How an out-of-home impression is counted, priced and proved.

Delivery is recorded by the ad player itself, transmitted over standard VAST tracking events, and signed at the edge so any figure in this report can be verified independently.

• Google IMA SDK is the player runtime on the screen. The same SDK used across digital video renders the spot and fires the tracking beacons — nothing proprietary sits in the counting path.
• Standard VAST events carry the record: start, firstQuartile, midpoint, thirdQuartile, complete. Only complete is billable.
• An Ed25519 signature is applied to every completed play at the edge, verifiable against a published public key by any third party.
• One completed play is one impression. The impression multiplier is 1.000 — no audience factor is applied to the billable count, so publisher and platform are counting the same event rather than two different quantities.

Completion rate is measured rather than assumed. The 95.37% start-to-complete rate is the observed rate across this footprint, consistent with the trailing fourteen-day baseline.

The flight closed on schedule at 100.9% of contracted volume. Delivery tracks taxi ridership rather than a building's opening hours: Friday is the peak and Sunday the trough, a 25% swing, with the working week sitting between them.

Every screen returns a structured read of the cabin in front of it every ten seconds. Attention is reported per play, from those reads — not inferred from screen position, traffic counts or venue class. 93.1% of reads show a single- occupant cabin, averaging 1.08 faces in frame.

Composition is observed across the delivered footprint over the flight window, classified per person per scene window. Every cell clears a minimum cohort size of five before release.

Reach is measured from device observations at the screen and deduplicated across the whole footprint and the whole flight. It is a census count, not a projection from a traffic panel.

Counts were matched on play identifier rather than on aggregate totals, so a variance can be traced to specific plays instead of argued in the abstract.

The MRC tolerance for programmatic delivery reconciliation is ±2.0%. Residual variance here is one sixteenth of that, so no true-up was applied.

Exclusions are applied before an impression becomes billable, not deducted afterwards — so no invoice is issued and later credited.

The footprint is a single venue type — OpenOOH 103, taxi & rideshare, all 2,966 screens mobile and in-cab — so rate is set by daypart, following measured attention and ridership rather than venue class. Rates were fixed for the flight term and agreed before launch.

Venue taxonomy follows OpenOOH v1.2. Every screen in this flight carries venue type 103 — taxi & rideshare — so the rate card varies by time of day, not by venue.

Exposed devices were matched against a control cohort drawn from the same markets and balanced on market, daypart profile and observed mobility. Visitation is measured against 2,847 Chick-fil-A locations inside the flight footprint.

Attribution window is 14 days from exposure. The control cohort is held out for the full flight term at 29.9% of exposed volume.

Player runtime

Google IMA SDK. The counting path uses the same SDK and the same event vocabulary as digital video.

Billable event

The VAST complete event for a rendered spot. Quartile events are reported but never billed.

Signature

Ed25519, applied per completed play at the edge. Algorithm, key identifier and verification steps are published in the methodology document.

Deduplication

Each play carries a deduplication key covering screen, creative and time window. Repeat submissions collapse to one billable record before invoicing.

Invalid traffic

General invalid traffic is filtered pre-billing. Filtered events are retained and inspectable at play-identifier level.

Obstructed view

Plays where the sensor confirmed the display was blocked from the viewing area are excluded from billing entirely.

Reach basis

Device observations at the screen, deduplicated across footprint and flight. Census, not panel-projected.

Attention basis

Structured scene reads at ten-second intervals, classified per person per window and joined to the play on screen and time.

Privacy floor

No cohort below five is released in any cell of this report. Device identifiers are salted per counterparty and cannot be joined across buyers.

Raw log access

Play-level detail behind every row is available to the buyer and to an independent auditor on request, keyed by play identifier.