{
  "schema": "anvil-serving.benchmark-decision-summary/v1",
  "campaign_id": "2026-09-19-qwen38-huihui-ninfer",
  "complete": true,
  "completion_scope": "Hard-gate rejection and restoration complete; full performance/long-context qualification was stopped, not passed.",
  "identity": {
    "configuration": "configuration.json"
  },
  "workload": {
    "manifest": "workload-manifest.json"
  },
  "measurement": {
    "path": "online direct",
    "warm_cold_state": "mixed sequential functional probes",
    "statistics": [
      "Repeated correctness pass counts only"
    ]
  },
  "gates": [
    {
      "name": "strict typed tool arguments",
      "candidate_off": "0/3",
      "candidate_reasoning_low": "0/3",
      "incumbent": "3/3",
      "status": "failed"
    },
    {
      "name": "unified diff",
      "candidate_off": "0/3",
      "candidate_reasoning_low": "3/3",
      "incumbent": "3/3",
      "status": "fails off, passes requested low"
    },
    {
      "name": "parallel triage",
      "candidate_off": "3/3",
      "candidate_reasoning_low": "3/3",
      "incumbent": "3/3",
      "status": "passed"
    },
    {
      "name": "image suite with vision enabled",
      "result": "12/12",
      "status": "passed functional; companion identity only"
    },
    {
      "name": "performance, MTP, long context, endurance, immutable recreation",
      "status": "not run; blocked by correctness gate"
    }
  ],
  "headline_results": [
    "Reject exact tested no-spec 8K served profile; retain incumbent recommendation.",
    "No demonstrated speed or footprint improvement.",
    "Media services restored HTTP 200."
  ],
  "failures": [
    "Required zip string tool argument failed every candidate attempt.",
    "Initial chat_template_kwargs rejected HTTP 400.",
    "Initial vision-disabled image run 0/12; corrected recipe run12/12."
  ],
  "limitations": [
    "Malformed raw tool arguments were discarded by harness; parser/template/model cause unconfirmed.",
    "Source-build apt/runtime dependencies are mutable; executable hash is local identity, not clean recreation proof.",
    "Incumbent context and runtime differ; correctness control cannot establish comparative performance.",
    "Other Huihui runtime/parser profiles and MTP remain unqualified."
  ],
  "decision": {
    "evidence_labels": [
      "functional",
      "quality",
      "compatibility-only"
    ],
    "decision_labels": [
      "no-promotion",
      "rejected-exact-profile"
    ],
    "promotion_authorized": false,
    "conditional_user_request": "Promotion requested only if better; conditions not met.",
    "next_gate": "Correct tool argument behavior in an exact pinned profile, retain raw diagnosis, then rerun qualification."
  },
  "evidence": [
    "multimodal-nospec-8k-vision.json",
    "multimodal-nospec-8k.json",
    "preflight-incumbent.json",
    "preflight-nospec-protocol.json",
    "preflight-nospec-server-control.json",
    "preflight-nospec-smoke.json",
    "quality-incumbent-control.json",
    "quality-nospec-8k.json",
    "quality-nospec-reasoning-low.json",
    "restoration.json",
    "independent-gate-review.md"
  ]
}