{
  "musique": {
    "dataset": "musique",
    "n_examples": 150,
    "chunks_per_query_min": 20,
    "chunks_per_query_max": 20,
    "chunks_per_query_target": 20,
    "chunks_per_query_uniform": true,
    "gold_substring_hits_source": 103,
    "gold_substring_hits_extended": 103,
    "gold_preservation_ratio_vs_source": 1.0,
    "structural_chunk_preservation": 1.0,
    "avg_chunk_chars": 2390.1,
    "avg_chunk_tokens_estimated": 597.5,
    "cacheblend_suitability": {
      "checks": {
        "chunks_per_query_ge_10": true,
        "gold_preservation_vs_source_ge_0.99": true,
        "structural_chunk_preservation_eq_1.0": true,
        "chunk_size_within_paper_range_300_800_tokens": true
      },
      "verdict_pass": true,
      "notes": [
        "multi-hop QA inherited from source (MuSiQue / 2WikiMQA)",
        "distractors sampled from other queries -> realistic cross-attention noise",
        "F1 (short-answer) metric directly applicable",
        "absolute substring hit < 100% reflects MuSiQue/2WikiMQA paraphrased answers, not our augmentation"
      ]
    },
    "source_file": "musique_s.json",
    "extra_per_query": 10,
    "seed": 42,
    "output_file": "/home/user/experiment-reproducer/workspace/code/data/musique_extended.json"
  },
  "wikimqa": {
    "dataset": "wikimqa",
    "n_examples": 200,
    "chunks_per_query_min": 20,
    "chunks_per_query_max": 20,
    "chunks_per_query_target": 20,
    "chunks_per_query_uniform": true,
    "gold_substring_hits_source": 105,
    "gold_substring_hits_extended": 109,
    "gold_preservation_ratio_vs_source": 1.0381,
    "structural_chunk_preservation": 1.0,
    "avg_chunk_chars": 2242.2,
    "avg_chunk_tokens_estimated": 560.5,
    "cacheblend_suitability": {
      "checks": {
        "chunks_per_query_ge_10": true,
        "gold_preservation_vs_source_ge_0.99": true,
        "structural_chunk_preservation_eq_1.0": true,
        "chunk_size_within_paper_range_300_800_tokens": true
      },
      "verdict_pass": true,
      "notes": [
        "multi-hop QA inherited from source (MuSiQue / 2WikiMQA)",
        "distractors sampled from other queries -> realistic cross-attention noise",
        "F1 (short-answer) metric directly applicable",
        "absolute substring hit < 100% reflects MuSiQue/2WikiMQA paraphrased answers, not our augmentation"
      ]
    },
    "source_file": "wikimqa_s.json",
    "extra_per_query": 10,
    "seed": 42,
    "output_file": "/home/user/experiment-reproducer/workspace/code/data/wikimqa_extended.json"
  }
}