[bokelley]

Vectors slice for #3504, added at the coordinator's request: the canonical spec coverage vectors and the two bugs they expose. It stacks on #3544 (T3) → #3543 → #3542 → #3534 and the rest of the #3504 stack. Review only test(reporting): run the canonical reporting-subscription coverage vectors. It targets main because CI runs only for PRs into main. T4 (naming) stacks on top.

Upstream pin

• Source: test(compliance): canonical reporting-subscription coverage vectors and reference model adcp#8616, commit 023bbd26e7c230dcc74bea4087336ea6bc8d0f6a, path static/compliance/source/test-vectors/reporting-subscriptions/.
• Mirror: test/fixtures/reporting-subscriptions/canonical/, byte for byte, excluded from Prettier. PINNED.json records the source:

File	sha256
vectors.json (113 cases)	b392df771fc2bf635c5c441ff2dd5107e1cda13ab483fe902a6a56661f5e7904
case.schema.json	0550b51097ea2c9a6bbd8721de5b5855acfcbd097dff31edeb0df2f397f7dc42
error-codes.json	4e8e5142ada936fc8923735155d786a0801258922fb85b492785730479713481
reason-codes.json	2574b2a32fb8260fc7f9a7d9cd1eac1b588aa22c54a350e54964a893b80e77c4

• The parity test fails when a hash differs from the pin. npm run sync:reporting-subscription-vectors -- --commit <sha> [--from <spec checkout>] refreshes the mirror from GitHub raw or a local checkout, and rewrites the pin.
• The Python SDK runs the same vectors.

What changes

• Parity test. test/lib/reporting-subscription-coverage.test.js runs every case strictly: rows and grouped obligations, or expected_error code and detail. It also validates every case against case.schema.json. test/fixtures/reporting-subscriptions/coverage-cases.json is deleted.
• tzdata. The vectors are pinned to 2026f and hold from 2026b. On a runtime whose bundled tzdata predates 2026b, a failing case that depends on a non-UTC zone is reported as t.skip with a reason naming the runtime's release. UTC-only and expected_error cases always run strictly. Locally, Node 24 with tzdata 2026b ran all 113 cases with no skips. CI's unit lane uses Node 20.x, so any skips there show in the test output.
• Bug fix: excluded[].offering_id is honoured. An entry with offering_id removes only that offering's periods. It used to be applied to every offering, which dropped a period in max-accounts-capacity-per-offering. The coverage input type gains ReportingSubscriptionExclusionEntryV1 (offering_id?), and the engine's ReportingSubscriptionExclusionV1 carries the field.
• Bug fix: P1W is refused. A selector period_duration outside the schedule pattern now throws CoverageInputError COVERAGE_INPUT_INVALID / invalid_period_duration. It used to resolve no offering and silently owe nothing.
• Corrected cases. The five managed offerings that were schema-invalid in the ported cases (missing method.provider/method.format, receipt offerings without canonicalization_*) existed only in the retired coverage-cases.json. The other in-repo managed-offering fixtures already carry provider and format.
• numeric offset timezone is refused now uses an account_timezone offering and passes unchanged (invalid_timezone).
• Error test. test/lib/reporting-subscription-errors.test.js now reads the vendored error-codes.json. Each of its 14 codes is checked against the TS class it names, its recovery and its dispatched rule.
• Package budget. The packed-byte bump now lives in feat(reporting): typed reporting-subscription errors with code, recovery and dispatched #3544 (26,560,000). Measured at this slice with main (beta.2): 26,526,144 bytes.

Deviations and follow-ups

• Engine exclusions. The engine still writes account-wide exclusions, cut at the latest end across calendars, rather than one entry per offering. The coverage function and the engine's own coverage now honour offering_id wherever it appears. Writing per-offering entries is seller-side work for a later slice.
• Reason codes. reason-codes.json is vendored but nothing in TS emits trace reasons yet (T12). No conflicting names were introduced.

Checks

npm run typecheck, lint:boundaries, check:package-size, Prettier, and the reporting-subscription, principal and reliable-reporting suites (381 tests).

Refs #3504

🤖 Generated with Claude Code

[aao-secretariat]

Ladon verdict: Approve

Approve — no changes to review.

The head tree (df6845d) is byte-for-byte identical to the previously approved SHA (e7a74e9); git diff between them resolves to 0 files, with the intervening commits being stack merges that left the tree unchanged. No critical/high/medium findings in this pass, and the prior pass recorded no blockers.

Decision table: rows 1–8 do not fire (no findings, gated_paths false, high_risk false, no team gate, prior decision was approve). Falls through to row 9 → approve.

[aao-secretariat]

Ladon verdict: Approve

Approve — subsequent pass, clean.

The delta since the prior approved pass is a single hunk in scripts/check-package-size.mjs adding three explanatory comment lines documenting the AdCP 3.3 reporting-subscription coverage budget rationale (~26.37M packed bytes, ~40 KB variance allowance). No code behavior change, no cap-constant change in the delta — MAX_PACKED_TARBALL_BYTES and the #3544 cap bump sit above this hunk and were already present in prior approved passes. Documentation prose in a build script; nothing clears the Medium bar.

No blocking findings, no medium findings. No high-risk flag, no gated paths, no author-team gate. Rows 1–8 do not fire; outcome falls through to row 9 (approve).

[aao-secretariat]

Ladon verdict: Approve

Approve — subsequent pass on a clean delta with no blocking findings.

This is a follow-up pass; the prior decision was also a clean approve. The delta is 3 source + 3 test files refining the reporting-subscription engine:

• calendar.ts:9-18 — fail-closed refusal of sub-daily billing_cycle alignment (adcp#8622) instead of silent local-midnight alignment. Correct shape.
• engine.ts:798-810 — refreshes an open action_required entry's pending_action on change (canonicalize-compared, preserves from); latestPeriodEnd (engine.ts:42-46) replaces withheld.at(-1) safely since periods are appended out of end order and periods[0]! is guarded by the caller.
• reporting-subscriptions.ts:89 (!== true → === false) correctly preserves a pending replace retry across flag-omitting writes.

Each change has a dedicated test in the same delta and changesets are present. No Critical/High/Medium findings.

Decision table: no critical/high findings (row 1 N/A); gated_paths false (row 2 N/A); high_risk false (rows 3/5 N/A); no medium findings (rows 4/8 N/A); no no-auto-approve team match (row 7 N/A); prior decision was approve so sticky escalation (row 6) does not apply. Falls through to row 9 → approve. review_decision REVIEW_REQUIRED does not block approval since gated_paths is false.

The merge-base changed after approval.