Five verdicts in one week reported an outcome the mechanism had not decided

Not five bugs. One shape, found five times in seven days, in three tracks' files. Filing the sweep instead of a sixth individual fix.

The shape

A verdict has fewer slots than the subject has outcomes, so the missing case takes the shape of whichever neighbour the control flow happens to reach.

It is not a bias toward alarm or toward comfort — it is an absence with no slot. The same missing distinction produced opposite errors depending on which loop it fell out of:

instance	outcomes in reality	slots	the missing case became
`pasmith_run.recheck`, multi-unit	reproduces / fixed / could not regenerate	2	"still reproduces" — manufactures work
`pasmith_run.recheck`, zero examples	same	2	"FIXED (0 seeds agree)" — manufactures a green
`pasmith_run.localize`, sentinel result	a checkpoint / no location exists	1	a confident wrong checkpoint
`pasmith_run.ledger_status`	what throttles / what is open	1	reported the set that did not govern
`recheck`'s population	throttles? / re-measure?	1	`dodged` never re-measured at all

Two more from the same week are the prose variant — an assertion with nothing re-deriving it, rather than a missing slot — and they belong in the same sweep because the repair is the same kind of thing:

tools/gui_shot.sh BLANK_MAX=4000, on the note "a blank frame is ~1-3 KB". True when written; by 2026-08-30 an empty display was 4013 B and a real xterm window 4068 B. Fifty-five bytes apart — the proxy was dead, not mis-calibrated, which is a different ticket with a different fix, and only measuring both sides showed which.
compiler/defs.inc:1149 — a comment describing the slot, not the variant (filed as bug-a-defs-inc-vt-promo-comment-describes-the-slot-not-the-variant).

Why it survives review

Every one of these reads as routine at the call site. reproduces = True under a comment saying "cannot judge: keep it open, loudly" looks like the careful choice — and the code was not loud, it printed the same string as a genuine reproduction. A report is a claim, and nobody re-derives a claim that reads as routine. The two halves drift because nothing imports one into the other.

The latch is what gives it teeth: a false "still reproduces" is not one wrong verdict, it is a wrong verdict that cannot ever become right, because the arm that would flip it is the arm that is broken. It gates work, it gates itself, and nothing in it decays.

The sweep

Track T tooling only — testmgr.py, twatch.py, pasmith_run.py, fuzz.sh, gui_shot.sh. For each function that returns or prints a verdict:

Enumerate the subject's real outcomes, including "I could not look". Compare with the slots the return type actually has. A bool has two.
Where a set answers two questions, check both answers are wanted. ledger_open was right about throttling and wrong about re-measuring.
Where a report names a location, a cause, or a number, check the deciding half computed it — not a neighbouring value that is usually equal.
Where prose asserts a measurement, re-derive it, and measure both sides of any comparison. One side alone cannot distinguish a mis-calibrated proxy from a dead one.

Bias every repair toward "I could not measure": that failure mode is recoverable, and a clean verdict from no data is not.

Not a rewrite

Each individual repair has been ~20 lines plus guards. The value is doing them in one pass with the shape named, so the sixth instance is recognised rather than rediscovered. Prio 30: nothing here is on fire, and every instance found so far was found while doing something else.

[[devdocs/dev/the-empty-tree-audit]] is the detection technique for the prose variant (a partition whose empty cell is the finding). The repaired instances are 10c8...-era Track T commits from 2026-08-29/30.

Gate: tools/*_devtest.py green for whatever is touched; pure guards preferred, each with a negative control that goes red for the right reason.

Deprioritised 2026-09-02 — the Track T tooling backlog was cut as a pile

This ticket is not being called wrong. It was moved as part of a pile, not judged individually, and nothing here disputes its finding.

Owner decision. 73 of the 74 open track: T tickets were filed between 2026-08-31 and 2026-09-02, 58 on one day. The pile was too large to work through and returned almost nothing, and a ticket nobody will fix does not sit neutrally — it stays in the ranker forever at zero value, which is the argument CLAUDE.md already makes for a terminal folder over a low prio.

Four were kept in the ranker on a purely structural test — an active umbrella or a hard blocked-by: edge from live work: umbrella-one-full-tier-run-with-no-red-tier, feature-t-freebsd-image-and-runner, and the two regression-test-core-* reds that block the umbrella.

Kept, not deleted, for two reasons: so the finding is not rediscovered and refiled from scratch by the next agent who trips over it, and so it can be pulled back if what it touches becomes load-bearing.