Fix BidCliq compliance suite fixtures

[bokelley]

Updates compliance snapshots and storyboards for BidCliq's 3.0/3.1 runs: refreshes invalid_transitions, makes state_machine create active buys from discovered products, and removes the hardcoded test-product path. Adds seed_account support for deterministic list_accounts pagination and wires the controller schema, advertised scenarios, and lint docs accordingly. Includes an empty changeset because the repo pre-push hook requires one. Validated with targeted storyboard/schema tests, build:compliance, build:schemas, precommit, and the pre-push storyboard matrices for current source and 3.0 compatibility.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
adcp	🟢 Ready	View Preview	Jun 3, 2026, 4:20 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[aao-release-bot]

LGTM. Follow-ups noted below. Right shape: new seed_account mirrors the existing seed_creative_format pattern, idempotency contract is explicit, and the 3.0-compat path now prefers session-scoped accounts before falling back to the fixture pool so the seed flow works in both compat windows.

Things I checked

• seedAccountFixture at server/src/training-agent/account-handlers.ts:456 — auth guard (durableAccountIdentityError) matches handleSyncAccounts / handleListAccounts gate; defaults (billing=operator, status=active, paymentTerms=net_30, accountScope=operator_brand, rateCard=sandbox, sandbox=true) match the sandbox path in handleSyncAccounts.
• handleListAccounts 3.0-compat ternary at server/src/training-agent/account-handlers.ts:1067-1075 — preferFixtureAccounts && scopedAccounts.length > 0 now returns seeded accounts; existing fallback to getComplianceAccounts() preserved. Test exercises both 3.1 and 3.0 paths and confirms has_more=true + cursor on max_results=2.
• Schema if/then at static/schemas/source/compliance/comply-test-controller-request.json:363-383 — same shape as the other seed_* conditionals; requires params.account_id, leaves fixture permissive. scenario is type: string (no enum), so additive by construction.
• audit-oneof.mjs unaffected — new construct is if/then inside allOf, not a oneOf; baseline scripts/oneof-discriminators.baseline.json does not need to move.
• start_time: \"asap\" in the state-machine storyboard validates against static/schemas/source/core/start-timing.json (oneOf includes const: \"asap\"). Clean substitution for the hardcoded ISO timestamp.
• Scenario count assertion at server/tests/unit/comply-test-controller.test.ts:199 lifted from 19 → 20; LOCAL_SCENARIOS, get_adcp_capabilities, and tenant router scenario lists all updated consistently.

Follow-ups (non-blocking — file as issues)

• Pagination storyboard now under-tests its own name. static/compliance/source/universal/pagination-integrity-list-accounts.yaml dropped both the has_more=false + cursor-absence terminal-page assertion and the total_count == 3 invariant — the field_value_or_absent checks are gone. The justification ("agents may have pre-existing sandbox accounts") is undercut by the same diff turning on controller_seeding: true, which is exactly the session-isolation precondition that made the terminal assertion safe in the sibling pagination-integrity.yaml (list_creatives) storyboard. Either rename to pagination-integrity-list-accounts-continuation to match what it actually validates, or restore the terminal-page + total_count checks under the session-isolation predicate. As-is, the storyboard no longer catches the under-reporting failure mode it exists to detect.
• Changeset is empty; prior art was minor. feat(compliance): add seed_creative_format (PR #3114, commit 926b07935) shipped a minor changeset for the structurally identical change — new scenario value + new if/then in the published comply controller schema. The "runners MUST accept unknown scenario strings" rule governs consumers, not the spec's own version bookkeeping. Either justify the empty bump in .changeset/wild-rocks-yell.md (the issue-5228 changeset is the precedent for empty-with-prose) or convert to minor. The storyboard and state-machine edits are orthogonal compliance-fixture work and don't change that calculus.
• Idempotency-equality fragility at server/src/training-agent/account-handlers.ts:511-513. JSON.stringify(accountStateToWire(...)) is order-dependent and silently breaks if accountStateToWire's key emission order changes or if a new conditional field gets added. Swap for a accountStateEquals(a, b) helper that compares semantic fields explicitly. Same gotcha applies to cross-session findAccountByIdAcrossSessions matches — when we early-return without accounts.set(key, state), the current session never gets a local copy. Fine today because handleListAccounts reads via accountsForPrincipal, but a future caller using getAccountMap directly would miss it.
• 3.0-compat filter chain bypass at server/src/training-agent/account-handlers.ts:1140-1150. When preferFixtureAccounts === true, req.sandbox / req.status / exactAccountFilter are skipped — mirrors prior behavior, but worth a one-line comment so a future reader doesn't "fix" it.

Minor nits (non-blocking)

1. Double-cast on brand in seedAccountFixture. server/src/training-agent/account-handlers.ts:497 re-casts brand to { domain: string; brand_id?: string; name?: string } even though brand?.domain was narrowed truthy above. A typed local would drop the cast.
2. The fixture-name domains (pagination-acme-1.example, etc.) drift from the test-kit's own acme-outdoor.example brand. Not wrong — the storyboard just wants three distinct sandbox accounts — but the prefix pagination- makes it harder to grep the storyboard back to its test-kits/acme-outdoor.yaml kit. Non-blocking.

Approving on the strength of the schema-pattern match against existing seed_* scenarios plus the round-trip test covering both 3.0 and 3.1 compat modes.