Proposal: AdCP Reference Media Ontology — a shared vocabulary for agentic advertising (3.1)

[mikulbhatt]

Summary

AdCP 3.0 provides taxonomies for specific operational purposes: channels for budget allocation, industries for brand classification, content topics for contextual matching. But there is no unified vocabulary for how agents describe audiences, publisher quality, campaign objectives, or the relationships between these concepts. This proposal defines an AdCP Reference Media Ontology — a shared, extensible vocabulary that agents can use to declare capabilities, express intent, and verify interpretation across the protocol. It provides the concrete vocabulary that the taxonomy declaration, semantic fidelity, and decision provenance proposals require to be practical.

From human negotiation to machine negotiation

When two media planners have worked together for years, they share vocabulary implicitly. "Auto intenders" means something specific. "Premium publishers" has a shared definition shaped by three years of campaign history. "Outdoor enthusiasts" is distinct from "sports fans" because they've had that conversation before.

When agents negotiate, there are no years of shared context. Every transaction starts cold. The buyer agent says "outdoor enthusiasts" and the seller agent maps it to whatever segment ID its system has. If there is no shared vocabulary — no common reference point that both sides can use — then every mapping is a guess, and every semantic fidelity finding is "the seller mapped your term to their term, but we don't know if that's right because there's no common frame of reference."

Anyone who has built an audience taxonomy at scale has seen this firsthand. You create a segment called "outdoor recreation" and three different demand partners map it to three different things: one maps it to a broad "sports" category, one maps it to a narrow "camping" category, and one maps it to their proprietary "active lifestyle" cluster. All three mappings are defensible. None of them are verifiable without a shared reference frame.

A reference ontology is that common frame of reference. It does not replace seller-specific or buyer-specific vocabularies. It provides a lingua franca that both sides can map TO — so governance can evaluate whether the mapping was reasonable.

The vocabulary stack

The reference ontology fits into AdCP's existing architecture as an integration layer:

┌─────────────────────────────────────────────────────────────┐
│  AdCP Reference Media Ontology (NEW — this proposal)        │
│  Concept relationships, cross-taxonomy mapping,             │
│  audience/publisher/objective vocabulary                     │
├─────────────────────────────────────────────────────────────┤
│  IAB Content Taxonomy 3.0    │  IAB Audience Taxonomy 1.1  │
│  (TMP context signals)       │  (industry standard)        │
├─────────────────────────────────────────────────────────────┤
│  AdCP Media Channel Taxonomy │  AdCP Advertiser Industry   │
│  (20 channels)               │  (73 categories)            │
├─────────────────────────────────────────────────────────────┤
│  AdCP Protocol Layer (get_adcp_capabilities, sync_plans,    │
│  check_governance, get_products, create_media_buy)          │
└─────────────────────────────────────────────────────────────┘

The bottom two layers exist today. The IAB taxonomies are referenced but not integrated. The reference ontology provides the integration layer — connecting channels to audiences to content to objectives with explicit, machine-readable relationships.

This is the difference between having a pile of dictionaries in different languages and having a translation table between them. The dictionaries are useful on their own. The translation table is what makes them interoperable.

What the ontology covers

The reference ontology defines five vocabulary domains. Each domain provides standardized concepts with explicit relationships — the relationships are what make this an ontology rather than just another flat taxonomy.

1. Audience concept vocabulary

Standardized terms for how buyers describe target audiences, with explicit relationships between concepts:

{
  "concept_id": "outdoor_recreation",
  "domain": "audience",
  "label": "Outdoor Recreation Enthusiasts",
  "description": "Consumers with demonstrated interest in hiking, camping, trail running, fishing, and related outdoor activities",
  "relationships": [
    { "type": "is_a", "target": "active_lifestyle" },
    { "type": "overlaps_with", "target": "extreme_sports", "overlap_estimate": 0.3 },
    { "type": "distinct_from", "target": "general_sports_fans" },
    { "type": "broader_than", "target": "hiking_enthusiasts" },
    { "type": "narrower_than", "target": "active_lifestyle" }
  ],
  "iab_audience_mapping": {
    "taxonomy": "iab-audience-1.1",
    "primary": "Sports & Recreation",
    "note": "IAB taxonomy does not distinguish outdoor recreation from general sports at this hierarchy level"
  },
  "suggested_signals": ["purchase_intent.outdoor_gear", "content_affinity.outdoor_media"]
}

{
  "concept_id": "auto_intenders",
  "domain": "audience",
  "label": "Auto Purchase Intenders",
  "description": "Consumers actively researching or considering a vehicle purchase within 90 days",
  "relationships": [
    { "type": "is_a", "target": "purchase_intenders" },
    { "type": "distinct_from", "target": "auto_enthusiasts" },
    { "type": "broader_than", "target": "new_vehicle_intenders" },
    { "type": "broader_than", "target": "used_vehicle_intenders" },
    { "type": "overlaps_with", "target": "auto_enthusiasts", "overlap_estimate": 0.4 }
  ],
  "iab_audience_mapping": {
    "taxonomy": "iab-audience-1.1",
    "primary": "Automotive > Auto Purchase Intent"
  },
  "suggested_signals": ["purchase_intent.vehicle", "search_behavior.auto_research", "content_affinity.auto_reviews"]
}

{
  "concept_id": "affluent_professionals",
  "domain": "audience",
  "label": "Affluent Professionals",
  "description": "Consumers in professional occupations with household income in the top 20% for their market",
  "relationships": [
    { "type": "is_a", "target": "high_income" },
    { "type": "overlaps_with", "target": "business_decision_makers", "overlap_estimate": 0.5 },
    { "type": "distinct_from", "target": "affluent_retirees" },
    { "type": "narrower_than", "target": "high_income" }
  ],
  "iab_audience_mapping": {
    "taxonomy": "iab-audience-1.1",
    "primary": "Demographics > Income > High Income"
  },
  "suggested_signals": ["demographic.income_bracket", "professional.occupation_tier"]
}

{
  "concept_id": "health_conscious_consumers",
  "domain": "audience",
  "label": "Health-Conscious Consumers",
  "description": "Consumers with demonstrated interest in wellness, nutrition, fitness, and preventive health",
  "relationships": [
    { "type": "overlaps_with", "target": "active_lifestyle", "overlap_estimate": 0.6 },
    { "type": "overlaps_with", "target": "organic_food_buyers", "overlap_estimate": 0.5 },
    { "type": "distinct_from", "target": "medical_condition_seekers" },
    { "type": "broader_than", "target": "fitness_enthusiasts" },
    { "type": "broader_than", "target": "nutrition_focused" }
  ],
  "iab_audience_mapping": {
    "taxonomy": "iab-audience-1.1",
    "primary": "Health & Fitness"
  },
  "suggested_signals": ["purchase_intent.health_products", "content_affinity.wellness_media", "app_usage.fitness_tracking"]
}

The key insight is the relationship types. Six relationship types make this an ontology, not just a taxonomy:

Relationship	Meaning	Governance use
is_a	Hierarchical parent	Validate that narrower targeting falls within stated audience
overlaps_with	Partial overlap with estimated degree	Flag when seller maps to an overlapping (but not identical) concept
distinct_from	Explicitly not the same concept	Catch misalignment when seller conflates distinct concepts
broader_than	This concept contains the target	Validate targeting specificity claims
narrower_than	This concept is contained by the target	Validate that delivery matches requested scope
maps_to	Direct equivalence in another taxonomy	Enable cross-taxonomy translation

2. Publisher quality vocabulary

Standardized terms for how buyers describe publisher quality expectations. This addresses the persistent ambiguity where "premium" means three entirely different things to three different parties:

{
  "concept_id": "premium_editorial",
  "domain": "publisher_quality",
  "label": "Premium Editorial",
  "description": "Publishers selected for editorial quality, journalistic standards, and audience trust — independent of pricing or ad format capabilities",
  "distinct_from": ["premium_cpm", "premium_format", "premium_reach"],
  "quality_signals": ["editorial_standards", "audience_trust_score", "content_originality"],
  "anti_signals": ["mfa_score > 50", "programmatic_only"],
  "evaluation_criteria": {
    "required": ["original_content_ratio > 0.7", "editorial_staff_verified"],
    "preferred": ["press_awards", "industry_recognition"]
  }
}

{
  "concept_id": "premium_cpm",
  "domain": "publisher_quality",
  "label": "Premium CPM",
  "description": "Publishers commanding above-market pricing, typically due to audience composition, viewability, or scarcity — independent of editorial quality",
  "distinct_from": ["premium_editorial", "premium_format"],
  "quality_signals": ["historical_cpm_percentile", "audience_composition_value", "viewability_rate"],
  "note": "Price-based quality indicator; does not imply editorial or brand safety quality"
}

{
  "concept_id": "premium_format",
  "domain": "publisher_quality",
  "label": "Premium Format",
  "description": "Publishers offering high-impact ad formats (takeovers, interstitials, native integrations) — independent of editorial quality or pricing tier",
  "distinct_from": ["premium_editorial", "premium_cpm"],
  "quality_signals": ["format_diversity", "high_impact_inventory_share", "custom_integration_capability"]
}

When a buyer agent requests "premium publishers," the ontology forces disambiguation. The governance agent can flag a semantic_fidelity finding if the buyer meant premium_editorial but the seller interpreted it as premium_cpm. This is not a hypothetical scenario — it is one of the most common sources of post-campaign dissatisfaction in the industry.

3. Campaign objective vocabulary

Standardized terms for campaign objectives that map to measurable outcomes and appropriate channels:

{
  "concept_id": "brand_awareness",
  "domain": "objective",
  "label": "Brand Awareness",
  "description": "Objective focused on increasing unaided and aided brand recall among the target audience",
  "relationships": [
    { "type": "measured_by", "target": "reach" },
    { "type": "measured_by", "target": "frequency" },
    { "type": "distinct_from", "target": "direct_response" },
    { "type": "complementary_to", "target": "consideration" }
  ],
  "typical_channels": ["olv", "ctv", "display", "social"],
  "typical_kpis": ["impressions", "reach_pct", "brand_lift"],
  "anti_patterns": ["optimizing_for_clicks", "last_touch_attribution"]
}

{
  "concept_id": "direct_response",
  "domain": "objective",
  "label": "Direct Response",
  "description": "Objective focused on driving measurable user actions — clicks, conversions, sign-ups, purchases",
  "relationships": [
    { "type": "measured_by", "target": "conversion_rate" },
    { "type": "measured_by", "target": "cost_per_acquisition" },
    { "type": "distinct_from", "target": "brand_awareness" },
    { "type": "complementary_to", "target": "retargeting" }
  ],
  "typical_channels": ["search", "social", "display", "native"],
  "typical_kpis": ["cpa", "roas", "conversion_rate", "ctr"]
}

{
  "concept_id": "consideration",
  "domain": "objective",
  "label": "Consideration / Mid-Funnel",
  "description": "Objective focused on driving active evaluation — site visits, content engagement, product research",
  "relationships": [
    { "type": "measured_by", "target": "engagement_rate" },
    { "type": "measured_by", "target": "site_visit_rate" },
    { "type": "complementary_to", "target": "brand_awareness" },
    { "type": "complementary_to", "target": "direct_response" }
  ],
  "typical_channels": ["native", "olv", "social", "display"],
  "typical_kpis": ["video_completion_rate", "time_on_site", "pages_per_session"]
}

4. Geographic concept vocabulary

Standardized terms for geographic intent with disambiguation. This addresses a real and persistent class of errors where NLP systems misinterpret geographic references:

{
  "concept_id": "us_state_new_mexico",
  "domain": "geography",
  "label": "New Mexico (US State)",
  "iso_3166_2": "US-NM",
  "disambiguation": {
    "not": ["country_mexico"],
    "common_confusion": "Natural language 'New Mexico' is frequently misinterpreted as the country Mexico by NLP systems"
  },
  "relationships": [
    { "type": "is_a", "target": "us_state" },
    { "type": "distinct_from", "target": "country_mexico" },
    { "type": "contained_by", "target": "us_southwest_region" }
  ]
}

{
  "concept_id": "us_dma_new_york",
  "domain": "geography",
  "label": "New York DMA",
  "dma_code": "501",
  "disambiguation": {
    "includes": ["us_state_new_york_partial", "us_state_new_jersey_partial", "us_state_connecticut_partial"],
    "common_confusion": "DMA geography does not align with state boundaries — NY DMA includes parts of NJ and CT"
  },
  "relationships": [
    { "type": "is_a", "target": "us_dma" },
    { "type": "overlaps_with", "target": "us_state_new_york" },
    { "type": "overlaps_with", "target": "us_state_new_jersey" },
    { "type": "distinct_from", "target": "us_state_new_york" }
  ]
}

Geographic disambiguation matters because budget allocation errors caused by geographic misinterpretation are expensive and difficult to detect in-flight. A governance agent armed with the reference ontology can validate that a seller's geographic targeting actually matches the buyer's intent before impressions are served.

5. Cross-taxonomy mapping layer

The glue that connects everything. Cross-taxonomy mappings show how a single audience concept translates across the taxonomies AdCP already supports:

{
  "mapping_id": "outdoor_recreation_cross_taxonomy",
  "source_concept": "outdoor_recreation",
  "mappings": [
    {
      "taxonomy": "iab-audience-1.1",
      "mapped_to": "Sports & Recreation",
      "confidence": 0.7,
      "precision_loss": true,
      "note": "IAB taxonomy is too broad — includes general sports fans who have no outdoor recreation interest"
    },
    {
      "taxonomy": "iab-content-3.0",
      "mapped_to": ["632:Outdoor Recreation", "633:Camping", "634:Hiking"],
      "confidence": 0.9,
      "precision_loss": false
    },
    {
      "taxonomy": "adcp-channel",
      "affinity_channels": ["display", "olv", "social"],
      "low_affinity_channels": ["linear_tv", "radio"]
    },
    {
      "taxonomy": "adcp-industry",
      "relevant_industries": ["sporting_goods", "travel_tourism", "outdoor_apparel"],
      "relationship": "advertiser_categories_with_high_audience_affinity"
    }
  ]
}

The cross-taxonomy mapping makes explicit what is currently implicit (and often wrong). When a buyer says "outdoor recreation enthusiasts" and a seller returns inventory tagged with IAB Content Taxonomy "Sports & Recreation," the mapping layer quantifies the precision loss: confidence 0.7, precision loss true, because the IAB category is broader than the requested concept.

This is the information a governance agent needs to produce a meaningful semantic_fidelity finding — not just "the mapping happened" but "the mapping lost 30% precision because the target taxonomy lacks the granularity of the source concept."

How the ontology integrates with the three companion proposals

Each of the three companion proposals defines a mechanism. The reference ontology provides the vocabulary those mechanisms operate on:

1. Taxonomy declaration (supported_taxonomies)

Sellers declare which ontology concepts they can resolve. A seller's capability declaration becomes concrete:

{
  "supported_taxonomies": [
    {
      "taxonomy_id": "adcp-reference-ontology",
      "version": "1.0",
      "domains_supported": ["audience", "publisher_quality", "geography"],
      "resolution_depth": 3,
      "note": "Audience concepts resolved via internal segment mapping; publisher quality self-assessed"
    }
  ]
}

A buyer agent can evaluate whether a seller's taxonomy support is sufficient for a given campaign before sending a plan. "This campaign requires outdoor_recreation audience targeting. Does the seller support the audience domain of the reference ontology?"

2. Semantic fidelity (campaign_intent to semantic_fidelity finding)

Buyers express intent using ontology concepts. The governance agent uses cross-taxonomy mappings to evaluate seller interpretation:

{
  "campaign_intent": {
    "audience": {
      "concept": "outdoor_recreation",
      "ontology": "adcp-reference-ontology-1.0",
      "constraints": [
        { "relationship": "distinct_from", "concept": "general_sports_fans" }
      ]
    }
  }
}

The governance agent can now produce a precise semantic_fidelity finding:

{
  "category_id": "semantic_fidelity",
  "severity": "should",
  "explanation": "Seller mapped buyer concept 'outdoor_recreation' to internal segment 'Sports Enthusiasts'. Reference ontology indicates 'outdoor_recreation' is distinct_from 'general_sports_fans' with overlap_estimate 0.3. Seller's segment appears to include general sports fans, violating the buyer's distinct_from constraint."
}

3. Decision provenance (decision_provenance)

Sellers attest that evaluation used specific ontology concepts, making their reasoning auditable:

{
  "decision_provenance": {
    "ontology_version": "adcp-reference-ontology-1.0",
    "concept_evaluated": "outdoor_recreation",
    "ontology_concepts_referenced": ["outdoor_recreation", "extreme_sports"],
    "cross_taxonomy_check": "iab-content-3.0 categories [632, 633, 634] used for contextual matching",
    "candidates_evaluated": 23,
    "candidates_matched": 7
  }
}

Design principles

• Extensible, not exhaustive. The reference ontology covers common concepts that recur across campaigns and verticals. Sellers and buyers can extend it with proprietary concepts that map TO the reference vocabulary. The ontology does not need to describe every possible audience segment — it needs to describe enough common ones that the mapping layer provides meaningful coverage.

• Relationship-first. What makes this an ontology (not just a taxonomy) is the relationships between concepts. distinct_from, overlaps_with, broader_than — these are the assertions that governance can verify. A flat list of audience labels would be another taxonomy. The relationships are the value.

• Mapping, not replacing. Sellers keep their internal taxonomies. They have invested years in building them. The reference ontology provides a mapping target, not a replacement. A seller with a detailed proprietary outdoor sports taxonomy can map their concepts to outdoor_recreation in the reference ontology, preserving their internal granularity while gaining interoperability.

• Machine-readable, human-reviewable. JSON-LD compatible structure that both agents and procurement teams can work with. Every concept has a human-readable label and description alongside machine-readable concept_id and relationships. A planner reviewing a governance finding should be able to understand what happened without decoding opaque IDs.

• Community-governed. The ontology should be maintained by the AAO working group, not by a single vendor. Contributions follow the same IPR policy as the protocol itself. No single participant's internal taxonomy should have privileged status.

Crawl, walk, run

Crawl (AdCP 3.1 initial release)

Publish the reference ontology as a standalone JSON-LD document alongside the specification. Sellers can optionally reference it in supported_taxonomies. No governance integration required. The audience and geographic domains are published first — these have the highest impact and the most common misalignment patterns. Acme Outdoor and StreamHaus use the ontology concept outdoor_recreation in a pilot campaign; Pinnacle Agency's governance agent logs ontology references but does not enforce them.

Walk (AdCP 3.1 mid-cycle)

Governance agents use the cross-taxonomy mapping layer to enrich semantic_fidelity findings. Buyers express intent using ontology concept IDs. The ontology covers audience, publisher quality, and geographic domains. Governance findings reference specific ontology relationships when flagging misalignment. StreamHaus declares support for the audience and publisher quality domains in its supported_taxonomies response. Pinnacle Agency's governance agent produces semantic_fidelity findings that reference the cross-taxonomy mapping: "Seller mapped outdoor_recreation to IAB Sports & Recreation — precision loss of approximately 30% per reference ontology mapping."

Run (AdCP 3.2+)

The reference ontology covers all five domains. Cross-taxonomy mappings are community-maintained with version control. Governance validates that seller interpretations are consistent with ontology relationships. New concepts require working group approval with a defined contribution process. The ontology becomes the standard reference frame for all AdCP semantic operations — buyer intent, seller capability declaration, governance evaluation, and decision provenance all use ontology concepts as their shared language.

Backward compatibility

The reference ontology is a standalone artifact published alongside the specification — it does not modify any existing AdCP 3.0 schema. Consumers that do not reference the ontology continue to work exactly as they do today. When a seller declares taxonomy_id: "adcp-reference-ontology" in supported_taxonomies, buyers that do not recognize the taxonomy ID ignore it per standard JSON handling. Governance agents that do not support ontology-aware evaluation skip ontology-enriched findings and produce standard findings using existing capabilities. The ontology adds vocabulary and relationships; it does not change protocol behavior for participants that do not opt in.

---

Stakeholder considerations

Stakeholder	Benefit	Concern	Mitigation
Buyer agents	Express intent precisely; reduce semantic drift between what they ask for and what they get	Learning curve for ontology concepts; may need to map existing workflows	Ontology concepts have human-readable labels; tooling can suggest concept mappings
Seller agents	Clear specification of what buyers mean; fewer post-campaign disputes about audience mismatch	Mapping internal taxonomy to reference ontology requires effort; precision loss in mapping	Mapping is optional in crawl phase; sellers map only the concepts they encounter
Governance agents	Concrete reference frame for evaluating semantic fidelity; can produce precise, actionable findings	Must maintain ontology awareness; versioning adds complexity	Ontology is versioned and backward-compatible; governance agents cache the latest version
Procurement teams	Can review governance findings that reference common vocabulary rather than opaque segment IDs	Another standard to learn	Ontology concepts are designed for human readability; descriptions explain meaning
AAO working group	Shared vocabulary reduces ambiguity in spec discussions; enables more precise interoperability requirements	Governance burden of maintaining the ontology; risk of scope creep	Crawl/walk/run phases limit initial scope; contribution process gates additions
Industry (IAB, etc.)	Reference ontology builds on IAB taxonomies rather than replacing them; increases IAB taxonomy adoption	Potential perception of competition with IAB taxonomy efforts	Explicit design principle: mapping, not replacing. IAB taxonomies are referenced as foundational layers

Open questions

1. Governance model: Who maintains the ontology? The AAO working group directly? A dedicated ontology subcommittee with domain expertise? What is the process for resolving disputes about concept definitions or relationships?

2. Serialization format: JSON-LD provides web-native linked data with broad tooling support. OWL (Web Ontology Language) provides formal reasoning capabilities but higher complexity. SKOS (Simple Knowledge Organization System) is lighter weight and designed for vocabulary management. Recommendation: JSON-LD with SKOS-compatible structure for the initial release, with a migration path to OWL if formal reasoning proves necessary.

3. Versioning strategy: How do you version an ontology? Semantic versioning on the document as a whole? Individual concept versioning to allow incremental updates? Both? If a concept's relationships change, does that constitute a breaking change?

4. Contribution process: How do sellers and buyers propose new concepts or relationships? Pull request model against a canonical repository? Formal proposal process through the working group? What is the bar for accepting a new concept — how many participants must find it useful?

5. Relationship to IAB taxonomies: The reference ontology is designed as a complement and mapping layer, not a superset. But as the ontology grows, it may cover ground that overlaps with IAB taxonomy evolution. How do we ensure alignment rather than divergence? Should IAB be invited to participate in ontology governance?

6. Scope for 3.1: Which domains to prioritize for initial release? Recommendation: audience and geographic domains first. These are the highest-impact domains (most common source of misalignment in real campaigns) and the most tractable (well-understood problem space with existing prior art to draw on).

7. Overlap estimation methodology: The overlap_estimate field in overlaps_with relationships is powerful but potentially contentious. How are overlap estimates determined? Self-reported by contributors? Empirically measured? Should the ontology include methodology metadata for each estimate?

8. Proprietary concept mapping confidentiality: When a seller maps their internal taxonomy to the reference ontology, the mapping itself may reveal competitive information about their data assets. How does the protocol handle mapping confidentiality while still enabling governance evaluation?

Working group context and prior decisions

This proposal provides the vocabulary layer that makes the three companion proposals — and the working group's existing decisions — practical:

Governance WG Slack discussion (April 2025): The working group agreed that governance requirements operate within the Policy Registry framework and that "different entities hold different compliance thresholds." A reference ontology makes this concrete: when two entities disagree on what "outdoor enthusiasts" means, the ontology provides the shared frame of reference to evaluate the disagreement. Without it, the Policy Registry can enforce rules about budget and brand safety, but has no vocabulary for semantic rules.

Brian O'Kelley's response and merged PR #3175: The audit-trail doc's field-by-field visibility tagging works for structured fields (budget amounts, policy IDs, approval status). But semantic fidelity findings — "the seller mapped 'outdoor_recreation' to 'Sports Fans' with 30% precision loss" — require a reference vocabulary to be meaningful. The reference ontology provides the concept definitions and cross-taxonomy mappings that make those findings precise and actionable rather than subjective.

Issues and PRs already resolved:

• Document campaign governance audit trail: internal vs shareable views #3139 → PR docs(governance): audit-trail internal-vs-shareable views + Addie anonymous knowledge tools #3175 (merged). The audit trail can now tag which fields are shareable. The reference ontology ensures the content of semantic findings references standardized concepts rather than implementation-specific segment IDs — making shareable findings actually interpretable by counterparties.
• Document Policy Registry sync pattern: version pinning, change handling, effective_date transitions #3140 — Policy Registry sync patterns. Ontology versions can be pinned in the Policy Registry alongside policy versions, ensuring that semantic evaluation uses a consistent vocabulary across effective_date transitions.

Production validation — why a reference ontology is necessary, not theoretical:

• A healthcare campaign targeting "Bay Area" was narrowed by the agent to San Francisco DMA only, missing Sacramento, Modesto, and other Northern California DMAs. A reference ontology with geographic decomposition (bay_area → overlaps_with: [sf_dma, sacramento_dma, modesto_dma]) would have prevented this.
• A buyer targeting "New Mexico" had it resolved to Mexico — different country, different language, different market. The geographic domain of the reference ontology encodes us_state_new_mexico as distinct_from: country_mexico explicitly.
• A global brand targeting "football fans" gets NFL in the US, soccer in Europe — different sport, different audience, different sponsorship category. Locale-aware concept resolution (football + US → NFL, football + EU → soccer) requires structured relationships that flat taxonomies cannot express.

Industry collaboration: Yahoo is co-leading the semantic matching workstream within AdCP, and partnering with Google on the open-source BigQuery Agent Analytics SDK where the same ontology-driven patterns are being applied to agentic systems at large. Production ontology implementations (encoding structured relationships across audience segments, contextual signals, product types, pricing models, brand safety classifications, and geographic taxonomies) inform the reference ontology's domain structure and relationship types. The SDK's YAML-driven ontology extraction with OWL import capabilities validates that the proposed serialization approach (JSON-LD with SKOS-compatible structure, migration path to OWL) works at production scale.

AAO Spotlight: The reference ontology is the foundational layer showcased in Yahoo's AAO/AdCP Foundry sizzle reel submission — demonstrating that "agents that know, not guess" requires a shared vocabulary where relationships between concepts (not just terms) are explicit, machine-readable, and auditable.

---

Relationship to other tracks

This proposal is the fourth in a series of four companion proposals for AdCP 3.1 semantic governance extensions:

• Proposal 1: Taxonomy Declaration — Defines supported_taxonomies in seller capability responses. The reference ontology is one of the taxonomies sellers can declare support for. See: GITHUB_ISSUE_taxonomy_declaration.md
• Proposal 2: Semantic Fidelity — Defines campaign_intent and semantic_fidelity governance findings. The reference ontology provides the vocabulary that campaign intent is expressed in and the cross-taxonomy mappings that semantic fidelity is evaluated against. See: GITHUB_ISSUE_semantic_fidelity.md
• Proposal 3: Decision Provenance — Defines decision_provenance in seller responses. The reference ontology provides the concept vocabulary that provenance attestations reference. See: GITHUB_ISSUE_decision_provenance.md

The three companion proposals define the plumbing. This proposal provides the first water to flow through it.

Prior work

• AdCP Media Channel Taxonomy (20 channels, already in AdCP 3.0 specification)
• AdCP Advertiser Industry Taxonomy (73 categories, already in AdCP 3.0 specification)
• IAB Content Taxonomy 3.0 (referenced in TMP context signals)
• IAB Audience Taxonomy 1.1 (industry standard for audience segment classification)
• Schema.org (prior art in machine-readable web vocabulary with relationship types)
• SKOS (Simple Knowledge Organization System) (W3C standard for knowledge organization, provides broader, narrower, related relationship primitives)
• Dublin Core (prior art in metadata vocabulary with cross-domain applicability)
• The feat/semantic-governance-extensions branch prototype (working implementation that informed this proposal)

---

Related: See also #3362 — Proposal: Taxonomy declaration as a core capability (3.1) | #3363 — Proposal: Semantic fidelity as a core governance capability (3.1) | #3364 — Proposal: Decision provenance and lineage as core governance capabilities (3.1)

[bokelley]

Thanks for filing!

Triage

Classification: RFC / Proposal
Bucket(s): spec / protocol
Status: ready-for-human

What the experts said: (RFC path — scope adoption is a human decision)

My take: Proposes a standalone JSON-LD reference ontology published alongside the spec — the shared vocabulary layer that #3362 (taxonomy declaration), #3363 (semantic fidelity), and #3364 (decision provenance) all depend on to be concrete. Additive and non-breaking (no existing schema modifications). Eight open questions — governance model, serialization format, versioning strategy, IAB alignment — are structural and need working-group decisions before any artifact can be published. The proposal to scope crawl-phase to audience + geographic domains only is a sensible starting point.

@bokelley, your call: (A) adopt into 3.1 spec process as a companion artifact; (B) defer to 3.2; or (C) bootstrap the ontology in a separate working-group track and reference from the spec once it has a stable identity (decouples from 3.1 timeline).

---

Triaged by Claude Code. Session: https://claude.ai/code/cse_01EBJw8sDDjm8PuyiC2dcxRt

---

Generated by Claude Code

[bokelley]

Thanks for filing!

Triage

Classification: Proposal / RFC
Bucket: spec / protocol
Status: deferred

My take: A shared reference ontology (audience, publisher quality, objective, geography, cross-taxonomy mapping) is the vocabulary layer the companion proposals (#3362–#3364) need to produce precise, machine-evaluable governance findings rather than approximate natural-language comparisons. This is part of a four-issue set (#3362, #3363, #3364, #3365) targeting 3.1; none are yet in the active 3.1 scoping list, so they're parked here pending the working group's scope decision.

---

Triaged by Claude Code. Session: https://claude.ai/code/session_017ovY7JJhKSdu4k2qBzi4WC

---

Generated by Claude Code