{
  "artifact_type": "empirical_public_advanced_disease_analysis_provenance",
  "clinical_claim": false,
  "dataset_boundary": "Advanced prostate and metastatic breast cancer versus healthy donors; not MCED, early detection, or screening validation.",
  "software": {
    "python": "3.14.6",
    "numpy": "2.5.1",
    "scikit_learn": "1.9.0",
    "algorithm_module": "mced_lpwgs/finaleme_cohort.py",
    "algorithm_module_sha256": "b9675a357edde1c91e0187b39b8fcd06c04879183326452c4f1d2bf879862a9f",
    "entrypoint": "scripts/analyze_finaleme.py",
    "entrypoint_sha256": "d18e58787acb9a264fb58b20d8860c142a4a407a499ae2a6babc127a4608a188"
  },
  "inputs": {
    "feature_summary": "real_data/derived/finaleme/feature_summary.tsv",
    "feature_summary_sha256": "d129e87d0f10355840d5fa10f749ca058e4cd3defe9cdbd994942c1265b59ddc",
    "extraction_provenance": "real_data/derived/finaleme/extraction_provenance.json",
    "extraction_provenance_sha256": "80b320a3fcc6a4de99b7e3544595f3e86b7097b9f66b14e8b42c0d646287abaf"
  },
  "analysis_contract": {
    "features": [
      "global_short_long_log_ratio",
      "coverage_logratio_mad",
      "median_fragment_length"
    ],
    "primary_unit": "patient_id after componentwise-median sample aggregation",
    "primary_validation": "leave-one-donor-out; scaling fitted inside each fold",
    "model": "fixed-C ridge logistic with fold-fitted scaling; no tuning",
    "bootstrap_repeats": 5000,
    "negative_controls": [
      "batch incidence only",
      "usable coordinate-record depth only"
    ],
    "stringent_batch_sensitivities": [
      "retain only batches containing both classes",
      "train on one mixed batch and test the other with donor quarantine"
    ]
  },
  "validation": {
    "sample_records": 77,
    "patient_id_groups": 42,
    "all_grouped_cv_folds_have_zero_donor_overlap": true,
    "mixed_class_batches": [
      "8",
      "10"
    ],
    "posthoc_advanced_disease_batch_confounded_donor_lodo_auc": 0.9694444444444443,
    "posthoc_advanced_disease_batch_only_donor_lodo_auc": 0.9638888888888888,
    "confounding_warning": "Batch-only discrimination is nearly as high as the biological-feature model; performance must not be interpreted as cancer generalization."
  },
  "outputs": {
    "real_data/derived/finaleme/benchmark.json": {
      "bytes": 84484,
      "sha256": "7e530022f9689cc3731481fb04083c699bb0e97f9288fc18760c292bb4c69ff9"
    },
    "real_data/derived/finaleme/donor_predictions.tsv": {
      "bytes": 11035,
      "sha256": "66be2031c642ba24581b7fc90d6b4e9d3809059639ab851ef452b55015a98440"
    },
    "real_data/derived/finaleme/sample_predictions.tsv": {
      "bytes": 13007,
      "sha256": "8bdbc5285f77736aaaadf74836572e5c012fd4f523dffea29d2aca153d9ccad9"
    },
    "real_data/derived/finaleme/batch_audit.tsv": {
      "bytes": 615,
      "sha256": "db6c1bc0e74bbf53fde7fec5ddb4d99b27fc7c9beb256b6f025a3bb14eddda2c"
    }
  }
}