No illustration exclusion in _slow_path_eligible_cards_queryset: ~3,233 already-resolved cards enter the human queue on the first -v2 run

[WilfordGrimley]

What

_slow_path_eligible_cards_queryset has no illustration exclusion. Cards already resolved by the illustration deduction path (stage-d-illustration-v2) are not filtered out of the slow-path human review queue. The command's own comment admits the omission.

Measured: ~3,233 already-resolved cards would be enqueued for human review that does not need to happen.

Why it matters

It is harmless today and only today, because -v2 has never been run. It fires on the first -v2 run — which is the run currently being planned.

PR #604 (run-scoped eligibility + vote archive) does not close this. It changes eligibility scoping, not the illustration predicate.

The cost asymmetry is the whole argument: one .exclude() now, versus ~3,233 wrongly-queued cards, human attention spent on questions the machine already answered, and a review queue whose contents no longer mean what the queue claims they mean.

DECISION NEEDED (owner)

Fix the slow-path illustration exclusion before the first -v2 run?

• (a) Yes — add the .exclude() now, before the run.
• (b) No — run -v2 first and accept ~3,233 wrongly-queued cards, then clean up (note Retraction tooling doesn't clear stale slow-path review-queue rows (hygiene gap) #361: retraction tooling does not clear stale slow-path review-queue rows, so the cleanup is not free either).

What would close it

1. One .exclude() on _slow_path_eligible_cards_queryset removing cards with a resolved illustration-deduction verdict.
2. A test that a card resolved by stage-d-illustration-v2 is not slow-path eligible.
3. The stale comment in the command removed, since it will no longer be true.
4. Re-measure: the ~3,233 figure goes to 0.

LIVE vs LATENT

LATENT. Nothing is wrong in production right now. It becomes LIVE the moment -v2 runs for the first time.

Refs

• Established by the PR Audit: does the pipeline still catch what it aims to? The composition check #616 audit.
• Related: Run-scoped eligibility, the value-comparing split, and an archive for superseded votes #604 (does not close this), Retraction tooling doesn't clear stale slow-path review-queue rows (hygiene gap) #361 (retraction tooling doesn't clear stale slow-path rows), Persist survivor_pks candidate shortlists for ambiguous slow-path routings #433.