The second-family reads of #341, #356, #363 and #380, folded in
Merged
DO
22d ago+44+44+01234567890123456789-11-11-01234567890123456789
The second-family reads owed on four merged pull requests, folded in: #341, #356, #363 and #380. Argon read each squash against its parent after it merged; Muse had read them before.
Taken
| PR | Finding | What changed |
|---|---|---|
| #380 | An approval submitted in the same second as the last reopening, or one whose time cannot be read, had no test at the bound: >= for >, or dropping the str guard, survived | tests/test_approve.py::test_an_approval_from_before_a_reopening_is_not_recovered holds both; two mutants |
| #380 | An approval sent on a pull request closed under the act was checked for its fault and key, not its state: failed → refused survived | test_a_pull_request_closed_under_the_act_claims_no_approval checks the state; one mutant |
| #341 | A permission read answering an id of True typed the account of id 1, and a resolver of id True was told as an identity of id 1, since True == 1: the guards had no case | One case each in tests/test_adapters.py and tests/test_reconcile.py; two mutants |
| #363 | The cutover's rule read as held, though its test holds only that the set exists | Decision 0010 and evals/AGENTS.md say the cutover's own gate waits on release/cutover.py, as 0010's row above already says of its side |
Not taken
- #341, finding 1: a resolution no Run recorded, called unmatched when the run records cannot be read whole. Left as decision 0067 rules ("run records not read whole bind none, and leave reconciliation short rather than calling the resolution unmatched"). The Run that recorded it may be in the part not read, so calling it unmatched would be a false alarm. And
reconciliation_unreadablealready says the pass was not read clean, so nothing goes silent. The Builder's ruling. - #341, finding 2: a run record not held to its schema raises.
reconcile's contract says the caller holdsrunsto their schema before passing them. - #341, finding 3, in part: the three guards on a thread's author in
adapters/threads.py. These mutants are equivalent._threadreads the opener throughAccounts.actorfirst, and that refuses a non-object author, an id ofTrue, and any type but User or Bot. - #363, finding 2: the manifest's count against its list. Already held by
test_the_manifest_counts_its_cases_and_holds_no_case. - #356: Argon found nothing. Muse's gap, no test or mutant for a Lens actor, was closed in #356's own review round by the mutant "a Lens's write access is read from a permission".
Verify
make verify: 1,298 tests, OK.python3 -m tools.mutants harness/approve.py: 21 of 21 killed.- The five new mutants, run alone: each killed by its test.
- Every mutant applies once.
No record: nothing here changes a rule.
🤖 Generated with Claude Code
Summary by CodeRabbit
-
Tests
- Expanded checks for approval handling around reopening times and pull requests closed under an approval act.
- Added coverage to ensure boolean identifiers are not mistaken for numeric account IDs when matching accounts or thread resolvers.
- Added mutation cases to verify these approval and identity-handling rules.
-
Documentation
- Clarified that the evaluation-set requirement is checked by both a named test and the cutover gate.