{
  "schema_version": 1,
  "run_id": "fullglm-frozen-32k-c1-a8227da6b542",
  "started_at": "2026-09-21T22:43:57.637799+00:00",
  "completed_at": "2026-09-21T22:47:30.761618+00:00",
  "status": "completed",
  "evidence_mode": "actual_inference",
  "role": "serving",
  "boundary": "role_completion",
  "phase": "performance",
  "suite_version": "static-performance-evidence-1",
  "suite_sha256": "54698fe094f4d8a440d7890e90a89f95d222a1f8e4f9e6feaebcbb4d5bd81aeb",
  "evaluator_sha256": "2977430a164f9ef681d6cff4dd7e5826318f55d95d9ffe8740bce494dec505a9",
  "comparison_key": null,
  "deployment": {
    "id": "glm53-uncensored-nvfp4-tp16-high",
    "model": "GLM-5.3 Uncensored",
    "source_url": "https://huggingface.co/dealignai/GLM-5.3-UNCENSORED-NVFP4",
    "revision": "409ddf484d16cb7b13d390e8b30942890164d5af",
    "variant": "uncensored",
    "quantization": "NVFP4 weights; FP8 MLA KV cache",
    "hardware": "NVIDIA DGX Spark GB10, 128 GB per node",
    "node_count": 16,
    "runtime": "vLLM with upstream structured-output repair",
    "runtime_version": "v19 + upstream PR44993",
    "context_limit": 1048576,
    "concurrency": 16,
    "profile": "Timing-only C1; eight naturally stopping requests per input target; endpoint usage unknown",
    "weights_verified": true
  },
  "counts": {
    "declared": 8,
    "primary_passes": 0,
    "outcomes": {
      "not_run": 8
    }
  },
  "metrics": [
    {
      "id": "completion_throughput_tps",
      "label": "End-to-end completion throughput",
      "value": 4.719937510581762,
      "unit": "tok/s",
      "samples": 8,
      "definition": "Observed completion tokens, including reasoning, divided by the full measured batch duration. Includes waiting, prefill, generation and private LAN latency; not decode TPS."
    },
    {
      "id": "first_generated_median_ms",
      "label": "First generated output, median",
      "value": 21399.9375,
      "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": 23088.7515,
      "unit": "ms",
      "samples": 8,
      "definition": "First nonempty content payload among complete answers independently checked for reasoning tags throughout their captured text. Correctness and usefulness were not graded."
    },
    {
      "id": "completion_median_ms",
      "label": "Request completion, median",
      "value": 24389.394500000002,
      "unit": "ms",
      "samples": 8,
      "definition": "Full client-observed request duration, including private LAN latency and any endpoint queueing."
    },
    {
      "id": "batch_duration_seconds",
      "label": "Measured batch duration",
      "value": 196.40090529201552,
      "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": 23626.454,
      "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": 24424.503,
      "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": 23612.672,
      "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": 24354.286,
      "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": 23560.852,
      "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": 24772.415,
      "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": 27467.533,
      "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": 24575.846,
      "summary": "Transport completed; answer correctness was not graded."
    }
  ],
  "performance": {
    "version": "serving-cell-v1",
    "target_input_tokens": 32768,
    "actual_input_tokens_min": 28884,
    "actual_input_tokens_max": 28915,
    "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": "Unknown; private LAN endpoint. AEVAmon was paused, but exclusive endpoint use was not verified."
    },
    {
      "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; 300-second request deadline."
    },
    {
      "label": "Answer checks",
      "value": "Correctness not graded; timing evidence only. No quality score or ranking."
    },
    {
      "label": "Timing observer",
      "value": "sglang-content-observation-4; complete captured answers independently checked for late reasoning tags."
    },
    {
      "label": "First-final-content coverage",
      "value": "8/8 complete answers without a <think> tag; 0 tagged, 0 truncated, 0 length mismatches."
    },
    {
      "label": "Timing observer SHA-256",
      "value": "2977430a164f9ef681d6cff4dd7e5826318f55d95d9ffe8740bce494dec505a9"
    },
    {
      "label": "Dataset SHA-256",
      "value": "54698fe094f4d8a440d7890e90a89f95d222a1f8e4f9e6feaebcbb4d5bd81aeb"
    },
    {
      "label": "Private timing receipt SHA-256",
      "value": "a8227da6b542e5abb26274cc3a0c41a33f55f4213cd432334a776922fe40a401"
    },
    {
      "label": "Observed output-token range",
      "value": "74\u2013220"
    },
    {
      "label": "Checkpoint verification",
      "value": "282 checkpoint shard hashes verified September 19 against the recorded upstream revision; matching source and staged-copy receipts retained."
    }
  ],
  "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 and unverified competing endpoint traffic limit comparability.",
    "Completion tokens include reasoning. First generated and first visible content are separate measurements.",
    "Frozen observer version 4 checks only the first 256 content characters for reasoning tags; first-final-content metrics additionally use a complete saved-answer audit. Frozen receipts were not rewritten.",
    "Checkpoint verification refers to the matching September 19 byte-hash and staging receipts; live tensors were not rehashed during this timing audit.",
    "This frozen campaign does not evaluate subsequent application repairs."
  ],
  "evidence": []
}