{
  "schema": "anvil-serving.benchmark-decision-summary/v1",
  "campaign_id": "cancelled-attempt-2",
  "assembled_at": "2026-10-25T02:59:40.633061+00:00",
  "source_revision": "890d51987f40cf02cfa8822fcd9f456c419baa7c",
  "source_sha256": "6fc9a0a5649cfcbb58fed7f2439037e13cc0bc0bc2bc95eb2b3a75084e9f43f2",
  "complete": false,
  "status": "cancelled_partial_pilot",
  "decision": "Retain the failed attempt for harness diagnosis; no model ranking or promotion.",
  "evidence_labels": [
    "diagnostic-only",
    "partial"
  ],
  "decision_labels": [
    "no-promotion"
  ],
  "promoted": false,
  "promotion_authorized": false,
  "authoritative_public_report": "campaign-report.json",
  "jobs": {
    "attempted": {
      "count": 9,
      "known_count": 9,
      "unknown_cells": 0
    },
    "cancelled": {
      "count": 1,
      "known_count": 1,
      "unknown_cells": 0
    },
    "completed": {
      "count": 6,
      "known_count": 6,
      "unknown_cells": 0
    },
    "failed": {
      "count": 2,
      "known_count": 2,
      "unknown_cells": 0
    },
    "planned": {
      "count": 96,
      "known_count": 96,
      "unknown_cells": 0
    },
    "scored": {
      "count": null,
      "known_count": 6,
      "unknown_cells": 1
    },
    "submitted": {
      "count": 9,
      "known_count": 9,
      "unknown_cells": 0
    },
    "unfinished": {
      "count": 87,
      "known_count": 87,
      "unknown_cells": 0
    }
  },
  "planned_batches": 60,
  "released_batches": 8,
  "unrun_batches": 51,
  "workload": {
    "configured": "12 tasks x C1/C4 x 2 repeats x 2 arms; fixed deterministic order; cache uncontrolled. Diagnostic sample, not complete model quality coverage.",
    "measurement": "Six native C1 incumbent jobs completed; two Aider launches failed before adapter entry/graded results; one Aider child cancelled before activation. Candidate arm unrun."
  },
  "controls": {
    "identity": "identity-control-map.json",
    "plan": "projections/campaign-control-records-0001.jsonl",
    "prepared_children": "projections/child-evidence-records-0001.jsonl",
    "configured_not_measured": "Planned concurrency, token/deadline budgets, runtime settings and GPU counts do not prove achieved concurrency, utilization or capacity."
  },
  "failure_diagnostic": "failure-diagnostic.json",
  "restoration": "restoration.json",
  "limits": [
    "96 planned jobs remain in the denominator:9 submitted,6 completed,2 harness failed,1 cancelled,87 unrun. Six native passes are not a6/96 campaign quality success rate.",
    "Job-level unrun count87 differs from sample-level unfinished89, which includes the two failed Aider partial jobs. Scored aggregate stays null with six known and one unknown.",
    "Failed Aider quality remains unknown. Their verified drain and four-plane telemetry do not repair the final cancelled pre-activation interval.",
    "Total active wall and request overlap remain unknown;437.0491850050166 seconds is only the sum of known intervals.",
    "Native six-cell token and latency totals exclude other campaign work. They are not TTFT, decode throughput, or whole-campaign timing.",
    "Sparse engine-zero samples do not prove no compute; the later zero observation cannot repair the unknown interval.",
    "Configured models and GPU counts differ; candidate has no measurements, so no ranking or engine-causal inference is available.",
    "No serving-comparison metric receipt or completeness claim is supplied. Private originals remain authoritative; sanitized derivatives cannot revalidate native hashes.",
    "The source repair is uncommitted and pending independent review at this cutoff; no successor campaign acceptance is claimed."
  ],
  "publication_status": "Final wrapper review and public matrix publication pending.",
  "unreleased_started_batches": 1,
  "timing": {
    "active_wall_seconds": null,
    "child_active_wall_seconds": null,
    "known_active_wall_seconds": 437.0491850050166,
    "observed_job_overlap": null,
    "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": 1,
    "unknown_intervals": 1
  },
  "native_completed_subpopulation": {
    "completed_native_cells": 6,
    "model_calls": 17,
    "prompt_tokens_sum": 6607,
    "completion_tokens_sum": 1433,
    "call_latency_ms_sum": 11848.445171955973,
    "harness_elapsed_s_sum": 11.852925054263324,
    "scope": "First six native C1 incumbent cells only; excludes preflight, drain, telemetry finalization, failed Aider and unrun cells.",
    "definition": "Call latency is the sum of retained turn.latency_ms; harness elapsed is the sum of native suite budget.elapsed_s. Neither is TTFT, decode throughput, full campaign wall time or a matched-arm comparison.",
    "evidence": [
      {
        "container": "projections/child-evidence-records-0001.jsonl",
        "record_id": "child/job-0001/evidence/0-agentic.json"
      },
      {
        "container": "projections/child-evidence-records-0001.jsonl",
        "record_id": "child/job-0002/evidence/0-agentic.json"
      },
      {
        "container": "projections/child-evidence-records-0001.jsonl",
        "record_id": "child/job-0003/evidence/0-agentic.json"
      },
      {
        "container": "projections/child-evidence-records-0001.jsonl",
        "record_id": "child/job-0004/evidence/0-agentic.json"
      },
      {
        "container": "projections/child-evidence-records-0001.jsonl",
        "record_id": "child/job-0005/evidence/0-agentic.json"
      },
      {
        "container": "projections/child-evidence-records-0001.jsonl",
        "record_id": "child/job-0006/evidence/0-agentic.json"
      }
    ]
  }
}