Cold review round 1 (two independent agents) found the first draft narrowed BOTH
readers of the verdict row, not just inheritance.
* `read_existing_verdict` is called twice. The second call is the last-moment re-read,
which asks "did a reviewer post a verdict while we were classifying" — the mirror-image
question. Gating the single flag made the job stop abstaining and post its exemption
straight over a mid-run row, and the post-write repair does NOT cover that: it is skipped
whenever the high-water mark could not be established. Reproduced red first, then fixed:
`test_the_MID_RUN_abstain_still_fires_for_a_NON_allowlisted_creator`.
* Splitting the flags broke the invariant that made the re-read work at all — the two calls
used to compute an IDENTICAL predicate, so anything firing at the second and not the first
had arrived in between. A pre-existing row declined for inheritance is still attributable,
so the re-read abstained on the very row the classification had decided to overwrite and
posted nothing. Seven tests went red on exactly that. The state/creator/description triple
from the first read is now snapshotted and compared, making "changed" explicit.
* The stated reason for keeping the post-write raced check broad was asserted, not checked:
a genuine reviewer is on the list by construction, so membership there would still count
every real verdict. Replaced with the real reason (the misconfiguration case — a second
human account forgotten from the literal) and the real cost (a verdict-shaped row landing
in the write window trips the sticky sentinel and costs that head its exemption).
* A `::warning::` now names the creator whenever the job re-derives over a verdict-shaped
status from an off-list account, so a reviewer left off the list is not discarded silently.
* The list guard uses `findall` + a single-assignment assertion: `search` validates the
FIRST assignment while the shell runs the LAST executed one.
* Mutation counts dropped from the decision record in favour of test names — nothing
re-executes a number written there.
Mutation-proven: mid-run site back to `ex_human` reddens the mid-run test; dropping the
changed-row comparison reddens the inheritance cases; `if false` at the abstain reddens both
mid-run tests.
fixes#742
Decisions-Edit: yes