Planning bug: MOTIR-3455's evidence grep was never RUN — it reports 3 files where its own named ref returns 19

A card whose EVIDENCE BLOCK is a quoted command output that the command does not produce.

MOTIR-3455 states:

Verified on `origin/main@b944dab5`:

$ git grep -l "@/lib/jobs/registry'" origin/main -- '*.ts' '*.tsx'
  app/api/inngest/route.ts
  scripts/worker.ts
  tests/jobs/engine-dispatcher.test.ts

Run at that exact ref, it returns nineteen files, not three.

$ git grep -l "@/lib/jobs/registry'" b944dab5 -- '*.ts' '*.tsx' | wc -l
19
$ git grep -l "@/lib/jobs/registry'" -- '*.ts' '*.tsx' | wc -l      # at 18d60791
19

Wrong when written, not drift since. Both refs return 19, and git log --diff-filter=A dates all sixteen unlisted files — every one under tests/ — before the card was created (2026-08-25 19:09 UTC); the newest, tests/jobs/engine-units.test.ts, landed 18:32 UTC.

Why it did not cost anything, and why it is still a defect

The conclusion survives intact. Exactly two of the nineteen are non-test, and those two are what the argument rests on: no module on any production emit path imports the registry. The run confirmed the defect and the ADR amendment shipped.

It is filed anyway because the failure mode is the shape, not the outcome. An enumeration presented as a captured command output carries more warrant than the same claim in prose — it says I ran this. A card that quotes a plausible-looking output nobody produced spends warrant it has not earned, and the next reader has no way to tell that block apart from one that was run. Here the error happened to be harmless; the same habit applied to a count that gates a decision is how a card's number goes unchecked into an ADR.

It also had a second-order cost worth naming. The sixteen omitted files are not noise — they are seventeen suites that evaluate the definition modules in their own process, which is precisely WHY the defect stayed invisible. The wrong enumeration hid its own best supporting evidence.

The rule this suggests

plan-rules/phase-deepen.md's ENUMERATION limb already says a card counting a population owes the ref the count was taken on. This card DID name a ref. What it did not do is run the command AT that ref. The limb's sharper form: a fenced block presented as command output must be pasted from a run, never composed — and a card that names a ref makes the check one command long for the reader, which is exactly what happened here.

Fix

Nothing to build. MOTIR-3455 is amended on the record with the true count and the dating, and docs/decisions/job-queue-foundation.md §11 carries the correction in its own One enumeration correction section, mirroring §8's precedent.

Found by running the card on parent/MOTIR-3415-event-jobs-cutover.