[MaxGhenis]

Follow-ups to #560, whose round-7 review approved it and left two low findings and one optional hardening.

Changes

1. Unset event path. With GITHUB_EVENT_PATH unset, removing TypeError from _event's handler survived both registry test modules. test_an_unset_event_path_raises now covers a push and a pull_request event with the path unset; each raises LookupError.
2. Zero push boundary (hardening). base_commit lets an all-zero before fall back to the merge base only when the push event also has created: true, which is how GitHub marks a push that creates its branch. Any other all-zero push raises LookupError, so the check never compares with a nearer commit and checks less history. test_a_zero_push_boundary_needs_a_created_branch covers created missing, false, "true" and 1.
3. Mutation record ids. Check 73 recorded two parametrized ids cut at a nested bracket. The runner now takes each failed id from pytest's own report.nodeid through a small reporting plugin. The whole record is rerun on the merged source (4688ec5c), with two new checks for the changes above: 77 in all, all killed.

Plan

• Code and tests (2d081af)
• Registry modules and birth-evidence artifact test, serially, one file at a time, at b111ad9f: 216 passed; 8 passed, 2 skipped; 13 passed (237 passed, 2 skipped)
• Mutation record rerun on the merged source (4688ec5c): 77 of 77 killed; restored-source SHA-256 9ca7fc52839d637f29596f369578ff255a528114fe8956c767e4fb3d84ddac5d. The run was resumable and restarted twice from its saved checks (a host disk hold and a memory-pressure stop), against the same source SHA-256, which the master merges leave unchanged.
• Tier recount: unit 18,339 → 18,348 (+9) against master 9967dfd6, merged at 8e195c41 and recounted at 2545856b (whole-suite --collect-only -m unit: 18,348 of 24,897)
• Tier resolution after merging master d8302233 (Parity ledger: CE0, OP0.1, the registry, U2's evidence record and F0/F1 status #587, PSID income and wealth readers and field registry (D1a) #575, SM1: differential mortality by lifetime earnings, education and race (normalized within age-sex-year cells) #570) at 9ccbbb12: unit 18,549 + 9 = 18,558, total 25,128, README table matched; CI green at 9ccbbb12, its full collection confirming the count. Master's changes touch neither the registry module (standard library and yaml imports only) nor its tests or specs.
• Tier resolution after merging master 0f69bbb2 (OP1: open-population harness around the sealed loop #566, Fix review r10's report findings for OP1 #589) at b111ad9f: unit 18,780 + 9 = 18,789, total 25,359, README table matched. CI's full collection confirmed it: green at b111ad9f. Master has since moved to 9cced8ff (Ledger: SM1, OP1 and D1's first part merged, with the in-review PRs they touch #590), which changes one docs file and neither tier file.
• Independent delta review (round 8) at b111ad9f: APPROVE (Opus 5.5; the Sol attempt stopped when the model was at capacity). All three round-7 items are resolved in substance. It left two low, nonblocking findings for a follow-up: _event lets a RecursionError through on a deeply nested payload (unchanged since Track B: held-out data registry for every Track B and open-population gate #560, and it fails closed), and the 64-digit all-zero before has no test.
• Ready for review

Invariants

• A push is compared with its own boundary: the event's before commit when it resolves, the merge base only on a push that creates its branch, and LookupError otherwise. No payload selects a nearer boundary.
• An event that cannot be read raises LookupError on both push and pull_request, never another exception type and never a silent skip.

axiom: n/a: test infrastructure for the held-out data registry

Check numbering: the messages of 4688ec5c and fca34901 count checks from 0. Counting from 1, as the record and the reviews do, the two new checks are 76 and 77, and the re-anchored ones are 26 and 70.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
social-security-model	Ready	Preview	Oct 26, 2026 6:59am UTC

[MaxGhenis]

Merged at head b111ad9f with --match-head-commit, as merge commit 172f690f.

• Checks. Every check passed on that head (gh pr checks exit 0; deploy skipped). The PR was MERGEABLE and CLEAN, not a draft, with no change requests.
• Review. Round 8, by an independent Opus 5.5 reviewer run through Subfleet, approved b111ad9f. It raised two low, nonblocking findings, which go to a follow-up PR:
  • L1: a payload nested 200,000 levels deep raises RecursionError, which fails closed;
  • L2: no test covers a zero before.
• Tier counts. Unit 18,789 and total 25,359, as tests/tier_counts.json records. The three commits on master that this head lacks all come from Ledger: SM1, OP1 and D1's first part merged, with the in-review PRs they touch #590 and change only docs/parity-ledger.md, so the count holds without a re-merge.

This PR held the tier-count slot. The next slot holder re-merges master and recounts.