{
  "scope": "Recorded 20-question run: sample-1000 plus its recorded benchmark-additions extension.",
  "unique_document_ids": 1045,
  "source_records": 1052,
  "chunks_per_database": 48053,
  "unique_chunk_ids": 48053,
  "chunks_per_database_match_recorded_preflight": true,
  "independent_sql_and_arrow_match": true,
  "parts": [
    {
      "workspace": "/home/kit/Documents/sfd-2025/data/pipeline/sample-1000",
      "filings": 1000,
      "chunks": 43105
    },
    {
      "workspace": "/home/kit/Documents/sfd-2025/data/pipeline/benchmark-additions/run",
      "filings": 52,
      "chunks": 4948
    }
  ],
  "sql": "SELECT count(*) chunks, count(DISTINCT id) chunk_ids, count(DISTINCT accession) document_ids, count(DISTINCT (source_file,source_row)) source_records FROM read_parquet(?)",
  "source_parquet_access": {
    "files": 6,
    "compressed_bytes": 6914044276,
    "note": "Agentic Search could additionally use these original source Parquet files; BM25/Vector/Hybrid searched the index."
  },
  "definitions": {
    "document_ids": "DISTINCT accession in indexed chunks; parsed filing accession, not company CIK.",
    "source_records": "DISTINCT (source_file,source_row), so repeated document IDs may occur.",
    "chunks": "Retrieval passages stored in each database, not documents."
  },
  "checked_utc": "2026-10-15T11:15:27.787217+00:00"
}