{"content":"GT#87 follow-up fix 1/5: quota-stall resilience for fleet-wide 429 token-plan exhaustion. GT#87 ended its 5.5h live observation parked on fleet-wide quota exhaustion: every member turn failed with '429 ... Token Plan quota has been exceeded', each attempt burned ~10min of provider retry ladders, and the 5-attempt retry budget was on track to DROP the three closing-gate triggers (round-5 diff, semantic-equivalence drill, ops re-verify). Quota failures are environmental, not member failures. Fix (commit c139a3a7): quota failures requeue UNCHARGED and arm a per-task quota stall; while stalled the deferred drain holds every entry and releases ONE probe dispatch per interval (default 15min); the owner gets one anomaly notice per episode through the origin session ('the task is parked, resumes automatically once the quota resets'); the first successful turn lifts the stall; past 24h the ordinary charged ladder resumes so the episode still terminates. Detection is an upstream error fingerprint (429 + quota), never natural-language matching. Suite 260/260 green.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}