{
  "schema_version": "trec-rag-2026-retrieval-submission-bundle-v1",
  "bundle_id": "cache-first-candidate-core-v1",
  "track": "TREC RAG 2026",
  "task": "R",
  "official_contract": {
    "checked_at": "2026-08-09",
    "guidelines_commit": "f281e88f61252662033c681df8b1ed2d0ceda97e",
    "test_data_commit": "a6255c10119a2984a874f46172d94045168ab1f3",
    "test_narratives_path": "trec-rag-data/trec-rag-2026/test-data/trec_rag_2026_queries.tsv",
    "test_narratives_sha256": "72dc2fd358d3eeda973397ccd7a8775545b19a6deaefc67709167eee6a9f8a2c",
    "expected_output_filename": "r_output_trec_rag_2026.tsv",
    "field_order": [
      "topic_id",
      "Q0",
      "docid",
      "rank",
      "score",
      "run_id"
    ],
    "maximum_runs_per_task": 10,
    "evalbase_priority_required": true,
    "submission_url": "https://ir.nist.gov/evalbase/conf/trec-2026"
  },
  "evalbase": {
    "form_observed_at": "2026-08-09",
    "run_tag_maximum_characters": 20,
    "run_tag_pattern": "[A-Za-z0-9][A-Za-z0-9_.-]{0,19}",
    "common_answers": {
      "run_type": "automatic",
      "submission_purpose": "real run",
      "agents_used_during_development": "Yes",
      "retrieval_access": "combination of both",
      "retrieval_pipeline_category": "sparse/lexical",
      "retrieval_staging": "multi-stage",
      "uses_neural_networks": "Yes",
      "uses_proprietary_models": "No",
      "uses_open_weight_models": "Yes"
    },
    "classification_note": "Candidate documents came from the organizer's Pyserini REST API and a custom local scoring and selection system reranked them; no custom index was used. BM25 is the sparse candidate retriever, while neural planning and reranking are declared separately. Runtime models are open-weight. Codex agents assisted development but did not select documents at runtime."
  },
  "source": {
    "retrieval_artifact": "outputs/facet-deepseek-b40-v3",
    "source_run_id": "facet-deepseek-b40-v3",
    "source_export_code_commit": "05fdf35d858bf52bec2843c699a4bfd85d4b8c61",
    "source_export_manifest_sha256": "cb5a81608c48a5c43692ada0c9121cd96b7160589f08f5bbaf4e20612001463a",
    "sealed_input_manifest_sha256": "f05bc8373b242b5dada7bf66a9f5d6d2526f14b8dfb2c517b6b9742bb62286ca"
  },
  "implementation": {
    "successful_publication_commit": "22b9feef18dddb6111beedb35b8c021516fab552",
    "successful_publication_tree": "c69339d23e173be62dcc024c2ed1769710827cb7",
    "successful_publication_module_sha256": "6deea9cdaf2a5232b9c483545ecfa97a8faf5e715550030f7bd23a2fda7429fe",
    "python_portability_fix_commit": "a2f5a5ffeebc6938dd946b2850876e3fc8f32903",
    "python_portability_fix_tree": "393b7c4ad56a0aefb68ca03d25b82198b81b63b2",
    "portal_projection_commit": "c8caca97060664ad323e89b89375f3a75aae9ee1",
    "portal_projection_tree": "8336049ec3e024730abf183735f4c303edbc024b",
    "current_module_sha256": "0d0b9606c7c5fe409ddeb9bc1dd9eeb383ef1a5e1866b109e82ad67d6a23b1e2",
    "portable_cache_sha256": "5de85b38a103ea0a129c7770f736ae57724b55af04b8f547622ee1ee1839057d"
  },
  "planner": {
    "backend": "trec_rag.facet_extraction.OpenRouterDeepSeekFacetBackend",
    "model": "deepseek/deepseek-v4-flash-20260423",
    "prompt_version": "subnarrative_query_generator_v1",
    "response_schema": "subnarrative_queries_v1",
    "execution": "bounded one-shot planning before retrieval",
    "weights_license": "MIT"
  },
  "candidate_selection": {
    "source_unit": "one authenticated original or facet:<subnarrative>:text retrieval lane",
    "threshold_population": "documents present in that authenticated source lane",
    "positive_mad_rule": "A_u(d) >= median_u + 2.5 * 1.4826 * MAD_u",
    "zero_mad_rule": "A_u(d) > median_u",
    "combination": "union of original-narrative and every facet-text lane admission",
    "empty_union_fallback": "one original-narrative argmax, then global best retrieval rank, then UTF-8 document ID",
    "shared_across_runs": true
  },
  "scorer": {
    "backend": "sentence-transformers-cross-encoder",
    "backend_version": "5.6.0",
    "model": "mixedbread-ai/mxbai-rerank-base-v2",
    "model_revision": "3ea9d4dffa7d12a4f366be8e275c349de9fc9865",
    "weights_license": "Apache-2.0",
    "input_policy": "topic_passage_query_text_v1",
    "score_representation": "raw_logits",
    "max_length": 1024,
    "inference_dtype": "bfloat16",
    "batch_size": 32
  },
  "chunker": {
    "backend": "trec_rag.chunking.SemanticTextChunker",
    "max_characters": 3500,
    "overlap_characters": 350,
    "trim": true
  },
  "runs": [
    {
      "name": "combo",
      "suggested_priority": 1,
      "run_id": "r26-narr-facet-v1",
      "path": "combo/r_output_trec_rag_2026.tsv",
      "evalbase_response_path": "combo/EVALBASE.md",
      "ranking": "0.5 narrative percentile + 0.5 * (0.7 best facet percentile + 0.3 second-best facet percentile)",
      "bytes": 235332,
      "line_count": 4246,
      "sha256": "29bc0c29dd51a752d49c734db456926ef94ad5202102cabd92d7fbf3e9dd15e8"
    },
    {
      "name": "breadth",
      "suggested_priority": 2,
      "run_id": "r26-facet-breadth-v1",
      "path": "breadth/r_output_trec_rag_2026.tsv",
      "evalbase_response_path": "breadth/EVALBASE.md",
      "ranking": "supported subnarrative count, strong passage count, combo score, retrieval rank, UTF-8 document ID",
      "bytes": 248070,
      "line_count": 4246,
      "sha256": "f42a794418cf692721adcd53df232ca0a3536d13d9cd137d90eb9821562e199d"
    },
    {
      "name": "narrative",
      "suggested_priority": 3,
      "run_id": "r26-narrative-v1",
      "path": "narrative/r_output_trec_rag_2026.tsv",
      "evalbase_response_path": "narrative/EVALBASE.md",
      "ranking": "narrative percentile, retrieval rank, UTF-8 document ID",
      "bytes": 231086,
      "line_count": 4246,
      "sha256": "80985a42e43333085975c27d88a1da82cc11cba6dd576c3fd506839c016f2feb"
    }
  ],
  "validation": {
    "topic_count": 119,
    "rows_per_run": 4246,
    "minimum_k": 1,
    "maximum_k": 121,
    "distinct_k_count": 61,
    "fallback_topic_count": 0,
    "same_topic_document_sets_across_runs": true,
    "dense_ranks_from_one": true,
    "nonincreasing_scores": true,
    "duplicate_topic_document_pairs": 0,
    "fresh_cache_replay": {
      "cache_hits": 167597,
      "cache_misses": 0,
      "model_batches": 0
    },
    "shared_cache_replay": {
      "cache_hits": 167597,
      "cache_misses": 0,
      "model_batches": 0
    },
    "privacy_review": {
      "contains_document_text": false,
      "contains_queries_or_narratives": false,
      "contains_credentials_or_private_urls": false
    }
  },
  "run_manifest": {
    "path": "retrieval-baseline-runs-manifest.json",
    "sha256": "afde8c8143e86e025b5481c4a7330e2b5c9547f3a9086597f9be103f06150246"
  }
}
