check_governance

Experimental. Campaign governance (sync_plans, check_governance, report_plan_outcome, get_plan_audit_logs) is part of AdCP 3.0 as an experimental surface — it may change between 3.x releases with at least 6 weeks’ notice. Sellers implementing it MUST declare governance.campaign in experimental_features. See experimental status for the full contract.

Universal governance check for campaign actions. Both the orchestrator (buyer-side) and the seller call this task. The governance agent infers the check type from the fields present:

Check type	Who calls	Discriminating fields	Purpose
Intent	Orchestrator	`tool` + `payload`	Validates the intended action before sending to a seller. No budget committed.
Execution	Seller	`planned_delivery` + `governance_context`	Validates what the seller will actually deliver. Budget is committed later via `report_plan_outcome`.

The governance agent maintains all state. Callers do not chain check IDs or track conversation history — they post the action, and the governance agent correlates by plan_id. On subsequent lifecycle checks, callers include governance_context from the prior response for continuity. An account binds to one governance agent (see sync_governance and One governance agent per account). All lifecycle calls for a governed action go to that same agent.

Per-specialist review surfaces here. Internal decomposition (legal, brand safety, category) is not exposed as separate endpoints — it appears as agent-internal labels in categories_evaluated and, on denied, conditions, or informational approved responses, in findings[].details. Treat these values as opaque audit values; do not pattern-match against a fixed list.

Check types

Intent checks (orchestrator)

The orchestrator calls check_governance with tool and payload before sending a tool call to a seller. The governance agent evaluates the intended action against the campaign plan.

Orchestrator decides to call a seller tool (e.g., create_media_buy)
Orchestrator calls check_governance with the tool name and full payload
If approved, orchestrator sends the tool call to the seller
If denied, orchestrator does not send the tool call
If conditions, orchestrator adjusts the payload and re-calls check_governance
If the governance agent needs human review, the task goes async and eventually resolves to approved or denied

Execution checks (seller)

The seller calls check_governance with governance_context and planned_delivery when processing a request on an account that has a governance agent configured (set via sync_governance). Execution checks are always binding — if the governance agent denies, the seller must not proceed. Before executing the check, the seller verifies the signed governance_context token that arrived on the protocol envelope from the buyer. The buyer produces an intent-phase token (per the JWS profile); the seller’s own execution check produces the purchase/modification/delivery-phase tokens bound to the assigned media_buy_id for the rest of the lifecycle.

on receive(create_media_buy request):
  token = request.envelope.governance_context
  persist(token)                                      # always persist for audit/forwarding
  verify(token, {                                     # per Security — Signed Governance Context
    sellerId:    my_adagents_url,
    planId:      request.plan_id,
    phase:       "intent",                             # buyer produces intent tokens
    mediaBuyId:  null,                                 # intent tokens have no media_buy_id
  })                                                  # throws on any of 15 checks failing
  call check_governance(planned_delivery, token)      # seller-side execution check — produces purchase-phase token
  proceed only if governance_agent verdict = approved

Sellers that have not yet implemented verification MUST still persist and forward the token unchanged — auditors and regulators rely on this. Verification is the ramp from “forward-only” compliance to cryptographic accountability and can be adopted incrementally. Execution checks cover the full media buy lifecycle through three phases:

Phase	When	What’s checked
`purchase`	Before confirming `create_media_buy`	Budget, geo, channels, flight dates, policies
`modification`	Before confirming `update_media_buy`	Change magnitude, reallocation, new parameters
`delivery`	Periodically during delivery	Pacing, spend rate, geo drift, channel distribution

Sellers can adopt committed governance checks incrementally:

Level 1: Purchase only — One call per create_media_buy. The minimum viable integration.
Level 2: + Modification — One call per update_media_buy.
Level 3: + Delivery reporting — Periodic calls during active delivery.

Invocation requirement

When a governance agent is configured on the plan, buyer agents MUST invoke check_governance before every spend-commit request (create_media_buy, update_media_buy, acquire_rights, update_rights, activate_signal, build_creative) — full stop. There is no dollar floor, no anomaly threshold, and no cold-start exemption; every commit goes through the governance agent. The buyer-side call is an intent check (tool + payload), which produces an intent-phase governance_context token the buyer attaches to its request to the seller. The governance agent decides internally whether to auto-approve, apply conditions, deny, or escalate to human review per the plan’s budget.reallocation_threshold and human_review_required fields. Seller-side enforcement makes the MUST real through the signed governance_context token: a seller receiving a spend-commit for a plan with a configured governance agent MUST require a valid, in-date intent-phase token (phase: "intent", sub equal to the plan_id, aud addressed to this seller), and MUST reject with PERMISSION_DENIED otherwise. The seller then performs its own execution check — calling check_governance with planned_delivery and the received governance_context — which produces the purchase-phase token bound to the newly assigned media_buy_id for the rest of the lifecycle. A buyer that skips the intent check cannot produce a valid intent token, so the commit is rejected before the seller ever reaches its execution check. When no governance agent is configured on the plan, check_governance invocation is neither required nor meaningful — there is nothing to call. Sellers MAY refuse to transact on plans lacking a configured governance agent as a matter of their own commercial policy. See the specification for the full definition, including audit requirements, seller-side retention MUSTs, and interaction with idempotency.

Status values

Status	Meaning	Caller action
`approved`	Proceed as planned.	Act before `expires_at` or re-call.
`denied`	Do not proceed.	Return error to upstream caller.
`conditions`	Approved if caller accepts adjustments.	Apply conditions, then re-call `check_governance` with adjusted parameters.

Expiration

expires_at is present when the verdict is approved or conditions. A lapsed approval is no approval — the caller must re-call check_governance before proceeding.

Conditions

When the verdict is conditions, the caller MUST re-call check_governance with adjusted parameters before proceeding. Conditions with a required_value are machine-actionable — the caller can programmatically apply the value. Conditions without a required_value are advisory — the caller should interpret the reason and adjust accordingly. Governance agents SHOULD return denied (not conditions) after 3 unsuccessful re-calls for the same action. This prevents infinite negotiation loops, particularly for seller-side checks where the seller has no visibility into the campaign plan.

Human review

When the governance agent determines that human review is required (e.g., the action exceeds the plan’s reallocation_threshold, or the plan carries human_review_required: true), it handles the escalation internally. The check_governance task goes async — the caller receives standard async task lifecycle statuses (submitted, working) and eventually gets approved or denied once the human acts. The caller does not need special handling for this case beyond supporting async tasks (see task lifecycle). For committed checks (seller-side), the seller sets a timeout. If the governance agent does not respond within the timeout, the seller treats it as denied and returns an error to the orchestrator. The orchestrator can re-initiate the media buy after the governance agent resolves.

Linking to outcomes

The response includes a check_id. Use this in report_plan_outcome to link outcomes to the governance check that authorized them.

When the governance agent is unavailable

If the governance agent is configured and the caller cannot reach it (timeout, network error), the caller MUST NOT proceed. Governance is a gate — when the gate is unreachable, the default is halt. The caller SHOULD retry with backoff and report the failure upstream.

Delivery cadence

The presence of next_check in a response is the signal that the governance agent expects ongoing delivery reporting. The seller SHOULD call no later than the next_check time. The governance agent MAY treat a missed deadline as a finding on the next delivery check.

Request

Intent check (orchestrator checking before sending to seller)

{
  "tool": "check_governance",
  "arguments": {
    "plan_id": "plan_q1_2026_launch",
    "caller": "https://orchestrator.example.com",
    "tool": "create_media_buy",
    "payload": {
      "product_id": "premium_video_300k",
      "budget": 150000,
      "currency": "USD",
      "geo": { "countries": ["US"] },
      "channels": ["olv"],
      "flight": {
        "start": "2026-03-15T00:00:00Z",
        "end": "2026-06-15T00:00:00Z"
      }
    }
  }
}

On the first check_governance call, the governance agent extracts what it needs from payload. The response includes a governance_context string that the caller attaches to the protocol envelope and includes on all subsequent governance calls for this governed action. In 3.0 the governance agent MUST emit a compact JWS signed per the AdCP JWS profile so sellers can verify authenticity, authorization scope, and freshness (the 15-step seller checklist). The token also carries a required plan_hash audit-layer claim — see Plan binding and audit for canonicalization rules, retention obligations, and the eleven reference vectors governance-agent implementers SHOULD validate against before shipping.

Intent check (rights license)

{
  "tool": "check_governance",
  "arguments": {
    "plan_id": "plan_acme_summer_2026",
    "caller": "https://buying.pinnacle-agency.example",
    "purchase_type": "rights_license",
    "tool": "acquire_rights",
    "payload": {
      "brand": { "domain": "acmeoutdoor.com" },
      "right_type": "image_generation",
      "pricing_option_id": "standard_monthly",
      "campaign": {
        "countries": ["US"],
        "start_date": "2026-04-01",
        "end_date": "2026-06-30"
      }
    }
  }
}

Execution check — purchase

{
  "tool": "check_governance",
  "arguments": {
    "plan_id": "plan_q1_2026_launch",
    "caller": "https://seller.example.com",
    "governance_context": "gc_from_buyer_envelope",
    "phase": "purchase",
    "planned_delivery": {
      "geo": { "countries": ["US"] },
      "channels": ["olv"],
      "start_time": "2026-03-15T00:00:00Z",
      "end_time": "2026-06-15T00:00:00Z",
      "total_budget": 150000,
      "currency": "USD",
      "frequency_cap": { "max_impressions": 3, "per": "user", "window": { "interval": 1, "unit": "days" } },
      "audience_summary": "Adults 25-54, US, premium video inventory",
      "enforced_policies": ["us_coppa"]
    }
  }
}

Execution check — modification

{
  "tool": "check_governance",
  "arguments": {
    "plan_id": "plan_q1_2026_launch",
    "caller": "https://seller.example.com",
    "governance_context": "gc_from_buyer_envelope",
    "phase": "modification",
    "modification_summary": "Budget increase from $150,000 to $200,000 and flight extension to 2026-07-15.",
    "planned_delivery": {
      "geo": { "countries": ["US"] },
      "channels": ["olv"],
      "start_time": "2026-03-15T00:00:00Z",
      "end_time": "2026-07-15T00:00:00Z",
      "total_budget": 200000,
      "currency": "USD",
      "frequency_cap": { "max_impressions": 3, "per": "user", "window": { "interval": 1, "unit": "days" } },
      "audience_summary": "Adults 25-54, US, premium video inventory",
      "enforced_policies": ["us_coppa"]
    }
  }
}

Execution check — delivery

{
  "tool": "check_governance",
  "arguments": {
    "plan_id": "plan_q1_2026_launch",
    "caller": "https://seller.example.com",
    "governance_context": "gc_from_buyer_envelope",
    "phase": "delivery",
    "planned_delivery": {
      "geo": { "countries": ["US"] },
      "channels": ["olv"],
      "start_time": "2026-03-15T00:00:00Z",
      "end_time": "2026-06-15T00:00:00Z",
      "total_budget": 150000,
      "currency": "USD",
      "frequency_cap": { "max_impressions": 3, "per": "user", "window": { "interval": 1, "unit": "days" } },
      "audience_summary": "Adults 25-54, US, premium video inventory",
      "enforced_policies": ["us_coppa"]
    },
    "delivery_metrics": {
      "reporting_period": {
        "start": "2026-03-15T00:00:00Z",
        "end": "2026-03-22T00:00:00Z"
      },
      "spend": 12500,
      "cumulative_spend": 12500,
      "impressions": 850000,
      "cumulative_impressions": 850000,
      "geo_distribution": { "US": 100 },
      "channel_distribution": { "olv": 100 },
      "pacing": "on_track",
      "audience_distribution": {
        "baseline": "platform",
        "indices": {
          "age:18-24": 0.8,
          "age:25-34": 1.4,
          "age:35-44": 1.3,
          "age:45-54": 1.1,
          "gender:female": 1.05,
          "gender:male": 0.95
        },
        "cumulative_indices": {
          "age:18-24": 0.85,
          "age:25-34": 1.35,
          "age:35-44": 1.25,
          "age:45-54": 1.1,
          "gender:female": 1.03,
          "gender:male": 0.97
        }
      }
    }
  }
}

Response

approved (intent check)

{
  "check_id": "chk_001",
  "verdict": "approved",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Proposed create_media_buy is within plan parameters. Budget: $150,000 of $500,000 plan total. Geo: US (within plan). Channel: OLV (within 40-70% target range).",
  "categories_evaluated": ["budget_authority", "geo_compliance", "channel_compliance", "flight_compliance", "delegation_authority"],
  "policies_evaluated": ["us_coppa", "alcohol_advertising"],
  "expires_at": "2026-03-15T01:00:00Z"
}

The orchestrator proceeds to send the create_media_buy to the seller before expires_at.

approved (execution check — purchase with delivery opt-in)

{
  "check_id": "chk_002",
  "verdict": "approved",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Planned delivery is within plan parameters. Budget: $150,000 of $500,000 plan total. Geo: US (within plan). Channel: OLV (within 40-70% target range).",
  "mode": "enforce",
  "expires_at": "2026-03-15T01:00:00Z",
  "next_check": "2026-03-22T00:00:00Z"
}

The seller proceeds with the media buy. The presence of next_check signals that the governance agent expects delivery reporting starting at that time.

approved (execution check — delivery)

{
  "check_id": "chk_003",
  "verdict": "approved",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Delivery on track. Week 1 spend: $12,500 of $150,000 (8.3%). Pacing is on target for 13-week flight. Geo and channel distribution match plan parameters.",
  "next_check": "2026-03-29T00:00:00Z"
}

The seller continues delivery and schedules the next governance check for next_check.

denied (intent check)

{
  "check_id": "chk_004",
  "verdict": "denied",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Proposed media buy targets CA (Canada) which is not within the plan's geography.",
  "findings": [
    {
      "category_id": "strategic_alignment",
      "severity": "critical",
      "explanation": "Geo targeting includes CA but plan only covers US.",
      "details": {
        "plan_countries": ["US"],
        "payload_countries": ["US", "CA"]
      }
    }
  ]
}

The orchestrator MUST NOT send the tool call to the seller.

denied (execution check — delivery geo drift)

{
  "check_id": "chk_005",
  "verdict": "denied",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Delivery has drifted outside plan parameters. 12% of impressions delivered in CA (Canada) which is not within the plan's geography.",
  "findings": [
    {
      "category_id": "strategic_alignment",
      "severity": "critical",
      "confidence": 0.98,
      "explanation": "Geo distribution shows 12% delivery in CA, but plan only covers US.",
      "details": {
        "plan_countries": ["US"],
        "actual_distribution": { "US": 88, "CA": 12 }
      }
    }
  ]
}

The seller MUST pause delivery immediately and correct the geo targeting before resuming.

conditions (execution check — purchase)

{
  "check_id": "chk_006",
  "verdict": "conditions",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Budget approved but frequency cap must be applied per brand policy.",
  "conditions": [
    {
      "field": "planned_delivery.frequency_cap",
      "required_value": { "max_impressions": 5, "per": "user", "window": { "interval": 1, "unit": "days" } },
      "reason": "Brand policy requires daily frequency cap of 5 or fewer impressions per user."
    }
  ],
  "expires_at": "2026-03-15T01:00:00Z"
}

The seller MUST adjust its planned delivery, then re-call check_governance with the updated parameters before proceeding.

conditions (execution check — delivery overpacing)

{
  "check_id": "chk_007",
  "verdict": "conditions",
  "plan_id": "plan_q1_2026_launch",
  "explanation": "Delivery is pacing 40% ahead of schedule. Cumulative spend of $42,000 after 2 weeks exceeds expected $23,000 for this point in the flight.",
  "conditions": [
    {
      "field": "pacing",
      "reason": "Reduce daily spend rate to align with the planned flight duration. At current pace, budget will be exhausted by week 7 of 13."
    }
  ],
  "next_check": "2026-03-31T00:00:00Z"
}

The seller MUST adjust pacing and re-call check_governance immediately. The next_check is set closer than normal so the governance agent can verify the correction.

Fields

Request

Field	Type	Required	Description
`plan_id`	string	Yes	Campaign governance plan identifier.
`caller`	string (URI)	Yes	URL of the agent making the request.
`purchase_type`	enum	No	The kind of financial commitment being validated: `media_buy` (default), `rights_license`, `signal_activation`, or `creative_services`. When omitted, the governance agent assumes `media_buy`.
`tool`	string	Intent	The AdCP tool being checked. Present on intent checks (orchestrator). The governance agent uses the presence of `tool` + `payload` to identify an intent check.
`payload`	object	Intent	Full tool arguments as they would be sent to the seller. Present on intent checks.
`governance_context`	string	No	Governance context token from a prior `check_governance` response. Include on subsequent lifecycle checks so the governance agent can maintain continuity. On execution checks, the governance agent uses `governance_context` + `planned_delivery` to identify the check. This is the sole lifecycle correlator across all purchase types. See Signed Governance Context for the JWS profile and seller verification; see Plan binding and audit for the `plan_hash` audit-layer claim.
`phase`	enum	Execution	`purchase`, `modification`, or `delivery`. Defaults to `purchase`. Present on execution checks.
`planned_delivery`	object	Execution	What will actually be delivered. Present on execution checks. See planned delivery.
`delivery_metrics`	object	Delivery	Actual delivery performance data. Required when `phase` is `delivery`.
`delivery_metrics.audience_distribution`	object	No	Audience demographic composition relative to a baseline. Used for bias/fairness drift detection.
`delivery_metrics.audience_distribution.baseline`	enum	Yes	Reference population: `census` (national population), `platform` (platform’s user base), or `custom`.
`delivery_metrics.audience_distribution.baseline_description`	string	No	Description of the baseline when `baseline` is `custom` (e.g., “US adults 18+ with broadband access”).
`delivery_metrics.audience_distribution.indices`	object	Yes	Index values for the current reporting period. Key format: `dimension:value` (e.g., `age:25-34`, `gender:female`). Value of 1.0 means parity with baseline; above 1.0 means over-indexed; below 1.0 means under-indexed.
`delivery_metrics.audience_distribution.cumulative_indices`	object	No	Index values across all reporting periods. Same format as `indices`. Helps governance agents detect trends vs. one-period noise.
`modification_summary`	string	No	Human-readable summary of what changed. SHOULD be present for `modification` phase.

Delivery metrics

Field	Type	Description
`reporting_period`	object	Reporting window with `start` and `end` timestamps (ISO 8601). Required.
`spend`	number	Spend during the reporting period.
`cumulative_spend`	number	Total spend since the media buy started.
`impressions`	integer	Impressions during the reporting period.
`cumulative_impressions`	integer	Total impressions since the media buy started.
`geo_distribution`	object	Actual geographic distribution. Keys are ISO 3166-1 alpha-2 codes, values are percentages.
`channel_distribution`	object	Actual channel distribution. Keys are values from the channels enum, values are percentages.
`pacing`	enum	`ahead`, `on_track`, or `behind`.
`audience_distribution`	object	Audience composition relative to a baseline. Contains `baseline` (enum), optional `baseline_description` (string, for custom baselines), `indices` (current period), and optional `cumulative_indices` (all periods). Keys are `dimension:value` strings, values are index numbers (1.0 means parity).

Response

Field	Type	Description
`check_id`	string	Unique identifier for this governance check. Use in `report_plan_outcome` to link outcomes.
`verdict`	enum	`approved`, `denied`, or `conditions`.
`plan_id`	string	Echoed from request.
`explanation`	string	Human-readable explanation of the decision.
`findings`	array	Per-category issues found. Present when `verdict` is `denied` or `conditions`. MAY also be present on `approved` for informational findings. Each finding has `category_id`, `severity`, `explanation`, and optionally `policy_id`, `details`, `confidence` (0-1), and `uncertainty_reason`. `category_id` is an agent-internal label, not a protocol-level enum — treat as opaque for display/audit, not for machine pattern-matching.
`conditions`	array	Present when `verdict` is `conditions`. Adjustments the caller must make before re-calling.
`categories_evaluated`	string[]	Governance categories evaluated during this check (e.g., `budget_authority`, `geo_compliance`, `channel_compliance`). Agent-internal labels — each string is defined by the governance agent’s policy model and is how internal specialist review (legal, brand safety, category) surfaces for audit from behind the agent’s single endpoint. Not a protocol enum; not safe to pattern-match against a fixed list.
`policies_evaluated`	string[]	Registry policy IDs evaluated during this check.
`mode`	enum	`audit`, `advisory`, or `enforce` — governance mode active when this check was evaluated. Recorded by the governance agent from its runtime configuration at check time, not from a plan field. Lets counterparties, regulators, and auditors distinguish whether an `approved` decision reflects deliberate `enforce` enforcement or `audit`-mode silent logging.
`expires_at`	string	Present when `verdict` is `approved` or `conditions`. The caller must act before this time or re-call. A lapsed approval is no approval.
`next_check`	string	When the seller should next call `check_governance` with delivery metrics. Present when the governance agent expects ongoing delivery reporting.
`governance_context`	string	Governance context token for this governed action. Present when `verdict` is `approved` or `conditions`. Attach to the protocol envelope and include on all subsequent governance calls. This is the sole lifecycle correlator for all purchase types. See Signed Governance Context for the JWS profile and seller verification; see Plan binding and audit for the `plan_hash` audit-layer claim. Sellers that do not verify MUST still persist and forward the token verbatim.
`authority_remaining`	object	Buyer-side plan budget authority remaining after this check — not the seller’s allocated budget. Present when `verdict` is `approved` or `conditions` for execution checks. Contains `budget_remaining` (number), `currency` (string), and `budget_used_pct` (number, 0-100). Orchestrators use this to track plan-level spend against the media plan’s total authority.

Error codes

Code	Recovery	Description
`PLAN_NOT_FOUND`	correctable	No plan with this ID. The buyer may not have synced the plan yet.
`AMBIGUOUS_CHECK_TYPE`	correctable	Request contains both intent fields (`tool` + `payload`) and execution fields (`governance_context` + `planned_delivery`). Send one set or the other.
`CAMPAIGN_SUSPENDED`	correctable	Campaign governance is suspended pending human review.
`SELLER_NOT_RECOGNIZED`	correctable	The caller URL is not in the plan’s `approved_sellers` list.

sync_plans — The plan this governance check validates against
report_plan_outcome — Report what happened after the action was confirmed
get_plan_audit_logs — View plan state and audit trail

Documentation Index

​check_governance

​Check types

​Intent checks (orchestrator)

​Execution checks (seller)

​Invocation requirement

​Status values

​Expiration

​Conditions

​Human review

​Linking to outcomes

​When the governance agent is unavailable

​Delivery cadence

​Request

​Intent check (orchestrator checking before sending to seller)

​Intent check (rights license)

​Execution check — purchase

​Execution check — modification

​Execution check — delivery

​Response

​approved (intent check)

​approved (execution check — purchase with delivery opt-in)

​approved (execution check — delivery)

​denied (intent check)

​denied (execution check — delivery geo drift)

​conditions (execution check — purchase)

​conditions (execution check — delivery overpacing)

​Fields

​Request

​Delivery metrics

​Response

​Error codes

​Related tasks

check_governance

Check types

Intent checks (orchestrator)

Execution checks (seller)

Invocation requirement

Status values

Expiration

Conditions

Human review

Linking to outcomes

When the governance agent is unavailable

Delivery cadence

Request

Intent check (orchestrator checking before sending to seller)

Intent check (rights license)

Execution check — purchase

Execution check — modification

Execution check — delivery

Response

approved (intent check)

approved (execution check — purchase with delivery opt-in)

approved (execution check — delivery)

denied (intent check)

denied (execution check — delivery geo drift)

conditions (execution check — purchase)

conditions (execution check — delivery overpacing)

Fields

Request

Delivery metrics

Response

Error codes

Related tasks