{
  "schema": "anvil-serving.benchmark-decision-summary/v1",
  "campaign_id": "mixed-c4-diagnostic",
  "complete": false,
  "scope": "One incumbent-only mixed-C4 measurement-closure diagnostic; excluded from full comparison populations.",
  "source": {
    "repository_revision": "d8f86a2208584b21f39c3cf600762142d7ad0057",
    "source_sha256": "4a87fc4fcd3d7f21b1a5f51197392f057f27ae103c09cb2792bb692656a17430"
  },
  "native_report": "campaign-report.json",
  "task_mapping": [
    {
      "cell": "cell-1",
      "task": "tool-recovery (synthetic error result)",
      "passed": true,
      "measurement_complete": true
    },
    {
      "cell": "cell-2",
      "task": "debug-loop-v2 (synthetic tool results)",
      "passed": true,
      "measurement_complete": true
    },
    {
      "cell": "cell-3",
      "task": "cpp/exercises/practice/all-your-base",
      "passed": true,
      "measurement_complete": true
    },
    {
      "cell": "cell-4",
      "task": "go/exercises/practice/alphametics",
      "passed": false,
      "measurement_complete": true
    }
  ],
  "counts": {
    "planned": 4,
    "submitted": 4,
    "completed": 4,
    "graded": 4,
    "task_passed": 3,
    "task_failed": 1,
    "measurement_complete": 4,
    "telemetry_rows": 60
  },
  "configured": {
    "concurrency": 4,
    "repeats": 1,
    "arms": 1,
    "total_budget_s": 900,
    "drain_reserve_s": 120,
    "temperature": 0.6,
    "top_p": 0.95,
    "reasoning_effort": "max",
    "thinking_mode": "default",
    "cache": "uncontrolled",
    "native_max_steps": 24,
    "native_max_completion_tokens": 8192,
    "aider_max_completion_tokens": 16384,
    "aider_max_transport_attempts": 12,
    "aider_retries": 0,
    "aider_upstream_edit_test_attempts": 2
  },
  "closure_observation": {
    "native_status": "closed",
    "drain": "verified",
    "planes": [
      "engine",
      "gpu",
      "host",
      "cgroup"
    ],
    "terminal_complete": true,
    "source_sample_counts": {
      "engine": 21,
      "gpu": 22,
      "host": 60,
      "cgroup": 60
    },
    "independent_terminal_review": {
      "status": "GO for repaired four-child mixed-C4 closure only",
      "reviewer": "independent Sol/high",
      "original_private_review_sha256": "49ee302e52c48f473b7ccd7c7df58b36752d26b0f8d066f9caf2198f8e33f8bb",
      "original_private_review_bytes": 8673,
      "earlier_GO_version_original_sha256": "7cc37b12955a5acc5ceacca3f281aede12b48d32e734acac383a0b82bae4effc",
      "version_delta": "Final review adds full-chain audit binding; earlier GO provenance retained.",
      "original_private_full_chain_audit_sha256": "63bbeb423295a78bdcdf64106ae881c27120dc91e5d07be791146c2109d6ffad",
      "public_full_chain_audit": "full-chain-audit.json",
      "bound_reference_count": 41,
      "source_python_file_count": 447,
      "source_and_reference_reads_unchanged_twice": true,
      "model_quality_ranking": false,
      "publication_approval": false,
      "promotion_authorized": false
    }
  },
  "timing": {
    "active_wall_seconds": 308.53438065899536,
    "child_active_wall_seconds": 292.2098744949326,
    "known_active_wall_seconds": 308.53438065899536,
    "observed_job_overlap": 4,
    "observed_request_overlap": null,
    "request_overlap_status": "unknown",
    "scope": "active wall includes release, verified drain and bounded telemetry finalization; child overlap is not request concurrency",
    "unknown_batch_intervals": 0,
    "unknown_intervals": 0
  },
  "limitations": [
    "Four-task diagnostic only; three passes are not a model-quality ranking or full96-cell result.",
    "Go alphametics remains failed after [false,false]; C++ native oracle outcomes [false,true]. No first-attempt root cause is assigned by this packet.",
    "Native debug/tool results are synthetic; Aider contained tests supply the executable coding oracle.",
    "Native request overlap remains unknown; process overlap4 does not prove four simultaneous model requests. No throughput, TTFT, speedup or percentile claims.",
    "Telemetry60 poll rows include repeated source observations: engine21 and GPU22 distinct source samples. Memory allocated does not prove compute.",
    "Go grader token totals are0 as emitted; not proof of zero generation. Two exhausted-context warnings do not replace exact finish-reason evidence.",
    "Aider process retained display prefixes are truncated; this is not the hard output-overflow failure from earlier canaries. Raw prompts/reasoning/output remain private.",
    "Native hashes inside sanitized objects bind private originals, never these projected bytes; original hashes and public file hashes are separate in projection-ledger.json.",
    "This compact packet is not the full transitive private evidence graph. Independent native-chain acceptance is recorded in terminal-review.json/full-chain-audit.json; publication acceptance remains pending.",
    "No candidate run, model swap, alias activation, promotion or prior R7 rehabilitation. Whole-experiment publication size approval is separate."
  ],
  "decision": {
    "evidence_labels": [
      "compatibility-only"
    ],
    "decision_labels": [
      "no-promotion"
    ],
    "promotion_authorized": false,
    "next_gate": "Independent public-packet review; any fresh full campaign has its own preparation/freeze/admission gates. No comparison or promotion accepted."
  },
  "request_accounting": {
    "scope": "Aider children only",
    "transport_starts": 6,
    "observed_http_response_events": 5,
    "cpp_starts": 4,
    "cpp_http_response_events": 3,
    "go_starts": 2,
    "go_http_response_events": 2,
    "usage_totals_claimed": false,
    "basis": "terminal-review.json and full-chain-audit.json; raw request events remain private",
    "limit": "Six attempts do not mean six successful responses or complete usage accounting."
  }
}