{
  "status": "SUPPLEMENT_COMPLETED_AND_COLLECTED",
  "measured_date": "2026-10-27",
  "model": "zai-org/GLM-5.3",
  "revision": "ba32f8c19c8738e2c57c71798c601d477a7bf844",
  "configured_context_tokens": 1048576,
  "requested_max_output_tokens": 262144,
  "conditions": [
    {
      "concurrency": 10,
      "planned": 10,
      "statuses": {
        "complete": 10
      },
      "complete": 10,
      "automatic_pass": 10,
      "completion_s": {
        "n": 10,
        "median": 597.3838916579843,
        "p95": 1085.2958258549916,
        "min": 110.60820785991382,
        "max": 1085.2958258549916
      },
      "first_generated_s": {
        "n": 10,
        "median": 595.9265916090226,
        "p95": 1082.6799619860249,
        "min": 109.06545504496899,
        "max": 1082.6799619860249
      },
      "first_final_s": {
        "n": 10,
        "median": 595.9582690069801,
        "p95": 1082.8590682810172,
        "min": 109.0950298809912,
        "max": 1082.8590682810172
      },
      "usage_prompt_tokens": {
        "n": 10,
        "median": 785941.0,
        "p95": 785941,
        "min": 785941,
        "max": 785941
      },
      "usage_completion_tokens": {
        "n": 10,
        "median": 189.0,
        "p95": 294,
        "min": 160,
        "max": 294
      },
      "per_request_timeout_s": 5400,
      "shared_stage_deadline": null,
      "requests_sha256": "c55a9472a1d28375c6c6e185bda34e179be212d09be41f4016ff3479fdd051be",
      "grades_sha256": "8eea9f9b0a50e9d050434c873a1556be3a825bb0d787447516edc3dc26cec8ff"
    },
    {
      "concurrency": 20,
      "planned": 20,
      "statuses": {
        "complete": 20
      },
      "complete": 20,
      "automatic_pass": 19,
      "completion_s": {
        "n": 20,
        "median": 1140.6131378384307,
        "p95": 2065.046157795936,
        "min": 109.78770281502511,
        "max": 2173.3835648170207
      },
      "first_generated_s": {
        "n": 20,
        "median": 1138.5261134034954,
        "p95": 2063.568897200981,
        "min": 108.02753589698114,
        "max": 2171.6674641539576
      },
      "first_final_s": {
        "n": 20,
        "median": 1138.5752973084454,
        "p95": 2063.57032176794,
        "min": 108.19423190003727,
        "max": 2171.7681592539884
      },
      "usage_prompt_tokens": {
        "n": 20,
        "median": 785941.0,
        "p95": 785941,
        "min": 785941,
        "max": 785941
      },
      "usage_completion_tokens": {
        "n": 20,
        "median": 199.5,
        "p95": 404,
        "min": 161,
        "max": 645
      },
      "per_request_timeout_s": 5400,
      "shared_stage_deadline": null,
      "requests_sha256": "5bd79e55d4072a730536037828ce19ee17402f9a5b09445735fb4998b6402dc7",
      "grades_sha256": "b59645315c8bb342fc905c3dd9cc9f3037d0374b87f4496eab5a56b0baef8587"
    }
  ],
  "original_source_metrics_sha256": "98b50ed08e38a7999d1a9ecb762bea57d813890906c6cde041d6dd334fe5d020",
  "protocol_sha256": "5ca18bc45b16ce63207804c312e5ebd731f50728b15c535eddaa03b8763a2381",
  "input_manifest_sha256": "8a4454f00baf333c94743d0b61f56480f0f67ee43c3ff6dd063efd4f4eef4bf2",
  "change": "Removed the shared 30-minute capacity-stage cutoff; retained 5400 seconds per request.",
  "unchanged": [
    "Model revision and precision",
    "Runtime image and launch configuration",
    "Saved input bodies",
    "Frozen numerical and citation grader",
    "Context and output limits",
    "Sampling and reasoning settings"
  ],
  "original_censored_observation": {
    "concurrency10": {
      "complete": 6,
      "timeout": 4,
      "effective_timeout_approximately_s": 738
    },
    "concurrency20": "not_started"
  },
  "limitations": [
    "Supplemental run; original 6/10 censored and 20 not started retained",
    "No rerun of GPQA or operational measurements",
    "Each capacity condition one wave, not sustained SLA",
    "Sequential 10 then 20 requests, no prefix cache",
    "Same frozen GLM model, prompts, grader, context, output cap and runtime; only shared stage cutoff removed",
    "Do not equate completed requests with simultaneous KV residency",
    "These are synthetic retrieval and arithmetic prompts, not complete scientific-paper understanding"
  ],
  "automatic_failure_review": {
    "concurrency20": {
      "automatic_pass": 19,
      "planned": 20,
      "reason": "One completed response returned two JSON objects and correction prose. Initial value388, corrected final value380 matches expected380; original strict single-object grading failure retained."
    }
  },
  "gpqa_results": "Original 198-question outcomes retained, including GLM two truncated responses."
}