Skip to main content
The protocol does not mandate what a buyer must share with a counterparty. It does give you the primitives to produce two distinct views of the same audit trail:
  • An internal view — the buyer’s full record of every check, finding, budget movement, and human review. Used for self-protection, regulator audits, and internal compliance review.
  • A shareable view — a scoped subset, addressed to one counterparty (a seller, an agency, a third-party auditor). Reveals only what that party needs to verify their own action.
Both are produced from get_plan_audit_logs. The split is a matter of filtering and field selection, not a separate protocol.

Why split at all

Three reasons buyers maintain distinct views:
  1. Strategy leakage. Total authorized budget, channel allocation, and remaining headroom across all sellers are competitive signals. Note that derived ratios are not safer than raw amounts — budget.utilization_pct is a one-step inverse problem if a seller knows their own committed share. A single seller does not need to know the buyer’s full plan to verify their own buy.
  2. Cross-counterparty isolation. Seller A has no business seeing the findings, governance contexts, or outcomes for Seller B. A shareable view is always scoped to the requesting party’s governance_context values.
  3. Internal review fields. Human review reasons, drift metrics, and oversight thresholds are the buyer’s own governance posture. They protect the buyer in a regulator audit; they are not addressed to a counterparty.

Field tagging

The table below classifies every field in the get_plan_audit_logs response by who can safely see it.

Producing a shareable view

To answer “what should I share with Seller X?”, filter the audit query to the contexts that belong to that seller and drop plan-level aggregates. Request scoped to one seller’s actions:
This narrows governed_actions and entries to the requested contexts. The buyer is still responsible for stripping budget.*, channel_allocation.*, summary.findings_count, summary.statuses, and summary.drift_metrics before forwarding — those summaries cover the entire plan and reveal totals across other sellers. Minimum compliance attestation (no entries): For counterparties who only need proof that a buy was governed — not the full trail — share a four-field attestation derived from the audit response:
The seller can verify the plan_hash against the plan revision they signed under, and the policies_evaluated list confirms which registry policies were applied. Nothing else about the buyer’s portfolio is disclosed.

Regulator and auditor views

A regulator typically asks for one of three things:
  1. Did human oversight occur where required? Share summary.escalations[] and the requires_human_review flag from the resolved policies. Reasons may be redacted or summarized; resolutions and timestamps should be intact.
  2. Did the buy comply with a specific regulation? Share entries scoped by policies_evaluated containing the regulation’s policy_id (e.g., us_coppa). Include plan_hash so the regulator can verify against the plan revision.
  3. Are oversight metrics within thresholds? Share summary.drift_metrics and the policy-derived thresholds. An escalation_rate below thresholds.escalation_rate_min is a signal worth explaining.
The protocol does not define a regulator API. Counterparty rules govern the disclosure; the audit trail makes the disclosure possible.

Worked example: a clean buy

A plan with $500K authorized for an OLV/display campaign. The orchestrator runs an intent check on `get_products`, then an execution check on `create_media_buy` for $150K, then reports the outcome. The full internal audit response:
The seller’s shareable view of the same trail is the governed_actions[] entry and the entries[] filtered to governance_context = 11ab64d0.... Plan-level budget.authorized, budget.remaining, and the summary aggregate stay buyer-side.

What violations look like

Findings are how the audit trail communicates risk. Two patterns worth knowing:

Security-shaped finding: an unauthorized seller

The plan declares approved_sellers: ["https://ads.seller-approved.example"]. A different seller calls check_governance and gets denied:
The corresponding plan summary records statuses.denied: 1 and findings_count: 1. No governance_context is issued; the audit entry captures the denied attempt through check_id.

Coaching-shaped finding: an Annex III prerequisite is missing

A plan with policy_categories: ["fair_lending"] (mortgage, consumer credit, etc.) auto-flips human_review_required: true. If the brand profile doesn’t expose a contestation contact, the check fails closed with an actionable explanation:
The denial is also a coaching moment — the operator sees exactly what to fix. This is the shape good governance findings should take: a category, a severity, and a path forward.

The operator’s dial: enforce / advisory / audit

mode is governance agent configuration. The buyer’s policy team sets it on the governance agent itself; it is not a plan field and is not sent in sync_plans. It is the operator’s primary lever, and an agent that reports mode records the mode in force on each check entry. The same caller, same payload, same finding produces three different outcomes depending on mode. mode: "enforce" — the buy is blocked at the governance layer:
mode: "advisory" — the buy proceeds; the operator gets a critical finding to act on post-hoc:
mode: "audit" — the buy proceeds silently; the finding is in the log for retrospective analysis:
The trade-off: The agent’s mode can change during a plan’s lifetime. A different mode on two audit entries reflects a change to the governance agent’s configuration, not a plan revision — the governance decisions in the audit trail tell the truth about how each historical buy was governed. The operating mode should be visible on every audit entry; today it appears on the per-check_governance response but not consistently on get_plan_audit_logs responses (#3156).

What the protocol guarantees

Two invariants you can rely on when designing your shareable view:
  • plan_hash is verifiable. It is base64url_no_pad(SHA-256(JCS(plan_payload))) over the plan revision the check was evaluated against. Any party with the plan revision can recompute and byte-compare the digest. See Plan binding and audit.
  • Inline policies cannot relax registry policies. A buyer’s bespoke policy entries can only add restrictions on top of registry-sourced policies. So when a shareable view shows policies_evaluated: ["us_coppa"], the counterparty can trust that the registry version of us_coppa was applied at its declared enforcement level — the buyer did not silently downgrade it. See Policy Registry.