{
  "phase": "Phase 2 - Cross-Domain Validation with BEIR (M-0002)",
  "coreArchitecture": "ARCH-0.5 (BM25 + BGE hybrid)",
  "datasets": [
    {
      "id": "scifact",
      "tier": 1,
      "bm25": 0.6906,
      "published": 0.665,
      "live": true,
      "searchLive": true,
      "note": "live BM25 search"
    },
    {
      "id": "nfcorpus",
      "tier": 1,
      "bm25": 0.3273,
      "published": 0.325,
      "live": true,
      "searchLive": true,
      "note": "live BM25 search"
    },
    {
      "id": "fiqa",
      "tier": 1,
      "bm25": 0.2536,
      "published": 0.236,
      "live": true,
      "searchLive": true,
      "note": "live BM25 search"
    },
    {
      "id": "arguana",
      "tier": 1,
      "bm25": 0.4739,
      "published": 0.472,
      "live": true,
      "searchLive": true,
      "note": "live BM25 search"
    },
    {
      "id": "scidocs",
      "tier": 1,
      "bm25": 0.1647,
      "published": 0.158,
      "live": true,
      "searchLive": true,
      "note": "live BM25 search"
    },
    {
      "id": "trec-covid",
      "tier": 2,
      "bm25": 0.6877,
      "published": 0.656,
      "live": false,
      "searchLive": false,
      "note": "tier 2: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "webis-touche2020",
      "tier": 2,
      "bm25": 0.3472,
      "published": 0.367,
      "live": false,
      "searchLive": false,
      "note": "tier 2: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "cqadupstack",
      "tier": 2,
      "bm25": 0.3253,
      "published": 0.299,
      "live": false,
      "searchLive": false,
      "note": "tier 2: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "quora",
      "tier": 2,
      "bm25": 0.8084,
      "published": 0.789,
      "live": false,
      "searchLive": false,
      "note": "tier 2: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "nq",
      "tier": 3,
      "bm25": 0.3263,
      "published": 0.329,
      "live": false,
      "searchLive": false,
      "note": "tier 3: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "hotpotqa",
      "tier": 3,
      "bm25": 0.6022,
      "published": 0.603,
      "live": false,
      "searchLive": false,
      "note": "tier 3: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "dbpedia-entity",
      "tier": 3,
      "bm25": 0.3205,
      "published": 0.313,
      "live": false,
      "searchLive": false,
      "note": "tier 3: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "msmarco",
      "tier": 3,
      "bm25": 0.2278,
      "published": 0.228,
      "live": false,
      "searchLive": false,
      "note": "tier 3: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "climate-fever",
      "tier": 3,
      "bm25": 0.1862,
      "published": 0.213,
      "live": false,
      "searchLive": false,
      "note": "tier 3: offline-evaluated (free-tier infra limit)"
    },
    {
      "id": "fever",
      "tier": 3,
      "bm25": 0.6493,
      "published": 0.753,
      "live": false,
      "searchLive": false,
      "note": "tier 3: offline-evaluated (free-tier infra limit)"
    }
  ],
  "liveSearchDatasets": [
    "scifact",
    "nfcorpus",
    "fiqa",
    "arguana",
    "scidocs"
  ],
  "matrix": {
    "aggregateBm25NdcgAt10": 0.4261,
    "publishedBm25Average": 0.43,
    "supersetDatasets": "Cranfield + BEIR Tier 1",
    "techniques": [
      {
        "name": "bge-hybrid",
        "scope": "universal",
        "role": "ARCH-0.5 core",
        "evidence": "improves 6/6, +9.5% to +64.0%, zero regressions"
      },
      {
        "name": "learning-to-rank",
        "scope": "portfolio",
        "role": "activatable; kept for Phase 4 behavioral",
        "evidence": "beats BM25 everywhere but plain hybrid on only 2/4 (Occam)"
      },
      {
        "name": "refined-prf-rerank",
        "scope": "domain-conditional",
        "role": "portfolio",
        "evidence": "helps 3/6; hurts FiQA/ArguAna"
      },
      {
        "name": "coverage-rerank",
        "scope": "domain-conditional",
        "role": "portfolio",
        "evidence": "helps 3/6, marginal"
      },
      {
        "name": "doc2query",
        "scope": "portfolio-candidate",
        "role": "compute-deferred; Phase 3 fit",
        "evidence": "NFCorpus +0.9%; broad eval GPU-gated"
      }
    ],
    "fever_family_note": "FEVER (-13.8%) and Climate-FEVER (-12.6%) run ~13% below published (shared corpus, localized BM25 config difference); documented, not a harness bug (ADL-0005)."
  }
}