RFC: declare the DOOH impression basis — typed tier, effective multiplier, and derivation link, at delivery and discovery

[bokelley]

Summary

A DOOH impressions figure is never an observed event. It is plays × audience multiplier, where the multiplier comes from a measurement currency, a traffic study, the seller's own model, or — increasingly — an on-site sensor. AdCP already knows this (plays is documented as "raw play count before any impression multiplier is applied"; dooh_metrics.calculation_notes exists to explain the multiplier in prose) but gives a buyer no typed way to learn which kind of number impressions is. Two DOOH products at the same CPM — one on a 2019 traffic study, one on per-play sensor counts — are different products, and today they are indistinguishable on the wire.

Classic OOH already solved this for itself. ooh_metrics.estimation_basis (experimental in 3.2) is a closed tier enum, currency_measured | seller_modeled, with provider identity on the row-level measurement_source slug and a schema dependencies rule that makes the basis mandatory whenever the estimate is reported. This RFC extends that pattern to DOOH, adds the tier that sensors make possible, and exposes the basis at discovery time so buyers can compare before they buy.

Split out of #7150 (§3) per the discussion there; #7177 deliberately did not carry it. Supersedes the audience_multiplier line item from #5537, which closed without it.

What exists today

Surface	DOOH	Classic OOH
Raw unit	plays (top-level)	none — static units have no play event
Delivery number	impressions (top-level, modelled)	ooh_metrics.estimated_impressions
Basis tier	none	ooh_metrics.estimation_basis (currency_measured / seller_modeled), required when estimated_impressions present
Provider identity	row-level measurement_source slug (shared)	same
Methodology prose	dooh_metrics.calculation_notes	ooh_metrics.calculation_notes
Discovery-time declaration	dooh_placement_attributes (slot/loop duration, resolution, motion) — no basis	n/a
Product-level vendors	delivery_measurement.vendors[] + deprecated provider + notes prose	same

Proposal

1. Shared basis enum — enums/impression-estimation-basis.json

Hoist ooh_metrics.estimation_basis's inline enum to a shared file and add the tiers DOOH needs:

Value	Meaning	Provider identity
currency_measured	An audience-measurement currency's estimate for the unit and period (Geopath, Route, COMMB, MOVE, and their DOOH products)	measurement_source slug
seller_modeled	The seller's own model — traffic study, venue-attendance factor, historical footfall. The honest fallback.	the seller
sensor_measured	Audience derived from an on-site instrument observing presence per play — camera, WiFi/BLE, radar, turnstile, ticketing. Measured, not modelled, though the instrument has its own error.	the vendor_metric_values[] row it derives from (see basis_metric)
hybrid	A model calibrated by sensors: sensor counts on a sample of screens scale a currency or seller model across the network	measurement_source and/or basis_metric

Tier, not provider, not instrument. Currencies change and instruments are an open corpus (DR-0004) — both stay as data (measurement_source, the vendor catalog's methodology_url), not enum values. Four closed tiers is the same shape the OOH block already committed to.

2. dooh_metrics.impression_basis (delivery)

{
  "plays": 964984,
  "impressions": 1481441,
  "dooh_metrics": {
    "impression_basis": "sensor_measured",
    "audience_multiplier": 1.535,
    "basis_metric": { "vendor": { "domain": "dooh-seller.example" }, "metric_id": "people_in_frame" },
    "loop_plays": 964984,
    "screens_used": 1894,
    "calculation_notes": "Per-play detected-person counts; screens with <10 plays fall back to venue model."
  },
  "vendor_metric_values": [
    { "vendor": { "domain": "dooh-seller.example" }, "metric_id": "people_in_frame",
      "value": 1481441, "unit": "persons", "measurable_plays": 964984, "vendor_relationship": "first_party" }
  ]
}

• impression_basis — the tier. SHOULD be present whenever a DOOH row reports impressions; proposed as a sales-dooh storyboard requirement once stable (mirrors OOH's dependencies rule, expressed on the parent since DOOH impressions is top-level).
• audience_multiplier (number ≥ 0) — the effective impressions / plays for the row, declared, not inferred. Lets a verifier check impressions ≈ plays × audience_multiplier within tolerance and lets a buyer compare rows. For sensor_measured it is the realised ratio for the period, which will vary by row — that is the point.
• basis_metric (optional { vendor, metric_id }) — when sensor_measured or hybrid, the vendor_metric_values[] row the audience figure derives from. This turns "impressions came from our cameras" into a join: the row carries coverage (measurable_plays), the seller–vendor relationship (vendor_relationship, feat(schema): typed vendor_relationship and play-based coverage denominators for vendor metrics #7177), and, via the vendor catalog, methodology_url / methodology_version / accreditations[]. No methodology prose is duplicated here.

venue_breakdown[] rows MAY carry impression_basis and audience_multiplier when the basis differs by venue (a mixed network: currency-measured roadside, sensor-measured retail).

3. dooh_placement_attributes.impression_basis (discovery)

Same enum, on the placement, with the same product-level-default / publisher-inheritance rule the block already uses for slot_duration_seconds. Optional audience_multiplier here is the planning value (what the seller's forecast will use). This is the half that changes buying: a buyer agent can filter or weight products by basis before create_media_buy, and get_products can grow a required_impression_basis filter later if demand appears (same filter-not-fail convention as required_vendor_metrics).

4. Forecast symmetry

delivery-forecast points already carry measurement_source and plays; add optional impression_basis so forecast and delivery reconcile on the same declaration, the way measurement_source already does.

5. Reuse, not duplication

• ooh_metrics.estimation_basis switches its inline enum to the shared file. Renaming the OOH field to impression_basis is available while the block is experimental (Editorial-class under the experimental downgrade) — recommended for symmetry, WG's call.
• delivery_measurement.notes and calculation_notes stay as row-specific prose. General methodology remains at the measurement catalog.

Non-goals

• No new pricing_model. Billing stays CPM; what changes is that the CPM's basis is legible. A sensor_measured CPM and a seller_modeled CPM are different products at the same rate, and that is now visible.
• No attention metric promotion. People-present and people-looking remain vendor_metric_values[] rows (RFC: verified attention as a reportable vendor metric #7150 §5).
• No instrument or vendor enum (DR-0004).

Open questions for the WG

1. Is hybrid load-bearing, or does every hybrid collapse to "whichever tier the majority of impressions came from" plus calculation_notes? Place-based audio (RFC: Support place-based audio as DOOH inventory, not screen-only or generic streaming audio #5878) is the test case: presence sensing exists but is sparse.
2. Should impression_basis be required for the sales-dooh claim at 3.3, or SHOULD-only for one release? OOH chose required-by-dependency from day one.
3. Is a declared audience_multiplier acceptable to sellers whose multipliers are contractual with a currency provider? If not, the tier alone still carries most of the value.
4. Rename ooh_metrics.estimation_basis → impression_basis while experimental, or keep both names and share only the enum?

Release target

3.3. 3.2 is at rc.1 with a feature already landing in that candidate; this is a new normative surface the WG has not seen, and the OOH field it mirrors is itself experimental until 3.3. Interim path for sellers is unchanged: calculation_notes prose plus, for sensor networks, vendor_metric_values[] with vendor_relationship: "first_party" and measurable_plays (#7177, rc.1).

Related

• RFC: verified attention as a reportable vendor metric #7150 §3 — the reframe this implements; feat(schema): typed vendor_relationship and play-based coverage denominators for vendor metrics #7177 — provenance + play denominators this builds on
• DOOH: Add structured selling-unit fields to placements (slot_duration, share_of_time, screen_resolution, delivery_unit) #5537 — audience_multiplier proposed on Placement, closed without it
• RFC: Support place-based audio as DOOH inventory, not screen-only or generic streaming audio #5878 — place-based audio as DOOH inventory (the no-screen, sparse-sensor case)
• WG consultation: seller obligation on vendor methodology_version pins (append-only on change) — signal, notification, acceptance #7235 — vendor methodology-version obligation (applies to basis_metric vendors)
• DR-0004 — no open-corpus enums; DR-0009 / Policy: closed (additionalProperties: false) row schemas vs DR-0009's open-by-default — decide once #7224 — dooh_metrics is open, so these are patch-shaped additions structurally, minor by policy

cc @snehdhruv (Trillboards — the sensor_measured case), @geneticallymodifiedfoodforthought (place-based audio)

[bokelley]

Triage

Classification: Feature request — RFC
Bucket(s): spec / protocol, media-buy
Status: ready-for-human

What the experts said:

• ad-tech-protocol-expert: Converge with conditions — 4-tier taxonomy maps to real procurement distinctions (Quividi/AIM = sensor_measured, Geopath/Route = currency_measured, traffic-study operators = seller_modeled); schema additions are cleanly non-breaking on dooh_metrics (additionalProperties: true); hybrid needs an explicit scoping rule so Geopath (internally sensor-calibrated) stays currency_measured, not hybrid; basis_metric pointer into vendor_metric_values[] must carry a MUST in prose (JSON Schema can't enforce cross-array existence); confirm no production estimation_basis emissions before renaming the OOH field.
• adtech-product-expert: Converge — discovery-time dooh_placement_attributes.impression_basis is the workflow change that matters for buying; audience_multiplier needs tier-conditional SHOULD (Geopath/Route reseller agreements contractually prohibit multiplier disclosure); SHOULD at 3.3 is right with a committed 3.4 hardening date written into the RFC before it closes; keep hybrid but require calculation_notes whenever declared.

My take: Both experts converge on shipping 3.3. Two amendments need WG resolution before the schema PR opens: a hybrid scoping rule (Q1) preventing inconsistent Geopath classification from breaking buyer pre-buy filtering on the most liquid DOOH segment; and tier-conditional SHOULD on audience_multiplier (Q3) to avoid suppressing adoption from currency-backed sellers whose reseller agreements prohibit multiplier disclosure.

---

Decision memo (provisional DR-0010, for WG ratification)

Ruling: DOOH impression-estimation basis is a closed four-tier enum (currency_measured, seller_modeled, sensor_measured, hybrid) declared at delivery (dooh_metrics) and discovery (dooh_placement_attributes); hybrid is restricted to seller-operated calibrated models and never to currency systems' internal sensor use; audience_multiplier SHOULD is tier-conditional, not universal.

Class: normative
Refs: #7253, #7150, #7177

---

Q1 — hybrid boundary (critical — must resolve before schema PR)

Geopath uses cameras to calibrate its panel model, but the output is a published currency figure the buyer accesses directly. Without an explicit scoping rule, the same inventory is currency_measured from one seller and hybrid from another — buyer filtering on dooh_placement_attributes.impression_basis returns contradictory results for equivalent inventory.

Option A — currency-access rule (expert recommendation):

  "currency_measured": "Published currency figure from Geopath, Route, COMMB, MOVE.
+   Applies regardless of whether the currency system internally uses sensors to
+   calibrate its model. Provider named in measurement_source."
  "hybrid": "A seller-owned model calibrated by seller-operated sensors — e.g.,
+   camera counts on a subset of the seller's screens scale a seller-authored
+   traffic study across a network. Excludes impressions drawn from a published
+   currency figure."

Option B — drop hybrid, simplify to three tiers:
Remove hybrid; sellers declare seller_modeled + mandatory calculation_notes for sensor-calibrated rows. Place-based audio (#5878) becomes seller_modeled + notes.

---

Q3 — audience_multiplier conditionality (high — blocks currency-seller adoption)

Geopath and Route reseller agreements often prohibit multiplier disclosure contractually. Universal SHOULD will suppress adoption from the premium currency_measured segment — the buyers who most want the tier signal will find it absent on the inventory they most want to filter.

Option A — tier-conditional SHOULD (expert recommendation):

// sensor_measured or hybrid — SHOULD include audience_multiplier
{
  "impression_basis": "sensor_measured",
  "audience_multiplier": 1.535,
  "basis_metric": { "vendor": { "domain": "dooh-seller.example" }, "metric_id": "people_in_frame" }
}

// currency_measured — audience_multiplier OPTIONAL; tier alone signals methodology
{
  "impression_basis": "currency_measured"
}

Option B — universal SHOULD (as written):
Sellers with contractual constraints declare impression_basis and omit the multiplier — conformant but weakens verifiability for the currency segment.

---

Q2 — SHOULD → required hardening date
SHOULD at 3.3 is correct given sensor-network diversity. WG should commit the 3.4 required date in this RFC now; "SHOULD without a sunset becomes permanent."

Q4 — ooh_metrics field rename
Rename ooh_metrics.estimation_basis → impression_basis only after confirming no production emissions on x-status: experimental surface. If any seller is live with the 3.2 experimental field, carry both names for one minor with a deprecation notice.

Recommendation: Adopt the proposal as specified; apply Option A for Q1 and Q3 in the schema PR; write the 3.4 hardening date into the RFC; basis_metric → vendor_metric_values[] cross-reference must be MUST in prose; rename estimation_basis only after production-emission check.

Dissent: None from expert consultation.

---

Triaged by Claude Code. Session: https://claude.ai/code/session_01NFC9Qx1A77fCMgNnPYqucj

---

Generated by Claude Code