{
  "artifactClass": "CURATED EXTRACT",
  "notice": "Selected local-preview data; not the original evidence bundle or a new measurement. Repository-relative paths do not imply public availability.",
  "result": {
    "resultId": "olmoe-0924-instruct-c60-residency39",
    "model": "allenai/OLMoE-1B-7B-0924-Instruct",
    "revision": "7f1c97f440f06ce36705e4f2b843edb5925f4498",
    "evidenceLevel": "E3",
    "status": "exploratory",
    "residentExperts": 60,
    "totalExperts": 64,
    "generatedTokens": 128,
    "metric": "post_first_token_decode_tokens_per_second",
    "value": 35.01159000794033,
    "unit": "token/s",
    "hardwareSummary": "NVIDIA GeForce RTX 5070 (frozen campaign declaration); shared Windows desktop.",
    "runDate": "2026-09-11",
    "sourceArtifact": "artifacts/residency-executors-39-20260911-01/consolidation-01/summary.json",
    "runContractSummary": "PERF-REAL-v1; 12 prompts repeated in 3 blocks; 36 measured responses of 128 tokens per configuration; greedy selection, EOS suppressed; 64-token warmup excluded. Canonical initial placement, evolving LRU through fixed prompts; concurrency 1.",
    "comparisonGroup": "olmoe0924-scalar-partial",
    "limitations": [
      "Exploratory E3 on a shared desktop; no quality evaluation or E4 promotion.",
      "Driver VRAM and process RSS are whole-worker peaks including preparation and warmup; VRAM includes desktop use.",
      "OLMoE c60 to c64 changes executor as well as residency; the transition is not a residency-only effect.",
      "Block spread is an observed range, not a confidence interval. Token parity concerns repetitions within each configuration only.",
      "TTFT and response time use internal request boundaries; isolated prefill is UNKNOWN and OLMoE HTTP observation is UNSUPPORTED."
    ],
    "modelKey": "olmoe_0924_instruct",
    "displayName": "OLMoE 0924",
    "arm": "olmoe-core",
    "executor": "PRODUCT-core scalar",
    "aggregation": "pooled-post-first-token",
    "measuredResponses": 36,
    "blocks": 3,
    "ttftSeconds": 3.0013486361111115,
    "responseSeconds": 6.628718886111112,
    "peakVramGiB": 5.670684814453125,
    "peakRssGiB": 4.956043243408203,
    "overallTokensPerSecond": 19.30991526404797,
    "h2dGiB": 37.615814208984375,
    "blockRates": [
      35.47005949236675,
      34.72178886437509,
      34.85199722045717
    ],
    "blockSpreadPercent": 2.137208358209259,
    "storagePrecision": "INT4 / block 256 / FP16 scales",
    "executionPrecision": "FP16 execution values from low-bit source; original checkpoint dtype is a separate plane."
  },
  "provenance": {
    "sourceArtifact": "artifacts/residency-executors-39-20260911-01/consolidation-01/summary.json",
    "sourceSha256": "e87b76aa9b84eb936ea6bbd0e971506becc86caa165f4e4b3edbb18ddd8c9ba7",
    "sources": [
      {
        "path": "artifacts/residency-executors-39-20260911-01/contract.json",
        "sha256": "bc0ea65fc12b1612fc41c71bc679f19679da08014933d4ab1b9bfadbd38d2455"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/consolidation-01/report.md",
        "sha256": "3be6d95d89fb312b4e6af0dbe1e64990212ea4369178f5acf2a56197102694f0"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/consolidation-01/summary.json",
        "sha256": "e87b76aa9b84eb936ea6bbd0e971506becc86caa165f4e4b3edbb18ddd8c9ba7"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/consolidation-01/per-prompt.json",
        "sha256": "5053725f887d822527d7ef782064a9b8ce1f5584a330cc757636f1078f23999e"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/consolidation-01/per-worker.json",
        "sha256": "f6a41673245761d4bd0ad85880f7fa33f536ee4f151cd93af7c5aa7e2ff5a600"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/consolidation-01/source-hashes.json",
        "sha256": "7873fe23a5e1440b33f55131327cf3a2149d12c77af549565e3a477d630b7a3d"
      },
      {
        "path": "docs/evidence/grouped-moe-real-model-task0-4-20260904-01/preflight.json",
        "sha256": "79d83ccc3721c80f624c074aa507a4f3c65b1b1a52974b8869f05af973798639"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/source/packages/backends/offload-qpacked/src/horizon_offload_qpacked/catalog.py",
        "sha256": "5a1b8e5011b9e6a8a4a2f83dabdfb1107b1cd214d592a5bb42ac8ef4c143781d"
      },
      {
        "path": "artifacts/residency-executors-39-20260911-01/01-olmoe_0924_instruct-c64-olmoe-t3/telemetry.jsonl",
        "sha256": "79235e2502fe852cbb6be677083a503dd3beb9f0bda6edbde5e9e6763da08ff9"
      }
    ],
    "selector": {
      "model_key": "olmoe_0924_instruct",
      "capacity": 60,
      "arm": "olmoe-core",
      "field": "decode_tps",
      "n": 36
    },
    "aggregation": "sum(N_i - 1) / sum(internal_decode_seconds_i); 36 measured responses, 128 tokens each, 3 blocks. Warmups excluded.",
    "latency": "Internal request boundaries; not uniformly observed HTTP latency. Isolated prefill UNKNOWN.",
    "memory": "Whole-worker peak; driver VRAM includes preparation, warmup and shared desktop use.",
    "validation": "Source hashes and exact aggregate fields checked by site build. Repeated token-ID parity is attributed to the original consolidation audit, within each configuration only."
  }
}