{
  "source_files": {
    "language_model": "experiments/depth_lm/results/result.json",
    "digits": "experiments/depth_digits/result.json"
  },
  "primary_endpoint": "Paired difference of prefix4 excess CE relative to full6 baseline",
  "cloud_spend_usd": 0,
  "experiment_budget_usd": 50,
  "final_tuning_runs": 90,
  "final_evaluation_runs": 45,
  "subset_evaluations": 2315,
  "lm_primary": {
    "n": 5,
    "mean": -0.04330816864967346,
    "std": 0.012431496225445482,
    "ci95_low": -0.05874390995662737,
    "ci95_high": -0.027872427342719556
  },
  "digits_primary": {
    "mean": -0.06770565509796142,
    "n": 5,
    "std": 0.014568466778308248,
    "ci95_low": -0.08579479580561439,
    "ci95_high": -0.04961651439030845
  },
  "notes": "Earlier partial LM tuning/pilot and complete digits boundary-search sweep are archived separately, not counted as independent final replicates."
}