handoff: four of five KNOWN-BROKEN rows were stale, in the file a successor reads first - #646

…cessor reads first

Attempting #451's close condition — "every workaround is in a durable artifact, or recorded
as deliberately left" — turned up one unrecorded item and, on the way, a table that had
outlived its defects.

⇒ THE AUDIT: 7 of 8 §1–§2 items ARE durable (merge-guard's three legs are in the tool,
the two-dot trap is in leg 4's comment, `git cherry`'s patch-id collapse is in
stranded-branches, the SUBJ_DIR form is in gate-selftests itself). ONE was not.

⛔ THE UNRECORDED ONE, and the only place it lived was a QUARANTINED file. #451 §1 records
"reading a pane's context % needs `lines: 4` on terminal.getStatus — 1 and 2 omit it". The
sole match in the tree is `tools/teamlead/boxwatch.sh`, which sits under QUARANTINE.txt with
its belonging an open question. ⇒ A fact recorded only in a quarantined file is not recorded.

⚠ AND THE NUMBER IS WRONG. Re-measured against a live pane, one call per value:
    lines 1 -> no %   ·   lines 2 -> no %   ·   lines 3 -> 76% PRESENT   ·   lines 4 -> present
The boundary is 3, not 4. The `[re-verified]` tag travelled with a figure one too high.
★ And a fixed number is the wrong thing to record: the % lives in the status BLOCK, whose
height depends on what the pane renders. So the entry says "ask for >= 3 and CHECK the % is
in recentOutput", not a constant — a constant rots silently, because a short read returns a
well-formed status object with the percentage simply absent.

⛔ FOUR OF FIVE KNOWN-BROKEN ROWS WERE STALE. Each said "⛔ live on main":
    index-watch --self-test hangs      -> exits 0 in 13.3s
    24 of 48 controls establish nothing -> 48 passed · 0 UNESTABLISHED · 1 UNVERIFIABLE
    bootstrap-audit's control FAILS     -> 0 FAILED, its own control block passes
    use-not-mention is UNVERIFIABLE     -> exits 2, not 0
Struck rather than deleted; the fifth (estate vocabulary is a closed list) still holds, and
the real UNVERIFIABLE is now pretooluse-guard.py, which the gate names.

★ AND THE BARE COMMAND WAS THE WRONG COMMAND — #451 §1's own warning, applied to my own
measurement mid-audit. `bash scripts/gate-selftests.sh` reports `ran 6 subject(s)`;
`SUBJ_DIR=tools ./scripts/gate-selftests.sh`, the form CI runs, reports `ran 63`. Same
minute. A row citing the bare form measures a tenth of the population it claims to.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>

…s six

Review finding, and it is right: my edit ADDED a row (pretooluse-guard.py, replacing the
struck use-not-mention.py), so "FOUR OF THE FIVE ROWS ABOVE" describes a table the reader
can no longer see. The sentence now says four of the five the table held BEFORE
2026-09-07, and names all six of what stands: four struck, one replacement, and
`estate vocabulary` — the one of the original five that still holds.

DECLINED — MD040, a language identifier on the fence. Measured: 26 fences in this file,
0 with a tag. Same ruling as #636 on the same file: tagging one of twenty-six makes it
less consistent, not more, and markdownlint does not gate here. If the house style should
be `text` everywhere, that is a whole-file change with its own PR.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>