{
  "deliverable": "week11_comparative_analysis",
  "primary_metric": "macro_f1",
  "seed": 42,
  "version_boundary": {
    "classical": "v1.1.0",
    "gemma": "v1.0.0",
    "comparison_rule": "Within-version comparisons are controlled; cross-version scores are descriptive only."
  },
  "rows": [
    {
      "experiment": "majority",
      "dataset_version": "v1.1.0",
      "test_macro_f1": 0.06666666666666667,
      "challenge_macro_f1": 0.06666666666666667,
      "status": "complete"
    },
    {
      "experiment": "tfidf",
      "dataset_version": "v1.1.0",
      "test_macro_f1": 0.5327056707870661,
      "challenge_macro_f1": 0.5237397469334658,
      "status": "complete"
    },
    {
      "experiment": "gemma_zero_shot",
      "dataset_version": "v1.0.0",
      "test_macro_f1": 0.07955974842767295,
      "challenge_macro_f1": 0.10549943883277217,
      "status": "complete"
    },
    {
      "experiment": "gemma_few_shot",
      "dataset_version": "v1.0.0",
      "test_macro_f1": 0.13651119643810122,
      "challenge_macro_f1": 0.17411531447026668,
      "status": "complete"
    },
    {
      "experiment": "gemma_qlora",
      "dataset_version": "v1.0.0",
      "test_macro_f1": 0.0826655554183644,
      "challenge_macro_f1": 0.13089133089133087,
      "status": "complete"
    }
  ],
  "weights": {
    "toy_lora_scaffold": "mmc/deliverables/outputs/week07_lora_adapter.npz",
    "gemma_qlora_dir": "results/gemma_qlora/",
    "gemma_qlora_present": true
  },
  "failure_lab": "docs/data/failure_lab.json",
  "notes": [
    "All five published experiment rungs have result receipts.",
    "Within Gemma v1.0.0, few-shot is strongest and QLoRA does not beat it.",
    "Classical v1.1.0 and Gemma v1.0.0 scores are not a controlled same-benchmark leaderboard."
  ]
}