{
  "comparison_protocol": "same_700_group_heldout_candidate_selection",
  "dataset_groups": 3500,
  "evaluation_groups": 700,
  "validation_group_ids_sha256": "a7e51209e227ee8b68090e7826368541f209e1365112ed718c465c3bb0f11d53",
  "seed": 0,
  "candidate_oracle_success_rate": 0.4185714285714286,
  "dovla_iaf": {
    "top1_action_selection": 0.6171428571428571,
    "selected_success_rate": 0.37857142857142856,
    "mean_selected_regret": 0.059859846833028967
  },
  "smolvla_expert_only_bc": {
    "checkpoint_revision": "c83c3163b8ca9b7e67c509fffd9121e66cb96205",
    "model_sha256": "7cd549ac2351fb069c0ddb3c34ad2d09cfc92b56a15dccdfc2e41467aaca01eb",
    "training_groups": 2800,
    "training_steps": 1000,
    "top1_action_selection": 0.5228571428571429,
    "selected_success_rate": 0.3457142857142857,
    "mean_selected_regret": 0.13656934786188815
  },
  "dovla_minus_smolvla": {
    "top1_action_selection": 0.0942857142857142,
    "selected_success_rate": 0.03285714285714286,
    "mean_selected_regret": -0.07670950102885918
  },
  "scope": "Both methods select among the same measured same-state CIL candidates. This artifact does not compare online policy rollout success."
}