{
  "seed": 85,
  "query_selection": "first100 sorted by SHA256 UTF8(85:query_id), independent of labels and scores",
  "query_ids": [
    "881",
    "1125",
    "1309",
    "201",
    "1193",
    "47",
    "652",
    "456",
    "154",
    "537",
    "1224",
    "308",
    "1145",
    "9",
    "761",
    "530",
    "963",
    "1310",
    "325",
    "1306",
    "925",
    "557",
    "665",
    "30",
    "448",
    "774",
    "90",
    "813",
    "1109",
    "1120",
    "973",
    "755",
    "828",
    "1302",
    "182",
    "703",
    "1249",
    "1053",
    "371",
    "572",
    "1005",
    "1027",
    "280",
    "68",
    "858",
    "368",
    "481",
    "1138",
    "74",
    "258",
    "1050",
    "1166",
    "361",
    "900",
    "815",
    "365",
    "1406",
    "1263",
    "776",
    "780",
    "309",
    "622",
    "1034",
    "203",
    "104",
    "764",
    "283",
    "542",
    "304",
    "1276",
    "1361",
    "1341",
    "242",
    "155",
    "32",
    "658",
    "1307",
    "616",
    "469",
    "591",
    "1165",
    "95",
    "1136",
    "160",
    "717",
    "638",
    "64",
    "990",
    "139",
    "817",
    "1401",
    "978",
    "156",
    "779",
    "835",
    "687",
    "667",
    "671",
    "461",
    "454"
  ],
  "split": "BEIR train development subset; original SciFact train",
  "corpus_documents": 5183,
  "top_k": 50,
  "batch_size": 16,
  "max_length": 512,
  "threads": 4,
  "device": "cpu",
  "dtype": "float32",
  "activation": "Identity raw logits",
  "text": "(title + single space + text).strip()",
  "prompt": "none",
  "decode": "not applicable, scalar scoring",
  "primary_metric": "ndcg_cut_10",
  "tie_break": "score DESC, document ID string DESC",
  "confirmation": "300 BEIR test queries not evaluated",
  "resource_rule": "smoke first5 selected queries top10; full100x50 if extrapolated inference<900s and peakRSS<4GiB; otherwise preserve incomplete, no score-driven shrink"
}