{
  "status": "retrieval-completed-requests-not-sent",
  "input_sha256": "1e7cf35c758aa2cfdd78c89b990e881286f1e0f563e446f23495ce130309c949",
  "runner_sha256": "1b7f61dee18425dbb5f26afe1c46a2548b80846b691990f47808b631c615c74e",
  "request_builder_sha256": "3fbd2767548ad0a74100acaa7ee80982f61d7592b274649dbc6d537dbdad9f98",
  "upstream_agent_sha256": "41b1ccff3e8676cace8b3d23377e8d8468aab88dd34ebc0b1d3efe4a21282b60",
  "runtime": {
    "python": "3.12.14",
    "langchain-community": "0.3.27",
    "langchain-core": "0.3.86",
    "rank-bm25": "0.2.2",
    "numpy": "2.5.3",
    "PyYAML": "6.0.3"
  },
  "official_full_benchmark_run": false,
  "requested_queries": 3671,
  "completed_queries": 3671,
  "completed_contexts": 146,
  "deviations": [
    "Verified upstream method slice, not full agent lifecycle",
    "BM25 index rebuilt per query; wall time is not a performance comparison",
    "Integration-selected locked CPU dependencies; upstream dependencies unpinned",
    "Prepared requests only; no model generation, judging or quality scores"
  ],
  "requests_sha256": "0a6a13059fa39af631af3d83ca41d76e4fa1ec3c33dc1ea9645f33966210dfab",
  "source_commit": "5d95846",
  "request_bytes": 608678000,
  "verified_task_counts": {
    "eventqa_65536": 500,
    "icl_banking77_5900shot_balance": 100,
    "detective_qa": 71,
    "factconsolidation_sh_6k": 100,
    "eventqa_131072": 500,
    "eventqa_full": 500,
    "longmemeval_s*": 300,
    "ruler_qa1_197K": 100,
    "ruler_qa2_421K": 100,
    "factconsolidation_mh_262k": 100,
    "factconsolidation_mh_32k": 100,
    "factconsolidation_mh_64k": 100,
    "factconsolidation_mh_6k": 100,
    "factconsolidation_sh_262k": 100,
    "factconsolidation_sh_32k": 100,
    "factconsolidation_sh_64k": 100,
    "infbench_sum_eng_shots2": 100,
    "icl_clinic150_7050shot_balance": 100,
    "icl_nlu_8296shot_balance": 100,
    "icl_trec_coarse_6600shot_balance": 100,
    "icl_trec_fine_6400shot_balance": 100,
    "recsys_redial_full": 200
  },
  "verification": "Verified complete request file hash; exact ordered case/task/context IDs; each query suffix; model, temperature and official per-task generation cap; runner and request-builder bytes against source commit.",
  "first_four_tasks_byte_identical_to_previous_run": true,
  "download_scope": "Verification receipt only; raw requests retained locally, not included in this download."
}