[bokelley]

Per-offering exclusions for #3504. It stacks on #3546 (naming) → #3545 → #3544 → #3543 → #3542 → #3534 and the rest of the #3504 stack. Review only feat(reporting): write per-offering exclusions across calendars. It targets main because CI runs only for PRs into main.

#3545 made the coverage function honour excluded[].offering_id. This slice makes the seller engine write it.

What changes

• Per-offering entries. When a condition covers offerings with different period boundaries, the engine writes one excluded[] entry per offering: offering_id plus that offering's cutover instants. Before, it wrote one account-wide entry cut at the latest end across calendars. This follows core/reporting-subscription-coverage.json: "when the account's covered offerings have different period boundaries; from and until are then that offering's cutover instants". It matches the canonical vector max-accounts-capacity-per-offering. One calendar, or several with the same schedule, still gets a single entry.
• Which conditions: capacity, excluded_by_buyer, action_required and overridden, across the subscription's calendars. Also a buy's offering_unsupported after a resolution change, across the calendars of the offerings it lost.
• Closing. A per-offering entry closes at its own offering's cutover, or at the configuration's active_until for an override.
• Release gate. perOfferingExclusions is read from the served bundle: excluded[].offering_id in reporting-subscription-coverage.json, which 3.3.0-beta.2 defines. A release without it gets the entries folded back into one entry cut at the latest instant, which is what the engine wrote before.
• Test. Capacity under match: all across a UTC calendar and an account-timezone calendar writes four entries, from 11-02T00 for the UTC offerings and 11-02T05 for the New York one. The buyer's computed owed set agrees. The folded projection is one entry at the latest instant.

Checks

npm run typecheck, lint:boundaries, check:package-size, Prettier, and 1,281 tests across reporting, principal, server-decisioning and packages/reporting-storage (PostgreSQL suites run in CI).

Refs #3504

🤖 Generated with Claude Code

[aao-secretariat]

Ladon verdict: Approve

Approve — clean review, no blocking or medium findings.

The reviewer examined the designated slice (commit 8997e1d, "write per-offering exclusions across calendars"; 4 files, 140 insertions). Key checks:

• engine.ts: perOfferingExclusions splits excluded[] per offering only when calendar keys differ; calendar resolution uses the correct offering_id-keyed Map; the settleExclusions Iterable→ReadonlyMap signature change has a single consistent caller.
• release.ts: collapsePerOfferingExclusions folds entries back for releases without excluded[].offering_id; grouping by canonicalize(condition) works because stripExclusion removes the per-offering key; from takes the latest cutover (pre-PR behavior); open offerings keep the folded entry open.
• Changeset: minor is correct (experimental surface, gated release field), and the DR-0033 fold-back pathway is present with a test.

No critical/high/medium findings. gated_paths is false, high_risk is false, no no-auto-approve team match, and no prior decision. Although review_decision is REVIEW_REQUIRED, row 2 only applies when gated_paths is true — it is false here, so no hard gate fires. None of rows 1–8 match; the outcome falls through to row 9: approve.

The merge-base changed after approval.