedia.com/t-mobile-l-vistar-media-partner Accessed 2026-10-24.

[493] Arity | A mobility data and analytics company (homepage). Arity, 2026. https://arity.com/ Accessed 2026-10-24.

[494] Arity to acquire INRIX, building a more complete understanding of transportation. Arity (blog), 2026-10-16. https://arity.com/move/inrix-announcement/ Accessed 2026-10-24.

[495] Media and Games Invest advances its market position and offering in programmatic mobile location-based display and video advertising with the acquisition of Verve's business assets. Media and Games Invest plc (press release, hosted on press.verve.com), 2020-01-23. https://press.verve.com/media-and-games-invest-advances-its-market-position-and-offering-in-programmatic-mobile-location-based-display-and-video-advertising-with-the-acquisition-of-verves-business-assets Accessed 2026-10-24.

[496] MGI - Media and Games Invest is now Verve. Verve Group SE (press release), 2024-06-13. https://press.verve.com/mgi-media-and-games-invest-is-now-verve Accessed 2026-10-24.

[497] Privacy updates for App Store submissions (third-party SDK privacy manifest and signatures; new use cases for APIs that require reasons). Apple Inc. (Developer News), 2023-12-07. https://developer.apple.com/news/?id=r1henawx Accessed 2026-10-24.

[498] Privacy manifest files (developer documentation). Apple Inc. (Developer Documentation), 2026. https://developer.apple.com/documentation/bundleresources/privacy-manifest-files Accessed 2026-10-24.

[499] Requesting authorization to use location services (developer documentation). Apple Inc. (Developer Documentation), 2026. https://developer.apple.com/documentation/corelocation/requesting-authorization-to-use-location-services Accessed 2026-10-24.

[500] Understanding location in the background permissions (Play Console Help). Google Play, 2026. https://support.google.com/googleplay/android-developer/answer/9799150 Accessed 2026-10-24.

[501] User Data (Play policy). Google Play, 2026. https://support.google.com/googleplay/android-developer/answer/10144311 Accessed 2026-10-24.

[502] Complaint, In the Matter of X-Mode Social, Inc. and Outlogic, LLC (FTC Matter No. 212-3038). Federal Trade Commission, 2024-01. https://www.ftc.gov/system/files/ftc_gov/pdf/X-Mode-Complaint.pdf Accessed 2026-10-24.

[503] FTC Order Prohibits Data Broker X-Mode Social and Outlogic from Selling Sensitive Location Data. Federal Trade Commission (press release), 2024-01-09. https://www.ftc.gov/news-events/news/press-releases/2024/01/ftc-order-prohibits-data-broker-x-mode-social-outlogic-selling-sensitive-location-data Accessed 2026-10-24.

[504] Form S-4/A (Amendment No. 8), KludeIn I Acquisition Corp. (business combination with Near Intelligence). KludeIn I Acquisition Corp. / Near Intelligence (SEC EDGAR, accession 0001213900-23-007945), 2023-02-03. https://www.sec.gov/Archives/edgar/data/1826671/000121390023007945/fs42023a8_kludein1.htm Accessed 2026-10-24.

[505] Form 10-K for the fiscal year ended June 30, 2020, Telenav, Inc.. Telenav, Inc. (SEC EDGAR, accession 0001474439-20-000029), 2020-08-21. https://www.sec.gov/Archives/edgar/data/1474439/000147443920000029/tnav630202010k.htm Accessed 2026-10-24.

[506] Location Sciences Group plc: Interim Results for the six months ended 30 June 2019 (RNS 7068N). Location Sciences Group plc (AIM: LSAI), 2019-09-26. https://www.investegate.co.uk/announcement/rns/location-sciences-group--lsai/half-year-report/5525746 Accessed 2026-10-24.

[507] Location Sciences Group plc: Half-year Report for the six months ended 30 June 2020. Location Sciences Group plc (AIM: LSAI), 2020-09-15. https://www.investegate.co.uk/announcement/rns/location-sciences-group--lsai/half-year-report/5998885 Accessed 2026-10-24.

[508] Location Data: Friend or Foe? An Investigation into the Reliability of Location Signals in Mobile Advertising (in-house presentation abstract). Amir Mehrjoo, IMDEA Networks Institute (NETCOM group, UC3M), 2024-07-17. https://networks.imdea.org/whatsnew/events-agenda/location-data-friend-or-foe-an-investigation-into-the-reliability-of-location-signals-in-mobile-advertising/ Accessed 2026-10-24. Access: partial extract.

[509] ADINT: Using Targeted Advertising for Personal Surveillance (project page; paper at the 2017 Workshop on Privacy in the Electronic Society). Paul Vines, Franziska Roesner, Tadayoshi Kohno (University of Washington, Paul G. Allen School of Computer Science & Engineering), 2017-10. https://adint.cs.washington.edu/ Accessed 2026-10-24. Access: partial extract.

[510] An Unprecedented Era of Mobile Marketing (summary of Fong, Fang and Luo, 'Geo-Conquesting: Competitive Locational Targeting of Mobile Promotions', Journal of Marketing Research 52(5), 726-735, 2015). American Marketing Association (summarising Nathan Fong, Zheng Fang, Xueming Luo), 2015-12-12. https://www.ama.org/2015/12/12/an-unprecedented-era-of-mobile-marketing/ Accessed 2026-10-24. Access: access error; text or metadata via another route.

[511] Mobile Targeting Using Customer Trajectory Patterns (Management Science 65(11), 5027-5049). Anindya Ghose, Beibei Li, Siyuan Liu, 2019-11-01. https://pure.psu.edu/en/publications/mobile-targeting-using-customer-trajectory-patterns/ Accessed 2026-10-24. Access: abstract only.

[512] Chicago Fed Advance Retail Trade Summary (CARTS): About CARTS. Federal Reserve Bank of Chicago, unknown. https://www.chicagofed.org/research/data/carts/about Accessed 2026-10-24.

[513] Measuring Ad Effectiveness Using Geo Experiments. Jon Vaver, Jim Koehler (Google Inc.), 2011. https://research.google/pubs/measuring-ad-effectiveness-using-geo-experiments/ Accessed 2026-10-24. Access: abstract only.

[514] Report: Quality of precise location data declining as more publishers offer it. Greg Sterling, MarTech, 2016-05-17. https://martech.org/report-quality-precise-location-data-declining/ Accessed 2026-10-24.

[515] The importance of accurate data (Apex Mobile Media and Thinknear presentation to IAB Canada, Toronto). Walder Amaya and Chris Lombardi (Apex Mobile Media), Danielle Hanson (Thinknear), 2018-05. https://iabcanada.com/wp-content/uploads/2018/05/Thinknear-Apex-IAB-Toronto-Presentation-.pdf Accessed 2026-10-24.

[516] How Azira and Black C Media Drove a 43% Lift in Store Traffic for Domestos. Azira, undated (footer '© Azira LLC 2026'). https://azira.com/case-studies/how-azira-and-black-c-media-drove-a-43-lift-in-store-traffic-for-domestos Accessed 2026-10-24.

[517] Cuebiq Attribution 2022 Benchmarks. Cuebiq, 2022 (uploaded 2023-09). https://www.cuebiq.com/wp-content/uploads/2023/09/Cuebiq-Attribution-Benchmarks-2022.pdf Accessed 2026-10-24.

[518] Reveal homepage: OOH Measurement & Attribution Platform. Reveal Mobile, Inc., undated (footer © 2026). https://revealmobile.com/?p=18188 Accessed 2026-10-24.

[519] Blis AI - The new AI-powered planning tool (product page). Blis, 2025-06-11; modified 2025-06-27. https://blis.com/blis-ai/ Accessed 2026-10-24.

[520] Fast-Food Brand Increases Visits for Multiple Locations With Simpli.fi's Advanced Location Targeting. Simpli.fi, undated (cites January-April 2021 industry context; footer © 2026). https://simpli.fi/resources/case-studies/fast-food-brand-increases-visits-for-multiple-locations-with-simpli-fis-advanced-location-targeting Accessed 2026-10-24.

[521] MRC grants Foursquare Visits first-ever accreditation. Foursquare (Jim Watson), 2020-08-13. https://foursquare.com/?p=7764 Accessed 2026-10-24.

[522] vistarmedia.com/measurement and vistarmedia.com/audience-targeting (HTTP 404). Vistar Media. https://www.vistarmedia.com/measurement Accessed 2026-10-24. Access: partial extract.

[523] Press Releases (InMarket newsroom index). inMarket Media, LLC, live index as read 2026-10-24. https://inmarket.com/press-releases/ Accessed 2026-10-24.

[524] Addressable Audiences, Data & Identity Solutions | Lotame (homepage). Lotame, 2026. https://www.lotame.com/ Accessed 2026-10-24.

[525] Verve | Let's make media better. (homepage). Verve Group, Inc., 2026. https://verve.com/ Accessed 2026-10-24.

[526] Samba TV (consumer site samba.tv). Samba TV, 2026. https://samba.tv/ Accessed 2026-10-24.

[527] Azira Privacy Policy (Last Updated August 06, 2025), with California Notice at Collection and Sensitive Data Policy. Azira LLC, 2025-08-06. https://azira.com/privacy-policy/ Accessed 2026-10-24.

[528] Foursquare Audience FAQ (campaign products documentation). Foursquare Labs, Inc., 2026-03-19. https://docs.foursquare.com/campaign-products/docs/audience-faq Accessed 2026-10-24.

[529] How to Use a Next-Generation Demand-Side Platform Powered by a Single Data Set to Boost Return on Ad Spend (Azira blog). The Azira Team, 2025-01-05. https://www.azira.com/blogs/how-to-use-a-next-generation-demand-side-platform-powered-by-single-data-set-to-boost-return-on-ad-spend Accessed 2026-10-24.

[530] Geofencing Solutions - Reveal. Reveal Mobile, Inc., 2026. https://revealmobile.com/geofencing/ Accessed 2026-10-24.

[531] Reveal | OOH Measurement & Attribution Platform (homepage). Reveal Mobile, Inc., 2026. https://revealmobile.com/ Accessed 2026-10-24.

[532] Blis Joins PMG's Alli Marketplace as Launch Partner (post). Blis, 2025-06-17. https://blis.com/blis-joins-pmgs-alli-marketplace-as-launch-partner-ushering-in-a-new-era-of-intelligence-led-advertising-access/ Accessed 2026-10-24.

[533] Our Ethos - Values & Mission | Blis. Blis, 2024-05-07; modified 2025-12-04. https://blis.com/our-ethos/ Accessed 2026-10-24.

[534] Sample Publishers by Channel (GroundTruth Help Center). GroundTruth, 2026-10-20. https://help.groundtruth.com/hc/en-us/articles/56579907879443-Sample-Publishers-by-Channel Accessed 2026-10-24.

[535] NAI Announces Voluntary Enhanced Standards to Safeguard Sensitive Locations (InMarket post). inMarket Media, LLC, 2024-10-30. https://inmarket.com/nai-announces-voluntary-enhanced-standards-to-safeguard-sensitive-locations/ Accessed 2026-10-24.

[536] What is Proximity?. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/campaign-products/docs/what-is-proximity Accessed 2026-10-24.

[537] Proximity FAQ. Foursquare, 2026-04-02 (page updatedAt). https://docs.foursquare.com/campaign-products/docs/proximity-faq Accessed 2026-10-24.

[538] Reach audiences with data-driven targeting | Broadsign (Broadsign Ads audiences). Broadsign, 2026. https://broadsign.com/broadsign-ads/audiences Accessed 2026-10-24.

[539] Adsquare | Vistar Media Partner. Vistar Media, undated (page footer (c) 2026). https://www.vistarmedia.com/partner-adsquare Accessed 2026-10-24.

[540] PlaceIQ | Vistar Media Partner. Vistar Media, undated (page footer (c) 2026). https://www.vistarmedia.com/partner-placeiq Accessed 2026-10-24.

[541] FTC Sues Kochava for Selling Data that Tracks People at Reproductive Health Clinics, Places of Worship, and Other Sensitive Locations. Federal Trade Commission (press release), 2022-08-29. https://www.ftc.gov/node/79559 Accessed 2026-10-24.

[542] Memorandum Decision and Order (Rule 11 sanctions denied; motion to unseal amended complaint granted), FTC v. Kochava Inc., Dkt. 63. US District Court for the District of Idaho, 2023-11-03. https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-1.pdf Accessed 2026-10-24.

[543] Memorandum Decision and Order (motion to dismiss granted with leave to amend), FTC v. Kochava Inc., Dkt. 24. US District Court for the District of Idaho, 2023-05-04. https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-0.pdf Accessed 2026-10-24.

[544] Memorandum Decision and Order (motion to dismiss First Amended Complaint denied), FTC v. Kochava Inc., Dkt. 71. US District Court for the District of Idaho, 2024-02-03. https://www.govinfo.gov/content/pkg/USCOURTS-idd-2_22-cv-00377/pdf/USCOURTS-idd-2_22-cv-00377-2.pdf Accessed 2026-10-24.

[545] Gravy Analytics data broker breach: trove of location data threatens privacy of millions. TechCrunch (Zack Whittaker), 2025-01-13. https://techcrunch.com/2025/01/13/gravy-analytics-data-broker-breach-trove-of-location-data-threatens-privacy-millions/ Accessed 2026-10-24.

[546] Attorney General Ken Paxton Sues Allstate and Arity for Unlawfully Collecting, Using, and Selling Over 45 Million Americans' Driving Data to Insurance Companies. Office of the Attorney General of Texas, 2025-01-13. https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-sues-allstate-and-arity-unlawfully-collecting-using-and-selling-over-45 Accessed 2026-10-24.

[547] Plaintiff's Original Petition, The State of Texas v. The Allstate Corporation et al. and Arity, LLC et al., District Court of Montgomery County, Texas. Office of the Attorney General of Texas (court filing), 2025-01-13. https://www.texasattorneygeneral.gov/sites/default/files/images/press/Allstate%20and%20Arity%20Petition%20Filed.pdf Accessed 2026-10-24.

[548] Attorney General Ken Paxton Sues General Motors for Unlawfully Collecting Drivers' Private Data and Selling It to Insurance Companies. Office of the Texas Attorney General, 2024-08-13. https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-sues-general-motors-unlawfully-collecting-drivers-private-data-and Accessed 2026-10-24.

[549] State of Texas v. General Motors LLC and OnStar LLC, Original Petition (cause 24-08-12392, Montgomery County). Office of the Texas Attorney General, 2024-08-13. https://www.texasattorneygeneral.gov/sites/default/files/images/press/General%20Motors%20Original%20Petition%20Filestamped.pdf Accessed 2026-10-24.

[550] General Motors Company Form 10-Q for the quarter ended 30 September 2026. General Motors Company (SEC EDGAR), 2026-10-20. https://www.sec.gov/Archives/edgar/data/1467858/000146785826000059/gm-20260930.htm Accessed 2026-10-24.

[551] Attorney General Griffin sues General Motors and OnStar for deceiving Arkansans and unlawfully selling data. Arkansas Attorney General, 2025-02-26. https://arkansasag.gov/news-release/attorney-general-griffin-sues-general-motors-and-onstar-for-deceiving-arkansans-and-unlawfully-selling-data/ Accessed 2026-10-24.

[552] Attorney General Mike Hilgers files lawsuit against General Motors for deceptive collection and sale of driver data. Nebraska Attorney General, 2025-07-08. https://ago.nebraska.gov/attorney-general-mike-hilgers-files-lawsuit-against-general-motors-deceptive-collection-and-sale Accessed 2026-10-24.

[553] Attorney General Brenna Bird sues General Motors for illegal business practices. Iowa Attorney General, 2026-02-26. https://www.iowaattorneygeneral.gov/newsroom/attorney-general-brenna-bird-sues-general-motors-for-illegal-business-practices Accessed 2026-10-24.

[554] When It Comes to Data Privacy, Consumers Must Be in the Driver's Seat: Attorney General Bonta announces settlement with General Motors. California Attorney General, 2026-05-08. https://oag.ca.gov/news/press-releases/when-it-comes-data-privacy-consumers-must-be-driver%E2%80%99s-seat-attorney-general Accessed 2026-10-24.

[555] Attorney General Bonta announces investigative sweep of location data industry. California Attorney General, 2025-03-10. https://oag.ca.gov/news/press-releases/attorney-general-bonta-announces-investigative-sweep-location-data-industry Accessed 2026-10-24.

[556] Attorney General Ellison secures $391.5 million settlement with Google over location tracking. Minnesota Attorney General, 2022-11-14. https://ag.state.mn.us/Office/Communications/2022/11/14_Google.asp Accessed 2026-10-24.

[557] Attorney General Bonta announces $93 million settlement regarding Google's location-privacy practices. California Attorney General, 2023-09-14. https://oag.ca.gov/news/press-releases/attorney-general-bonta-announces-93-million-settlement-regarding-google%E2%80%99s Accessed 2026-10-24.

[558] Attorney General Ken Paxton secures historic $1.375 billion settlement with Google related to Texans' data privacy. Office of the Texas Attorney General, 2025-05-09. https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-secures-historic-1375-billion-settlement-google-related-texans-data Accessed 2026-10-24.

[559] Attorney General Ken Paxton finalizes historic settlement with Google and secures $1.375 billion. Office of the Texas Attorney General, 2025-10-31. https://www.texasattorneygeneral.gov/news/releases/attorney-general-ken-paxton-finalizes-historic-settlement-google-and-secures-1375-billion-big-tech Accessed 2026-10-24.

[560] AG reaches settlement with advertising company prohibiting 'geofencing' around Massachusetts healthcare facilities. Massachusetts Attorney General, 2017-04-04. https://web.archive.org/web/20220505002332id_/https://www.mass.gov/news/ag-reaches-settlement-with-advertising-company-prohibiting-geofencing-around-massachusetts-healthcare-facilities Accessed 2026-10-24.

[561] Data Broker Registry. Texas Secretary of State. https://www.sos.texas.gov/statdoc/data-brokers.shtml Accessed 2026-10-24.

[562] Data Broker Registry. Oregon Division of Financial Regulation. https://dfr.oregon.gov/business/licensing/data-broker-registry/Pages/index.aspx Accessed 2026-10-24. Access: partial extract.

[563] 9 V.S.A. § 2446. Data brokers; annual registration. Vermont General Assembly. https://legislature.vermont.gov/statutes/section/09/062/02446 Accessed 2026-10-24.

[564] NAI Precise Location Information Solution Provider Voluntary Enhanced Standards (June 22, 2022 version). Network Advertising Initiative, 2022-06-22. https://thenai.org/accountability/precise-location-information-solution-provider-voluntary-enhanced-standards Accessed 2026-10-24.

[565] Leading Companies Commit to Privacy-First NAI Enhanced Standards for Precise Location Information. Network Advertising Initiative, 2022-06-22. https://thenai.org/press/leading-companies-commit-to-privacy-first-nai-enhanced-standards-for-precise-location-information/ Accessed 2026-10-24.

[566] Accredited Services (accreditation landing page). Media Rating Council, 2026. https://mediaratingcouncil.org/accreditation Accessed 2026-10-24.

[567] Media Rating Council Accreditation Decisions, July 2026 through September 2026. Media Rating Council, 2026-10-16. https://mediaratingcouncil.org/sites/default/files/News/Accreditation-Matters/MRC_Accreditation_Status_Quarterly_Update_Q3_2026.pdf Accessed 2026-10-24.

[568] Media Rating Council Accreditation Decisions, April 2026 through June 2026. Media Rating Council, 2026-07-13. https://mediaratingcouncil.org/sites/default/files/News/MRC%20Accreditation%20Status%20Quarterly%20Update_Q2%2026%207.14.26.pdf Accessed 2026-10-24.

[569] Foursquare 'Visits' Data Gains MRC Accreditation. MediaPost (Laurie Sullivan), 2020-08-13. https://www.mediapost.com/publications/article/354710/foursquare-visits-data-gains-mrc-accreditation.html Accessed 2026-10-24.

[570] MRC accreditation letter to Epsilon (continuing accreditation). Media Rating Council (George W. Ivie), 2025-10-01. https://mediaratingcouncil.org/sites/default/files/Accreditation%20Letters/Epsilon/GI100125_Epsilon%20Accreditation%20Letter.pdf Accessed 2026-10-24.

[571] Mobile Location Data Buyer's Guide (landing page). IAB Mobile Location Data Working Group, 2014-07-01. https://iab.com/insights/mobile-location-data-buyers-guide Accessed 2026-10-24. Access: partial extract.

[572] Digital Out-Of-Home (DOOH) Measurement Guide (landing page). IAB, 2025-07-08. https://www.iab.com/guidelines/dooh-measurement-guide/ Accessed 2026-10-24. Access: partial extract.

[573] Cracking the Code of DOOH: IAB's New Measurement Guide Sets a Standard for the Future (blog). IAB (Jason Adamski, Director, Media Center), 2025-07-08. https://www.iab.com/blog/cracking-the-code-of-dooh/ Accessed 2026-10-24.

[574] IAB has released a DOOH measurement guide - free access. Digital Signage Pulse, 2025-07-08. https://digitalsignagepulse.com/news/iab-has-released-a-dooh-measurement-guide/ Accessed 2026-10-24.

[575] Digital Out-Of-Home (DOOH) Measurement Guide. IAB, 2025-07. https://assets.contentstack.io/v3/assets/bltbeaed4aed52c223a/blt494e044b82a9dcb0/68cb0a53da7537f88855f47c/iab-dooh-measurement-guide.pdf Accessed 2026-10-24.

[576] MRC Out-of-Home Measurement Standards, Phase 1 & 2 Combined, Final December 2025. Media Rating Council, 2025-12. https://mediaratingcouncil.org/sites/default/files/Standards/MRC%20OOH%20Standards%20Combined_FINAL.pdf Accessed 2026-10-24.

[577] MRC Digital Place-Based Audience Measurement Standards, Version 1.0 (Final), March 15, 2017. Media Rating Council, 2017-03-15. https://mediaratingcouncil.org/sites/default/files/Standards/DPB%20Standards%20Version%201.0%20%28Final%29%203-15-17.pdf Accessed 2026-10-24.

[578] Geopath Media Measurement Methodology Overview (Confluence). Geopath, 2023-05-31. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323447831/Geopath+Media+Measurement+Methodology+Overview Accessed 2026-10-24. Access: partial extract.

[579] 2026 Preview Dataset Live in the Insights Suite and API. Geopath (blog), 2026-09-30. https://blog.geopath.org/2026/09/30/2026-preview-dataset-live-in-the-insights-suite-and-api/ Accessed 2026-10-24.

[580] Geopath homepage. Geopath, undated. https://geopath.org/ Accessed 2026-10-24.

[581] Our Organization. Geopath, undated. https://geopath.org/our-org/ Accessed 2026-10-24.

[582] Geopath OOH Measurement Methodology (Confluence index page). Geopath (Chris Luken, contributor), 2025-08-11. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323447817/Geopath+OOH+Measurement+Methodology Accessed 2026-10-24. Access: partial extract.

[583] Geopath Movement Patterns (Confluence). Geopath, 2023-05-31. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323449552 Accessed 2026-10-24. Access: partial extract.

[584] Geopath Audience Measures & API Outputs. Geopath, 2023-05-31. https://geopath.atlassian.net/wiki/spaces/GEOP/pages/2323447870/Geopath+Audience+Measures+API+Outputs Accessed 2026-10-24. Access: partial extract.

[585] What's Next for OOH Advertising in America (OOH Media Con session page). Ipsos, 2026. https://www.ipsos.com/en-us/whats-next-ooh-advertising-america Accessed 2026-10-24.

[586] OpenRTB 2.6 specification (main branch, 2.6.md). IAB Tech Lab, 2023-03. https://raw.githubusercontent.com/InteractiveAdvertisingBureau/openrtb2.x/main/2.6.md Accessed 2026-10-24.

[587] AdCOM Specification v1.0 (FINAL, main branch). IAB Tech Lab, 2022-03. https://raw.githubusercontent.com/InteractiveAdvertisingBureau/AdCOM/main/AdCOM%20v1.0%20FINAL.md Accessed 2026-10-24.

[588] WOO launches world-first global measure of Programmatic DOOH Advertising Spend. World Out of Home Organization (WOO); data aggregated by PricewaterhouseCoopers (PwC), 2026-06-05. https://www.worldooh.org/news/woo-global-pdooh-spend-study-2025 Accessed 2026-10-24.

[589] Can You Really Measure Footfall from DOOH Ads?. Perion (Perion Marketing), undated. https://perion.com/blog/can-you-really-measure-footfall-from-dooh-ads/ Accessed 2026-10-24.

[590] Perion homepage (hivestack.com redirects here). Perion, undated. https://perion.com/ Accessed 2026-10-24.

[591] Methodology. Route Research Limited, undated. https://route.org.uk/methodology.php Accessed 2026-10-24.

[592] Route homepage. Route Research Limited, undated. https://route.org.uk/ Accessed 2026-10-24.

[593] COMMB homepage content (Gatsby page-data JSON for the client-rendered site). Canadian Out-of-Home Marketing and Measurement Bureau (COMMB), undated. https://commb.ca/page-data/index/page-data.json Accessed 2026-10-24. Access: partial extract.

[594] Vistar Media Partners with Leading Data Providers for Comprehensive Measurement Suite for Digital Out-of-Home. Vistar Media (byline Kerry Coppinger), 2019-01-30. https://www.vistarmedia.com/news/data-providers-measurement-suite-dooh Accessed 2026-10-24.

[595] California Data Broker Registry: 2025 registry file (registry2025.csv), row xAd, Inc. (GroundTruth). California Privacy Protection Agency (CalPrivacy), 2025 (registry year). https://cppa.ca.gov/data_broker_registry/registry2025.csv Accessed 2026-10-24.

[596] California Data Broker Registry, 2024 registration CSV (registry2024.csv). California Privacy Protection Agency, 2024. https://cppa.ca.gov/data_broker_registry/registry2024.csv Accessed 2026-10-24.

[597] AdCP CHANGELOG.md at tag v3.2.6. adcontextprotocol (GitHub), 2026-10-23. https://raw.githubusercontent.com/adcontextprotocol/adcp/v3.2.6/CHANGELOG.md Accessed 2026-10-24.

[598] VIOOH's Seller Agent automated 100+ DOOH deals in H1 2026. PPC Land, 2026-06-27. https://ppc.land/vioohs-seller-agent-automated-100-dooh-deals-in-h1-2026/ Accessed 2026-10-24.

[599] Displayce launches agentic DOOH agents via MCP at Cannes Lions 2026. PPC Land, 2026-06-27. https://ppc.land/displayce-launches-agentic-dooh-agents-via-mcp-at-cannes-lions-2026/ Accessed 2026-10-24. Access: partial extract.

[600] MRC Issues Location Based Advertising Measurement Guidelines (press release). Media Rating Council (release hosted by IAB), 2017-03-29. https://www.iab.com/wp-content/uploads/2017/03/March2017-Location-Based-Guidelines-PR_final.pdf Accessed 2026-10-24.

[601] Standards & Guidelines (index page listing every MRC standard with issued and last-updated dates). Media Rating Council, 2026. https://mediaratingcouncil.org/standards-and-guidelines Accessed 2026-10-24.

[602] The First Agent-Native Composable Advertising Platform - Infillion (homepage). Infillion, 2026. https://infillion.com/ Accessed 2026-10-24.

[603] Some final words on Foursquare. Dennis Crowley (Substack), 2026-10-20. https://teendrama.substack.com/p/some-final-words-on-foursquare Accessed 2026-10-24.

[604] Good morning, hello, how are you (newsletter). Rick Webb (Buttondown), 2026-09-24. https://buttondown.com/rickwebb/archive/good-morning-hello-how-are-you-1820/ Accessed 2026-10-24.

[605] Notification of a personal data breach to the Norwegian Data Protection Authority (Datatilsynet), Gravy Analytics / Unacast (redacted). BakerHostetler on behalf of Gravy Analytics / Unacast (published by NRK via DocumentCloud), 2025-01. https://s3.documentcloud.org/documents/25485205/datatilsynet-unacast-security-incident-notification-redacted.pdf Accessed 2026-10-24.

[606] OOH Industry Advances Effort to Modernize Audience Measurement and Selects Ipsos for Pilot Program (press release, syndicated copy). Geopath and Out of Home Advertising Association of America (OAAA), via GlobeNewswire; copy on MarTech Edge, 2026-03-31. https://martechedge.com/press-release/ooh-industry-advances-effort-to-modernize-audience-measurement-and-selects-ipsos-for-pilot-program Accessed 2026-10-24.

[607] Geopath and OAAA Selects Ipsos for Audience Measurement Pilot Program. Billboard Insider (press release copy), 2026-04-06. https://billboardinsider.com/geopath-and-oaaa-selects-ipsos-for-audience-measurement-pilot-program/ Accessed 2026-10-24.

[608] Atlas (product page). Talon, undated. https://talonooh.com/products/atlas/ Accessed 2026-10-24.

[609] Arrivalist: Media Measurement & Travel Analytics (homepage). Arrivalist, 2026. https://www.arrivalist.com/ Accessed 2026-10-24.

[610] Adaptive Advertising Platform | Locala (homepage, asklocala.com). Locala (formerly S4M), 2026. https://asklocala.com/ Accessed 2026-10-24.

[611] Adsmovil — Connecting Brands Through Data and Culture (homepage). Adsmovil, 2026. https://www.adsmovil.com/ Accessed 2026-10-24.

[612] Real-World Media Platform | Programmatic DSP | Simpli.fi (homepage). Simpli.fi, 2026. https://simpli.fi/ Accessed 2026-10-24.

[613] Location Data Platform for Global POI Data | SafeGraph (homepage). SafeGraph Inc., 2026. https://www.safegraph.com/ Accessed 2026-10-24.

[614] Location Intelligence Company | Footfall Data Provider | Huq (homepage). huq Industries, 2026. https://huq.io/ Accessed 2026-10-24.

[615] The World's Smartest Location & Geospatial Company - Tamoco (homepage). Tamoco, 2026. https://www.tamoco.com/ Accessed 2026-10-24.

[616] PassBy - AI-powered Market Intelligence for Retail (homepage). Passby Technologies Ltd, 2026. https://passby.com/ Accessed 2026-10-24.

[617] Zartico (homepage). Zartico, Inc., 2026. https://www.zartico.com/ Accessed 2026-10-24.

[618] Demographic Mapping & Site Selection Software | ArcGIS Business Analyst. Esri, 2026. https://www.esri.com/en-us/arcgis/products/arcgis-business-analyst/overview Accessed 2026-10-24.

[619] Platform Overview | Location Data Platform | HERE Technologies. HERE, 2026. https://www.here.com/platform Accessed 2026-10-24.

[620] Mapbox | Maps, Navigation, Search, and Data (homepage). Mapbox, 2026. https://www.mapbox.com/ Accessed 2026-10-24.

[621] Mobvista: The Growth Hub for Developers (homepage). Mobvista Inc. (HKEX: 1860), 2026. https://www.mobvista.com/en/ Accessed 2026-10-24.

[622] PlaceIQ Product Privacy Notice. Precisely Software Incorporated, 2025-10-30. https://www.precisely.com/legal/placeiq-product-privacy-notice/ Accessed 2026-10-24.

[623] Advan Data Dictionaries (Weekly Patterns+ v2.8, Neighborhood Patterns+ v1.8 and other products). Advan Research Corporation, undated (live documentation; accessed 2026-10-24). https://docs.advanresearch.com/advan_data_dictionaries.md Accessed 2026-10-24.

[624] PlaceIQ Location Data (Precisely product page, Precisely Visits and Precisely Audiences sections). Precisely Software Incorporated, 2026 (footer copyright). https://www.precisely.com/product/precisely-placeiq/ Accessed 2026-10-24.

[625] Place Intelligence Metrics: getting started guide (Precisely white-paper landing page). Precisely Software Incorporated, 2026 (footer copyright). https://www.precisely.com/resource-center/whitepapers/place-intelligence-metrics-getting-started-guide Accessed 2026-10-24. Access: abstract only.

[626] Validated Footfall Data | Huq Data Methodology. huq Industries, 2026 (footer "© 2026 huq Industries"). https://huq.io/data-methodology Accessed 2026-10-24.

[627] Website Visitor Privacy Policy (last updated October 1, 2026). The Collective Data Solutions, LLC, 2026-10-01. https://www.collectivedata.io/website-visitor-privacy-policy/ Accessed 2026-10-24.

[628] Validating social spend: Visit Vermont achieves a sub-$50 Cost Per Visit for tourism. Adsquare, 2025-12-16. https://adsquare.com/resources/the-lab/visit-vermont-achieves-a-sub-50-cost-per-visit-for-tourism Accessed 2026-10-24.

[629] Adsquare Platform Privacy Policy. adsquare GmbH, 2026. https://www.adsquare.com/privacy/platform-privacy-policy Accessed 2026-10-24.

[630] Consumers Privacy Control (Arrivalist). Arrivalist Co., 2026-10-23 (Framer publish stamp of the site). https://www.arrivalist.com/consumer-privacy-control Accessed 2026-10-24.

[631] Azira | Data That's Clean, Current, and Compliant (platform data page). Azira LLC, 2026. https://www.azira.com/platform/data Accessed 2026-10-24.

[632] Dwell Time (Azira knowledge base article). Azira (Pinnacle Knowledge Base), unknown. https://knowledge.azira.com/dwell-time Accessed 2026-10-24. Access: partial extract.

[633] How Azira and Tambourine Measured Over 1 Thousand Property Visits for Lotte New York (case study). Azira LLC, 2026-02-06. https://www.azira.com/case-studies/how-azira-and-tambourine-measured-over-1-thousand-property-visits-for-lotte-new-york Accessed 2026-10-24.

[634] Azira | Unlock Real-World Insights Without Compromising Privacy (privacy page). Azira LLC, 2026. https://www.azira.com/platform/privacy Accessed 2026-10-24.

[635] Cuebiq Privacy Policy. Cuebiq Group, LLC, 2026-10-22. https://cuebiq.com/privacy-policy/ Accessed 2026-10-24.

[636] Cuebiq API (OpenAPI specification "Create and manage Campaigns and Geoset via Cuebiq external APIs", version 0.1.0). Cuebiq Group, LLC, undated (spec version 0.1.0; linked from the cuebiq.com footer "Documentation: Cuebiq API" via https://openapi.cuebiq.com/). https://api.cuebiq.com/openapi Accessed 2026-10-24.

[637] Footfall Measurement (Cuebiq measurement page). Cuebiq Group, LLC, undated (footer (c) 2026). https://cuebiq.com/measurement/ Accessed 2026-10-24.

[638] Privacy Center (Cuebiq). Cuebiq Group, LLC, undated (footer (c) 2026). https://cuebiq.com/privacy-center/ Accessed 2026-10-24.

[639] Forrester determines 500%+ ROI in Foursquare Attribution. Foursquare (reporting a Forrester Consulting TEI study), 2022-06-07. https://foursquare.com/resources/reports-and-insights/report/forrester-determines-500-roi-in-location-foursquare-attribution/ Accessed 2026-10-24. Access: partial extract.

[640] Proskin v. Foursquare Labs, Inc. — Class Action Complaint (Doc. 1). US District Court, D. Massachusetts (via CourtListener RECAP), 2023-07-24. https://storage.courtlistener.com/recap/gov.uscourts.mad.259540/gov.uscourts.mad.259540.1.0.pdf Accessed 2026-10-24.

[641] Market Share (GroundTruth Help Center). GroundTruth, 2025-11-15. https://help.groundtruth.com/api/v2/help_center/en-us/articles/37775905880083.json Accessed 2026-10-24.

[642] Groundtruth Reporting API, OpenAPI specification (external_api_spec.js) loaded by reporting.groundtruth.com/api. GroundTruth (xAd, Inc.), undated (live file). https://cdn.xad.com/external_api_spec.js Accessed 2026-10-24.

[643] DOOH Advertising | Outcomes Measurement | Kochava. Kochava Inc., undated (footer (c) 2008 - 2026). https://www.kochava.com/product/strategic-services/ooh-dooh-advertising-measurement/ Accessed 2026-10-24.

[644] Privacy Policy - Reveal. Reveal Mobile, Inc., 2024-11-08. https://revealmobile.com/privacy Accessed 2026-10-24.

[645] Vistar Media Marketplace DSP API OpenAPI specification. Vistar Media, undated. https://developers.vistarmedia.com/_spec/apis/dsp/index.yaml Accessed 2026-10-24.

[646] Driving footfall at scale: How Soreen tripled audience reach with ID-Precise Targeting (case study). adsquare GmbH, 2025-06-04. https://www.adsquare.com/resources/the-lab/tripled-audience-reach-with-id-precise-targeting Accessed 2026-10-24.

[647] Privacy Centre (Adsquare): Ten Principles, documents list and privacy request form. adsquare GmbH, 2025-05. https://www.adsquare.com/privacy Accessed 2026-10-24.

[648] Arity Audiences | Segmentation for auto insurance marketers. Arity, 2026-10-08. https://arity.com/products/arity-audiences/ Accessed 2026-10-24.

[649] Arity Privacy Statement (Effective date: June 17, 2026). Arity (Arity LLC, Arity 875, LLC, Arity Services, LLC), 2026-06-17. https://arity.com/privacy/ Accessed 2026-10-24.

[650] Azira Audiences | Build & Activate Location-Based Audiences. Azira LLC, 2026. https://www.azira.com/solutions/azira-audience Accessed 2026-10-24.

[651] Azira knowledge base (Pinnacle help centre) landing page. Azira LLC (hosted on Elevio), undated. https://knowledge.azira.com/ Accessed 2026-10-24. Access: access error.

[652] Execute Media Campaigns That Drive Real-World Outcomes (Azira Campaigns product page). Azira LLC, 2026. https://www.azira.com/solutions/azira-campaigns Accessed 2026-10-24.

[653] How Azira and Visit Fort Worth Increased Destination Awareness and Drove 8K+ Visits During a Competitive Travel Window (case study). Azira LLC, 2026-02-04. https://www.azira.com/case-studies/how-azira-and-visit-fort-worth-increased-destination-awareness-and-drove-8k-visits-during-a-competitive-travel-window Accessed 2026-10-24.

[654] TRUSTe privacy seal validation page for Azira LLC (Data Certification program). TrustArc (TRUSTe), undated (live page). https://privacy.trustarc.com/privacy-seal/validation?rid=79c90252-7a59-4f23-8681-9180af7a58de Accessed 2026-10-24.

[655] Blis Launches Blis AI – New Media Planning Tool Delivers Smarter Audiences in Seconds (release). Blis, 2025-06-12. https://blis.com/blis-launches-blis-ai-new-media-planning-tool-delivers-smarter-audiences-in-seconds/ Accessed 2026-10-24.

[656] Target Audiences with Location Intelligence | Cuebiq. Cuebiq Group, LLC, 2026-02-24. https://cuebiq.com/target-audiences/ Accessed 2026-10-24.

[657] TRUSTe privacy seal validation: Cuebiq Group, LLC (Data Certification program). TrustArc (TRUSTe), undated (live validation page). https://privacy.trustarc.com/privacy-seal/validation?rid=73914205-a1dc-4c37-aed9-6d1d1d9b3605 Accessed 2026-10-24.

[658] What is Audience?. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/campaign-products/docs/what-is-audience Accessed 2026-10-24.

[659] Build an Audience Segment (Foursquare Audience documentation). Foursquare Labs, Inc., 2026-03-19. https://docs.foursquare.com/campaign-products/docs/build-an-audience-segment Accessed 2026-10-24.

[660] Proskin v. Foursquare Labs, Inc. — Notice of Voluntary Dismissal Without Prejudice (Doc. 16). US District Court, D. Massachusetts (via CourtListener RECAP), 2023-10-26. https://storage.courtlistener.com/recap/gov.uscourts.mad.259540/gov.uscourts.mad.259540.16.0_1.pdf Accessed 2026-10-24.

[661] Overview of Third-Party Audiences (GroundTruth Help Center). GroundTruth, 2026-06-17. https://help.groundtruth.com/hc/en-us/articles/52692203951123-Overview-of-Third-Party-Audiences Accessed 2026-10-24.

[662] InMarket and Basis: Turning Retail Media Spend into Verifiable Outcomes. InMarket, 2026-05-27. https://inmarket.com/inmarket-and-basis-turning-retail-media-spend-into-verifiable-outcomes/ Accessed 2026-10-24.

[663] Mobilewalla Privacy Center (Business Services Privacy Notice). Mobilewalla, Inc., 2026-01-09. https://www.mobilewalla.com/privacy-center Accessed 2026-10-24.

[664] Location-Based Marketing (explainer page). Mobilewalla, 2023. https://www.mobilewalla.com/location-based-marketing Accessed 2026-10-24.

[665] Simpli.fi Services Privacy Policy (Effective Date: June, 2026). Simpli.fi (Simplifi Holdings), 2026-06. https://simpli.fi/services-privacy-policy Accessed 2026-10-24.

[666] Solutions for Marketers. Vistar Media, undated (footer (c) 2026). https://www.vistarmedia.com/solutions-for-marketers Accessed 2026-10-24.

[667] T-Mobile Advertising Solutions Unveils New Capabilities on the Heels of Vistar and Blis Acquisitions. T-Mobile US (newsroom), 2025-05-06. https://www.t-mobile.com/news/business/t-mobile-advertising-solutions-unveils-new-capabilities Accessed 2026-10-24.

[668] ADVAN Privacy (privacy program and policy). Advan Research Corporation, 2026-09-14. https://advanresearch.com/privacy Accessed 2026-10-24.

[669] Visits Overview (Foursquare Data Products documentation). Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/visits-overview Accessed 2026-10-24.

[670] Standard Visits Feed. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/standard-visits-feed Accessed 2026-10-24.

[671] Offline Conversions (Foursquare Visits for Offline Conversions). Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/data-products/docs/offline-conversions Accessed 2026-10-24.

[672] On demand mobility data with Mapbox Movement (Movement Data product page). Mapbox, undated (live page; accessed 2026-10-24). https://www.mapbox.com/movement-data Accessed 2026-10-24.

[673] Mapbox Movement: Guides (license, sample data access, production data delivery timelines). Mapbox, undated (live documentation; accessed 2026-10-24). https://docs.mapbox.com/data/movement/guides/ Accessed 2026-10-24.

[674] Data format (Mapbox Movement documentation). Mapbox, undated (live documentation; accessed 2026-10-24). https://docs.mapbox.com/data/movement/guides/data-format.md Accessed 2026-10-24.

[675] Activity index (Mapbox Movement documentation). Mapbox, undated (live documentation; accessed 2026-10-24). https://docs.mapbox.com/data/movement/guides/activity-index.md Accessed 2026-10-24.

[676] Our data - PassBy. Passby Technologies Ltd, undated (live page; accessed 2026-10-24). https://passby.com/our-data/ Accessed 2026-10-24.

[677] Our Data: Behind the Numbers. Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/resources/our-data Accessed 2026-10-24.

[678] Placer.ai Deepens Bloomberg Collaboration with Data Entitlements Offering. Placer Labs, Inc., 2026-02-17. https://www.placer.ai/anchor/articles/placer-ai-deepens-bloomberg-collaboration-with-data-entitlements-offering Accessed 2026-10-24.

[679] Firehose Visits Dataset (Precisely Data Guide). Precisely Software Incorporated, 2026 (footer copyright). https://www.precisely.com/data-guide/products/firehose-visits-dataset Accessed 2026-10-24.

[680] Daily Visit Count (PlaceIQ solution sheet). Precisely Software Incorporated, 2024-11 (file name 241104; "©2024 Precisely"). https://www.precisely.com/wp-content/uploads/2024/11/PS_Daily-Visit-Count_241104_E_final.pdf Accessed 2026-10-24.

[681] GPS (United States signal source) | Pine59 Console documentation. Pine59, undated (live docs). https://console.pine59.com/docs/datasets/usa/supply/gps-usa Accessed 2026-10-24.

[682] Places (United States catalog: Safegraph Places) | Pine59 Console documentation. Pine59, undated (live docs). https://console.pine59.com/docs/datasets/locations/points_of_interest Accessed 2026-10-24.

[683] FAQ | Pine59. Pine59, undated (© 2026). https://pine59.com/faq Accessed 2026-10-24.

[684] Geolocation Data (Zartico ZDOS data set page). Zartico, Inc., 2023-10-02 (dateModified in page metadata). https://www.zartico.com/geolocation-data Accessed 2026-10-24.

[685] Location Data Feeds | Complementics. ECONTEXT.AI LLC dba Complementics, 2026-10-24. https://complementics.com/location-feeds Accessed 2026-10-24.

[686] Privacy & Compliance | Complementics. ECONTEXT.AI LLC dba Complementics, 2026-10-24. https://complementics.com/privacy-compliance Accessed 2026-10-24.

[687] Data Quality | Complementics. ECONTEXT.AI LLC dba Complementics, 2026-10-24. https://complementics.com/data-quality Accessed 2026-10-24.

[688] Observation Panel - Outlogic. Outlogic, LLC, 2022-10-11. https://outlogic.io/observation-panel/ Accessed 2026-10-24.

[689] Quadrant | Insightful, accurate, and reliable mobility data (Mobile Location Data product page). Quadrant, an Appen company, 2026. https://www.quadrant.io/mobile-location-data Accessed 2026-10-24.

[690] Technical Docs | Company Overview. Quadrant, an Appen company, 2026. https://www.quadrant.io/technical-docs-company-overview Accessed 2026-10-24.

[691] Quadrant | Technical Documentation - Delivery Options. Quadrant, an Appen company, 2026. https://www.quadrant.io/data-delivery-options Accessed 2026-10-24.

[692] Appen inks agreement to acquire Quadrant. iTWire, 2021-08-27. https://itwire.com/it-industry-news/listed-tech/appen-inks-agreement-to-acquire-quadrant Accessed 2026-10-24.

[693] Geospatial Data - Feed, Real-time, Batched | High Accuracy - Tamoco. Tamoco, 2021-08-02. https://www.tamoco.com/geospatial-raw-location-data/ Accessed 2026-10-24.

[694] Our Data | Tamoco. Tamoco, 2022-12-05. https://www.tamoco.com/our-data/ Accessed 2026-10-24.

[695] Terms of Service (page also carries Unacast privacy statements dated 12 March 2020 and 30 August 2021). Unacast, Inc., 2026-01 ("Last Updated: January 2026"). https://www.unacast.com/terms-service Accessed 2026-10-24.

[696] Movement | Veraset (dataset product page). Veraset, 2026-10-20. https://www.veraset.com/datasets/movement Accessed 2026-10-24.

[697] Movement - Documentation | Veraset Documentation (with embedded Movement Schema, October 2024, PDF). Veraset, 2024-10. https://www.veraset.com/resource/doc/datasets-schema/docs Accessed 2026-10-24.

[698] Our data | Veraset. Veraset, 2026-10-24. https://www.veraset.com/about/our-data Accessed 2026-10-24.

[699] Delivery Methods & Compatibility - Documentation | Veraset. Veraset, 2026-10-24. https://www.veraset.com/resource/doc/data-access-and-evaluation/delivery-methods-compatibility---documentation-veraset Accessed 2026-10-24.

[700] API: Overview - Documentation | Veraset. Veraset, 2026-10-24. https://www.veraset.com/resource/doc/data-access-and-evaluation/api-overview---documentation-veraset Accessed 2026-10-24.

[701] Rezolve - Order Orchestration & Fulfillment Documentation (introduction). Rezolve AI plc (Bluedot docs at docs.bluedot.io), undated (live docs). https://docs.bluedot.io/ Accessed 2026-10-24.

[702] Point SDK Overview. Rezolve AI plc (Bluedot), undated (live docs). https://docs.bluedot.io/Point%20SDK/Overview Accessed 2026-10-24.

[703] Web SDK Overview. Rezolve AI plc (Bluedot), undated (live docs). https://docs.bluedot.io/Web%20SDK/Overview Accessed 2026-10-24.

[704] Point SDK Android: Geo-triggering. Rezolve AI plc (Bluedot), undated (live docs). https://docs.bluedot.io/Point%20SDK/Android/Geo-triggering Accessed 2026-10-24.

[705] Version Release Notes. Rezolve AI plc (Bluedot), 2026-09-23 (latest entry). https://docs.bluedot.io/Version%20release%20notes Accessed 2026-10-24.

[706] Integrations Overview. Rezolve AI plc (Bluedot), undated (live docs). https://docs.bluedot.io/integrations/Overview Accessed 2026-10-24.

[707] Regions URLs. Rezolve AI plc (Bluedot), undated (live docs). https://docs.bluedot.io/Regions%20URLs Accessed 2026-10-24.

[708] Rezolve Privacy Policy. Rezolve AI plc, unknown. https://rezolve.com/privacy-policy/ Accessed 2026-10-24.

[709] Movement SDK Overview. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/developer/docs/movement-sdk-overview Accessed 2026-10-24.

[710] Movement SDK Data Usage. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/developer/docs/movement-sdk-data-usage Accessed 2026-10-24.

[711] Movement SDK Feature Breakdown. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/developer/docs/movement-sdk-feature-breakdown Accessed 2026-10-24.

[712] Set Up Visit Detection Notifications & Criteria (Movement SDK, Developer Console). Foursquare Labs, Inc., undated (page 'Updated 8 months ago', about 2026-02). https://docs.foursquare.com/developer/docs/set-up-visit-detection-notifications-criteria Accessed 2026-10-24.

[713] Movement SDK 3rd Party Integrations. Foursquare, 2026-02-12 (page updatedAt). https://docs.foursquare.com/developer/docs/movement-sdk-third-party-integrations Accessed 2026-10-24.

[714] Release Notes - iOS Movement SDK. Foursquare, 2026-02-12 (page updatedAt); release entries undated. https://docs.foursquare.com/developer/docs/movement-sdk-ios-release-notes Accessed 2026-10-24.

[715] Gimbal Documentation (overview). Gimbal, Inc. (Infillion), undated (footer © 2014-2021). https://docs.gimbal.com/ Accessed 2026-10-24.

[716] Gimbal Android Version 4 API Developer Guide. Gimbal, Inc. (Infillion), undated (names SDK 4.2.0 as current; footer © 2014-2021). https://docs.gimbal.com/android/v4/devguide.html Accessed 2026-10-24.

[717] Gimbal Manager REST API. Gimbal, Inc. (Infillion), undated (footer © 2014-2021). https://docs.gimbal.com/rest.html Accessed 2026-10-24.

[718] Infillion Privacy Policy. Infillion, undated. https://infillion.com/privacy-policy/ Accessed 2026-10-24.

[719] Blis Strikes Exclusive Deal with Virgin Media O2 Business to Enhance Advertising Offering (release). Blis, 2024-10-28 (page; dateline 'LONDON, UK – October 24, 2024'). https://blis.com/blis-strikes-exclusive-deal-with-virgin-media-o2-business-to-enhance-advertising/ Accessed 2026-10-24.

[720] T-Mobile Acquires Blis to Expand Advertising Capabilities and Accelerate Growth (CEO post). Blis (Greg Isbister, CEO), 2025-03-05. https://blis.com/t-mobile-acquires-blis-to-expand-advertising-capabilities-and-accelerate-growth/ Accessed 2026-10-24.

[721] xAd: The Foundation of GroundTruth Tech. GroundTruth, 2026-06-15 (article:modified_time). https://www.groundtruth.com/xad/ Accessed 2026-10-24.

[722] InMarket Revolutionizes Advertising Measurement with AI-Powered "Unified Measurement" (company news post). inMarket Media, LLC, 2025-07-29. https://inmarket.com/inmarket-revolutionizes-advertising-measurement-with-ai-powered-unified-measurement/ Accessed 2026-10-24.

[723] About Us (leadership and board). inMarket Media, LLC, undated (as read 2026-10-24). https://inmarket.com/about-us/ Accessed 2026-10-24.

[724] About Us. Placer Labs, Inc., 2026 (build stamp 2026-10-22). https://www.placer.ai/company/about-us Accessed 2026-10-24.

[725] EDGAR submissions record, Placer Labs Inc. (CIK 0001786302). US Securities and Exchange Commission, live record (accessed 2026-10-24). https://data.sec.gov/submissions/CIK0001786302.json Accessed 2026-10-24.

[726] Form D, Placer Labs Inc. (filed 2022-01-14). Placer Labs Inc. (filer), SEC EDGAR, 2022-01-14. https://www.sec.gov/Archives/edgar/data/1786302/000178630222000001/primary_doc.xml Accessed 2026-10-24.

[727] Form D, Placer Labs Inc. (filed 2022-07-01). Placer Labs Inc. (filer), SEC EDGAR, 2022-07-01. https://www.sec.gov/Archives/edgar/data/1786302/000178630222000002/primary_doc.xml Accessed 2026-10-24.

[728] Form D, Notice of Exempt Offering of Securities, Placer Labs Inc. (filed 2024-07-23). Placer Labs Inc. (filer), US Securities and Exchange Commission (EDGAR), 2024-07-23. https://www.sec.gov/Archives/edgar/data/1786302/000178630224000003/primary_doc.xml Accessed 2026-10-24.

[729] Geofencing Company Radar Raises $55 Million. Location Business News, 2022-02-17. https://locationbusinessnews.com/geofencing-company-radar-raises-55-million Accessed 2026-10-24.

[730] Radar Raises $55 Million Series C to Drive Next-Gen Location-Aware CX. Gabrielle Rodgers, CMSWire, 2022-02-22. https://www.cmswire.com/customer-experience/radar-raises-55-million-series-c-to-drive-next-gen-location-aware-cx/ Accessed 2026-10-24.

[731] From location infrastructure to location intelligence: Radar’s latest brand refresh (blog). Audrey Na, Radar Labs, Inc., 2026-08-18. https://radar.com/blog/from-location-infrastructure-to-location-intelligence-radars-latest-brand-refresh Accessed 2026-10-24.

[732] Form D, Gravy Analytics, Inc. (CIK 1526946), accession 0001526946-22-000002. Gravy Analytics, Inc. (SEC EDGAR), 2022-01-06. https://www.sec.gov/Archives/edgar/data/1526946/000152694622000002/primary_doc.xml Accessed 2026-10-24.

[733] Enhetsregisteret record: UNACAST AS (org. no. 913820746). Brønnøysund Register Centre (Norway), live record. https://data.brreg.no/enhetsregisteret/api/enheter/913820746 Accessed 2026-10-24.

[734] Form D, Unacast, Inc. (CIK 1674974), accession 0001674974-16-000002. Unacast, Inc. (SEC EDGAR), 2016-06-09. https://www.sec.gov/Archives/edgar/data/1674974/000167497416000002/primary_doc.xml Accessed 2026-10-24.