File the class a correctness review structurally cannot see: a mechanism outliving the population it was built for

[gunbai-bot]

Open, not merged. One file: a RecurringFailureMode row. Membership is the directory, so no registration edit exists to make.

The class

A mechanism lands for a population a root fix has already removed. A change that classifies or handles an observed failure signature merges after another change eliminated that signature at its root. Nothing in the declaration is false — the population was real when observed — and the mechanism is correct, complete, tested and reviewed. Only the world changed, in another change, between observation and merge.

Why this is not "we shipped something unnecessary"

The harm is that correctness review structurally cannot see it. On the specimen the mechanism was an auto-admission arm inside a required gate: it could not go red, could not be falsified by any run, and would thereafter have been cited as coverage of a class it never sees. That is §5's populationless arm — worse than an absent one, because absence is visible.

Base staleness is the mechanism, not the context. A required run's evidence is pinned at push against the base as it then was. It can answer is this correct with full rigour and cannot answer is this still needed. No participant is negligent and the defect lands anyway.

Specimen

root fix	#10856 aba7939b54 — 2026-09-08T18:51:35-04:00
the arm	#10871 d2559c36e3 — 23:19:18, 4½ h later
reviews	four, all approving
firings after landing	0, across three floor runs on two branches
revert	#10889

Recognition rule

After landing, grep the required gate's own log for the symbol you added. Zero firings is the finding. One grep, and it is what none of the four reviews did.

Two supporting tells available before merge: the motivating artifact's base is not an ancestor of current main; and the mechanism's only executing consumers are the fixtures shipped with it — which reads as full coverage and is the same reading either way, since a fixture cannot distinguish a live population from an empty one.

Why its own row, and not folded into the frontier row

declared_frontier_outlives_the_firing_of_its_own_trigger (#10842) is a declaration that announced it was waiting — admitted under §3c with a named later consumer, detectable by re-reading the trigger beside it. This is a mechanism that announced it was already needed: no frontier, no trigger, undetectable from the declaration at all.

Same family, opposite tense. Folding them would be a §3 meaning fork — one name over two materially different invalid states — and would blunt the rule that just fired.

Ceiling, stated below where it looks

Deciding is this still needed in general is asking whether a population is empty, which is not decidable from the declaration. Structural is not the ceiling, and claiming it would be inflation. What is decidable is the tell: a landed symbol with zero firings in the gate's executed log over a window — a lens over an executed artifact, rung 2 at best.

It sits at rung 1 today, mitigation, because nothing runs it: the tell was performed by hand on the specimen and no enrolled check performs it. The next-rung trigger is that lens existing and executing on the merge path — not someone remembering to grep.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GwvzQ3SQoWWDgHkAeffMeF