{
 "published": "2026-10-11",
 "provenance": "Generated by /assets/chunking-benchmark.py (seed 7, deterministic). Runs reproduce the numbers in the Oct 2026 data report exactly.",
 "runs": [
  {
   "config": {
    "articles": 20,
    "questions": 800,
    "chunk_tokens": 200,
    "overlap_tokens": 0,
    "model": "all-MiniLM-L6-v2",
    "dataset": "SQuAD v1.1 dev",
    "seed": 7
   },
   "results": {
    "fixed": {
     "recall@1": 0.6112,
     "recall@3": 0.8013,
     "recall@5": 0.8712,
     "chunks": 818,
     "avg_tokens": 197.3,
     "duplication_pct": 0.0,
     "embedded_tokens": 161426
    },
    "sentence": {
     "recall@1": 0.6687,
     "recall@3": 0.8562,
     "recall@5": 0.905,
     "chunks": 925,
     "avg_tokens": 174.5,
     "duplication_pct": 0.0,
     "embedded_tokens": 161426
    },
    "recursive": {
     "recall@1": 0.7075,
     "recall@3": 0.8812,
     "recall@5": 0.9137,
     "chunks": 1010,
     "avg_tokens": 159.8,
     "duplication_pct": 0.0,
     "embedded_tokens": 161426
    }
   }
  },
  {
   "config": {
    "articles": 20,
    "questions": 800,
    "chunk_tokens": 200,
    "overlap_tokens": 30,
    "model": "all-MiniLM-L6-v2",
    "dataset": "SQuAD v1.1 dev",
    "seed": 7
   },
   "results": {
    "fixed": {
     "recall@1": 0.6575,
     "recall@3": 0.85,
     "recall@5": 0.9137,
     "chunks": 959,
     "avg_tokens": 197.7,
     "duplication_pct": 17.5,
     "embedded_tokens": 189596
    },
    "sentence": {
     "recall@1": 0.7025,
     "recall@3": 0.8738,
     "recall@5": 0.92,
     "chunks": 1320,
     "avg_tokens": 176.9,
     "duplication_pct": 44.6,
     "embedded_tokens": 233473
    },
    "recursive": {
     "recall@1": 0.7325,
     "recall@3": 0.8838,
     "recall@5": 0.9263,
     "chunks": 1379,
     "avg_tokens": 208.8,
     "duplication_pct": 78.3,
     "embedded_tokens": 287882
    }
   }
  }
 ]
}