{
  "schema_version": 1,
  "run_id": "flash-frozen-4k-c1-bc80cb97a2aa",
  "started_at": "2026-09-21T19:14:28.007566+00:00",
  "completed_at": "2026-09-21T19:15:05.484978+00:00",
  "status": "completed",
  "evidence_mode": "actual_inference",
  "role": "serving",
  "boundary": "role_completion",
  "phase": "performance",
  "suite_version": "static-performance-evidence-1",
  "suite_sha256": "53ddfba9e794833f0ddb92f9785dc1c012c566ef6d438df0b4dd153ca074860f",
  "evaluator_sha256": "2977430a164f9ef681d6cff4dd7e5826318f55d95d9ffe8740bce494dec505a9",
  "comparison_key": null,
  "deployment": {
    "concurrency": 4,
    "context_limit": 327680,
    "hardware": "Two NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation GPUs, 96 GB per GPU",
    "id": "glm53-flash-abliterated-rtx-tp2-high",
    "model": "GLM-5.3 Flash Abliterated",
    "node_count": 1,
    "profile": "Timing-only C1; eight naturally stopping requests per input target; shared endpoint",
    "quantization": "EXL3 4 bpw weights; NVFP4 MLA KV cache",
    "revision": "c8f58e6aa9117c73607d692978b22f091d80450c",
    "runtime": "vLLM",
    "runtime_version": "0.26.1rc0+infernal.invocation.cu133.r19.vllm174c789.b12x12c4263",
    "source_url": "https://huggingface.co/lovesenko/GLM-5.3-Flash-tr3-4bpw-Abliterated",
    "variant": "abliterated",
    "weights_verified": false
  },
  "counts": {
    "declared": 8,
    "primary_passes": 0,
    "outcomes": {
      "not_run": 8
    }
  },
  "metrics": [
    {
      "id": "completion_throughput_tps",
      "label": "End-to-end completion throughput",
      "value": 39.410854356583755,
      "unit": "tok/s",
      "samples": 8,
      "definition": "Observed completion tokens, including reasoning, divided by the full measured batch duration. Includes waiting, prefill, generation and relay latency; not decode TPS."
    },
    {
      "id": "first_generated_median_ms",
      "label": "First generated output, median",
      "value": 1246.246,
      "unit": "ms",
      "samples": 8,
      "definition": "First nonempty reasoning or content payload observed by the client. This is not server-internal token timing."
    },
    {
      "id": "first_final_content_median_ms",
      "label": "First final content, median",
      "value": 1558.7134999999998,
      "unit": "ms",
      "samples": 8,
      "definition": "First nonempty content payload. Correctness and usefulness were not graded."
    },
    {
      "id": "completion_median_ms",
      "label": "Request completion, median",
      "value": 1858.121,
      "unit": "ms",
      "samples": 8,
      "definition": "Full client-observed request duration, including shared-endpoint queueing and relay latency."
    },
    {
      "id": "batch_duration_seconds",
      "label": "Measured batch duration",
      "value": 15.300353413913399,
      "unit": "s",
      "samples": 1,
      "definition": "Client measurement interval excluding process setup; the throughput denominator."
    }
  ],
  "cases": [
    {
      "case_id": "request-01",
      "title": "Synthetic timing request 01",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 1765.468,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-02",
      "title": "Synthetic timing request 02",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 1905.353,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-03",
      "title": "Synthetic timing request 03",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 1900.409,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-04",
      "title": "Synthetic timing request 04",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 1799.227,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-05",
      "title": "Synthetic timing request 05",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 1815.833,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-06",
      "title": "Synthetic timing request 06",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 1758.255,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-07",
      "title": "Synthetic timing request 07",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 2116.511,
      "summary": "Transport completed; answer correctness was not graded."
    },
    {
      "case_id": "request-08",
      "title": "Synthetic timing request 08",
      "outcome": "not_run",
      "primary_outcome": "unverified",
      "failure_stage": "answer_checks_not_run",
      "duration_ms": 2233.749,
      "summary": "Transport completed; answer correctness was not graded."
    }
  ],
  "performance": {
    "version": "serving-cell-v1",
    "target_input_tokens": 4096,
    "actual_input_tokens_min": 3704,
    "actual_input_tokens_max": 3735,
    "client_concurrency_limit": 1,
    "observed_peak_inflight": 1,
    "cache_state": "unknown",
    "transport_successes": 8,
    "token_accounting": "completion_includes_reasoning",
    "stream_interval_basis": "not_measured"
  },
  "conditions": [
    {
      "label": "Endpoint usage",
      "value": "Shared; includes relay latency. AEVAmon was paused."
    },
    {
      "label": "Cache condition",
      "value": "Warm serving engine; no cache flush or verified prefix-cache state."
    },
    {
      "label": "Request limit",
      "value": "One request at a time, eight requests per input target."
    },
    {
      "label": "Answer checks",
      "value": "Not run; timing evidence only. No quality score or ranking."
    },
    {
      "label": "Timing observer SHA-256",
      "value": "2977430a164f9ef681d6cff4dd7e5826318f55d95d9ffe8740bce494dec505a9"
    },
    {
      "label": "Dataset SHA-256",
      "value": "53ddfba9e794833f0ddb92f9785dc1c012c566ef6d438df0b4dd153ca074860f"
    },
    {
      "label": "Private timing receipt SHA-256",
      "value": "bc80cb97a2aaf41f078dd3e1ea23f3091e71928c478124fded3e5fc284962da2"
    },
    {
      "label": "Observed output-token range",
      "value": "54\u2013103"
    }
  ],
  "limitations": [
    "These 16 timing requests are separate from the 192 native-agent quality attempts; they do not change either quality score.",
    "Every timing request completed. Case outcomes mark answer checks as not run, not a transport failure or a zero accuracy score.",
    "C1 only; no maximum context, concurrency or sustained decode-throughput claim.",
    "Short naturally stopping responses, shared endpoint and relay latency limit comparability.",
    "Completion tokens include reasoning. First generated and first visible content are separate measurements.",
    "Checkpoint repository and revision are recorded, but exact tensor identity has not been reverified."
  ],
  "evidence": []
}