[MaxGhenis]

Round 8's two low findings on #580 (merged as 172f690f), with the mutation record rerun on the new source.

Changes

• L1: _event turns a RecursionError from the JSON parser into LookupError, as _parse_registry already does. Before, an event nested about 200,000 levels deep let RecursionError through. That broke the rule that an unreadable event raises LookupError, though it failed closed. New test test_a_too_deeply_nested_event_raises covers a push and a pull_request.
• L2: test_a_zero_push_boundary_needs_a_created_branch now runs on 40- and 64-digit all-zero befores. Before, a fallback limited to 40 digits survived both modules.
• N3: base_commit's docstring now says the zero before falls back only with created set to true.

Invariants (unchanged)

• An event that cannot be read raises LookupError and never another exception type.
• A push is compared with its own boundary: its before, or the merge base only for an all-zero before with created: true.

Checks

• Each new test fails on its mutant and passes on the fix:
  • RecursionError dropped from the handler: 2 failed;
  • the fallback limited to 40 digits: 4 failed.
• Registry module: 222 passed. Committed-registry module: 8 passed, 2 skipped. Serial, single file each, at afcbf820.
• Black and ruff are clean.
• Tiers: unit 18,789 + 6 = 18,795, total 25,365. The README table matches.
• CI: gh pr checks exits 0 at e83fb252.
• Mutation record, docs/design/track_b_holdout_registry_mutations.json (e83fb252): all 79 checks rerun on the source at afcbf820 (SHA-256 0b31ff14…), one at a time in a clean clone, single core at nice 10. 79 killed, none survived; the source was restored after each.
  • Check 76 ("an unset event path leaks TypeError") is re-anchored to the handler tuple that now includes RecursionError.
  • Two new checks: "a too deeply nested event leaks RecursionError" (2 tests fail) and "a 64-digit zero push boundary falls back without created" (4 fail).
  • The zero-boundary check's failed ids now carry the 40/64 parameter. Every other check's failed-test list is unchanged.
• Independent review naming the head: round 1 (Opus 5.5, the Codex lanes being closed) APPROVED e83fb252.
  • L1, L2 and N3 are resolved in substance. Every unreadable event it tried raises LookupError, including arrays 100,000 to 5,000,000 deep.
  • A differential of this head against master over 1,592 well-formed and malformed events found no difference. The only syntax-tree change is _event's handler tuple.
  • The record matches as the commit message lists: all 145 unique failed ids collect, and 13 sampled checks reproduce their failed ids exactly.
  • Tier arithmetic is correct.
  • One new low, nonblocking gap, L-A: no test asserts the created: true fallback for a 64-digit all-zero before, so a zero check that knows only 40 digits survives. It fails closed, and this repository uses SHA-1. It goes to a follow-up with nit N-C (each check's command omits the id-recording plugin).
• Coordinator's tier-count slot: requested.

🤖 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 27, 2026 11:08am UTC