{
  "schema": "anvil-serving.benchmark-public-run-plan/v1",
  "finalist_controls": {
    "mode": "text-only",
    "vision": false,
    "context_tokens": 32768,
    "visible_token_budget": 4096,
    "reasoning_headroom": 8192,
    "combined_request_cap": 12288,
    "temperature": 0.6,
    "reasoning_effort": "medium",
    "configured_concurrency": 4,
    "request_concurrency": 1,
    "shared_kv_tokens": 131072,
    "top_p": 0.95,
    "fixed_seed": null
  },
  "workloads": [
    "three-repetition independently graded coding fixture",
    "18-scenario tools and recovery scout"
  ],
  "decision_boundary": "no-promotion",
  "execution_order": [
    "QUASAR preflight, agentic scout, three-repetition coding",
    "unload QUASAR",
    "fresh Swift15 preflight, agentic scout, three-repetition coding",
    "unload test profile and restore original Swift15 262K"
  ],
  "posthoc_analysis": "Tuple-edge representation sensitivity was added after frozen grading; no prompts, cases, expected outputs or model responses changed."
}