Files
ersatztv/scripts/tests
timothy 9ed30a6b7c fix(742): the allow-list governs an inherited success, NOT a failure
Round-2 cold review (two independent agents) returned BLOCKED. Both found the same class
of hole from different angles.

FAIL-OPEN, and a regression against main. The short-circuit branched on success OR failure
under one membership test, so an off-list `failure` was re-derived. Trace it: a real
reviewer left off the literal posts BLOCKED; ex_human=no so the short-circuit declines; the
PR is docs-only so the classifier exempts it; the re-read triple never changed so the
mid-run abstain declines; the row predates the high-water mark so the post-write repair
cannot see it. The job posts success over an explicit rejection — reached by the very
misconfiguration the allow-list's own diagnostic is about. `failure` now needs only
attributability, which costs nothing the allow-list bought: inheriting a rejection cannot
green anything, and the worst a forged one achieves is a stall that any human clears in one
command (post-review-verdict.sh POSTs unconditionally).

THE CHANGED-ROW COMPARISON WAS NOT PINNED. A presence-only version (`[ -z "$pre_state" ]`)
passed all 158 tests, because every mid-run test started from an empty head and exercised
only the ARRIVAL case. The REPLACEMENT case is the one that co-occurs with #742's own
scenario. Stub knob STUB_PRE_ROW plus a test; the presence-only mutant now reddens.

Also from the round: the ::warning:: fired with a self-contradictory message for an
allow-listed creator in a non-terminal state, and interpolated an API-supplied value into a
workflow command unstripped; a quoted test count that no longer matched the tree; the
function's own contract line omitted the flag this change adds; a comment naming ex_human
where it is now ex_attributable; and an unprobed claim about how Gitea renders ::warning::,
now withdrawn rather than restated.

MEASURED rather than asserted: two POSTs of context probe742/combined-newest to one sha
(pending, then success) came back as ONE row on /commits/{sha}/status — the newest — and as
BOTH on /statuses/{sha}. That claim became load-bearing with the changed-row comparison;
main depended on presence alone. The probe left the sha's combined state `success`.

Recorded, not silently gained: an attributable row in a state other than success/failure
used to stall a PR forever on main (neither short-circuit fired, the old unconditional
abstain did). The changed-row comparison removes that stall.

scripts/tests: 969 passed, 2 skipped.

fixes #742

Decisions-Edit: yes
2026-08-27 00:07:43 +02:00
..