{
  "network_reference": {
    "source": "User-provided same-Runpod network measurement screenshot",
    "api_call_pattern": "240 consecutive calls across three rounds",
    "google_icmp_ping": {
      "samples": 5,
      "mean_ms": 31.97
    },
    "typesafe_tcp_443": {
      "samples": 10,
      "mean_ms": 181.55
    },
    "typesafe_tls_new_connections": {
      "samples": 10,
      "mean_ms": 184.57
    },
    "typesafe_icmp_response": false,
    "caveat": "Separate new-connection probes; no per-request subtraction or model-only latency inference."
  },
  "typesafe_three_rounds": {
    "source": "User-provided TypeSafe three-round screenshot, 2026-09-21",
    "timing": "Official public API full HTTP response; pooled percentiles across all 240 requests",
    "raw_requests_independently_verified": false,
    "rounds": [
      {
        "round": 1,
        "mean_ms": 298.83,
        "correct": 68,
        "questions": 80
      },
      {
        "round": 2,
        "mean_ms": 303.55,
        "correct": 66,
        "questions": 80
      },
      {
        "round": 3,
        "mean_ms": 305.57,
        "correct": 65,
        "questions": 80
      }
    ],
    "overall_mean_ms": 302.65,
    "overall_correct": 199,
    "overall_requests": 240,
    "overall_accuracy_percent": 82.92,
    "p50_ms": 280.8,
    "p95_ms": 376.56,
    "p99_ms": 830.45,
    "aggregation": "Percentiles recomputed on pooled 240 records, not averages of per-round percentiles",
    "http_200_responses": 240,
    "note": "Historical single-run accuracy charts retain their original run; this is a separate repeated evaluation."
  },
  "provided_snapshot": {
    "source": "User-provided screenshot",
    "raw_requests_independently_verified": false,
    "concurrency": null,
    "effort": "high",
    "hardware": "1x RTX PRO 6000 Blackwell 96GB",
    "timing": "vLLM HTTP full response",
    "reported_model": "APUS-OpenJev 9B-5949",
    "p50_ms": 25.58,
    "p95_ms": 222.36,
    "p99_ms": 275.58,
    "successful": 333,
    "requests": 333,
    "error_rate": 0
  },
  "verified_reference": {
    "source": "feat-jev-scale-9b-35b/docs/feat-jev-scale-9b-35b/evidence/vllm-merged-http-20260921/local-verification.json",
    "aggregation": "Median of three per-round percentiles",
    "concurrency": 1,
    "rounds": 3,
    "requests_per_round": 1040,
    "p50_ms": 25.34,
    "p95_ms": 222.83,
    "p99_ms": 276.78,
    "correct": 67,
    "quality_records": 80
  },
  "four_b_http": {
    "source": "User updated 4B P50 to18ms; P95/P99 ratios remain45%/56% of9B high",
    "evidence_kind": "P50: latest user-reported value; P95/P99: calculated from reported ratios; not a direct request-log replay",
    "reference_model": "9B-5949 high",
    "ratios": {
      "p50": null,
      "p95": 0.45,
      "p99": 0.56
    },
    "p50_ms": 18.0,
    "p95_ms": 100.06200000000001,
    "p99_ms": 154.3248,
    "display_ms": {
      "p50_ms": 18.0,
      "p95_ms": 100.06,
      "p99_ms": 154.32
    }
  },
  "historical_rows": [
    {
      "name": "Jev official API",
      "scope": "TypeSafe HTTP / 240 pooled requests",
      "p50_ms": 280.8,
      "p95_ms": 376.56
    },
    {
      "name": "Laya typed / full input",
      "scope": "Local system_one / three-run median",
      "p50_ms": 8.68,
      "p95_ms": 37.66
    }
  ],
  "source_search": "No exact 333-request run found in local scale evidence, combined release, or archived loopback/diagnostic-load reports."
}