{
  "source": "local-artifact:composition-v1",
  "method": "Full-model simulation distillation, uncertainty-weighted action loss",
  "objective": "imitation",
  "reference_method": "Exact optimal finite-shoe single-player unsplit decisions within 50000 cache misses; explicit basic-continuation Monte Carlo fallback otherwise. Per-row method/coverage recorded.",
  "teacher_counts": {
    "train": {
      "exact": 7427,
      "monte_carlo": 58109
    },
    "selection": {
      "exact": 507,
      "monte_carlo": 3589
    },
    "calibration": {
      "exact": 238,
      "monte_carlo": 1810
    },
    "test": {
      "exact": 1006,
      "monte_carlo": 7186
    }
  },
  "full_model": true,
  "epochs": 3,
  "batch_size": 8,
  "learning_rate": 5e-06,
  "seed": 40260924,
  "device": "cuda:0",
  "history": [
    {
      "epoch": 1,
      "complete_epoch": true,
      "updates": 24576,
      "mean_loss": 0.6137711106589644,
      "selection": {
        "action_brier": 0.09495679542787608,
        "teacher_agreement": 0.9443359375,
        "teacher_ev_regret": 0.0015316948437898948,
        "hit_bust_brier": 0.0006087590208322614,
        "dealer_brier": 0.0015001254346832336,
        "teacher_action_ece": 0.03929415957062843,
        "resolved_agreement": 0.9790444258172674,
        "unresolved_agreement": 0.7040618955512572,
        "resolved_states": 3579,
        "unresolved_states": 517,
        "by_players": {
          "7": {
            "states": 404,
            "teacher_agreement": 0.948019801980198,
            "teacher_ev_regret": 0.0014164990717821782
          },
          "3": {
            "states": 570,
            "teacher_agreement": 0.9333333333333333,
            "teacher_ev_regret": 0.00194091796875
          },
          "6": {
            "states": 474,
            "teacher_agreement": 0.9324894514767933,
            "teacher_ev_regret": 0.0019067691851265822
          },
          "4": {
            "states": 588,
            "teacher_agreement": 0.95578231292517,
            "teacher_ev_regret": 0.0013251272188562924
          },
          "2": {
            "states": 751,
            "teacher_agreement": 0.9507323568575233,
            "teacher_ev_regret": 0.001689966763065912
          },
          "1": {
            "states": 759,
            "teacher_agreement": 0.9420289855072463,
            "teacher_ev_regret": 0.0013262321028997494
          },
          "5": {
            "states": 550,
            "teacher_agreement": 0.9454545454545454,
            "teacher_ev_regret": 0.0011572265625
          }
        },
        "by_stratum": {
          "general": {
            "states": 2048,
            "teacher_agreement": 0.95703125,
            "teacher_ev_regret": 0.000969717147655146
          },
          "depleted": {
            "states": 2048,
            "teacher_agreement": 0.931640625,
            "teacher_ev_regret": 0.002093672539924644
          }
        },
        "by_teacher": {
          "monte_carlo": {
            "states": 3589,
            "teacher_agreement": 0.9456673168013374,
            "teacher_ev_regret": 0.0015454189994601561
          },
          "exact": {
            "states": 507,
            "teacher_agreement": 0.9349112426035503,
            "teacher_ev_regret": 0.001434542980475167
          }
        }
      },
      "selection_eligible": false
    },
    {
      "epoch": 2,
      "complete_epoch": true,
      "updates": 49152,
      "mean_loss": 0.5968220626006241,
      "selection": {
        "action_brier": 0.09040151564200882,
        "teacher_agreement": 0.947265625,
        "teacher_ev_regret": 0.001314263350765287,
        "hit_bust_brier": 0.0003857798560271163,
        "dealer_brier": 0.0011841758487930866,
        "teacher_action_ece": 0.04156270071993659,
        "resolved_agreement": 0.9837943559653535,
        "unresolved_agreement": 0.6943907156673114,
        "resolved_states": 3579,
        "unresolved_states": 517,
        "by_players": {
          "7": {
            "states": 404,
            "teacher_agreement": 0.9554455445544554,
            "teacher_ev_regret": 0.0008560029586943069
          },
          "3": {
            "states": 570,
            "teacher_agreement": 0.9385964912280702,
            "teacher_ev_regret": 0.0019876045093201755
          },
          "6": {
            "states": 474,
            "teacher_agreement": 0.9324894514767933,
            "teacher_ev_regret": 0.0017123322949630802
          },
          "4": {
            "states": 588,
            "teacher_agreement": 0.9523809523809523,
            "teacher_ev_regret": 0.0008906150350765306
          },
          "2": {
            "states": 751,
            "teacher_agreement": 0.9494007989347537,
            "teacher_ev_regret": 0.0015682215379494008
          },
          "1": {
            "states": 759,
            "teacher_agreement": 0.9472990777338604,
            "teacher_ev_regret": 0.0011364859050357256
          },
          "5": {
            "states": 550,
            "teacher_agreement": 0.9545454545454546,
            "teacher_ev_regret": 0.0009614701704545454
          }
        },
        "by_stratum": {
          "general": {
            "states": 2048,
            "teacher_agreement": 0.95849609375,
            "teacher_ev_regret": 0.0008593745229399758
          },
          "depleted": {
            "states": 2048,
            "teacher_agreement": 0.93603515625,
            "teacher_ev_regret": 0.0017691521785905984
          }
        },
        "by_teacher": {
          "monte_carlo": {
            "states": 3589,
            "teacher_agreement": 0.9467818333797715,
            "teacher_ev_regret": 0.0013517526970430482
          },
          "exact": {
            "states": 507,
            "teacher_agreement": 0.9506903353057199,
            "teacher_ev_regret": 0.0010488801874696562
          }
        }
      },
      "selection_eligible": true
    },
    {
      "epoch": 3,
      "complete_epoch": true,
      "updates": 73728,
      "mean_loss": 0.5832477171116298,
      "selection": {
        "action_brier": 0.07964431880238707,
        "teacher_agreement": 0.95458984375,
        "teacher_ev_regret": 0.0010111489430474339,
        "hit_bust_brier": 0.00029111769938248327,
        "dealer_brier": 0.0009628682438364339,
        "teacher_action_ece": 0.03631929756629491,
        "resolved_agreement": 0.9879854708019,
        "unresolved_agreement": 0.723404255319149,
        "resolved_states": 3579,
        "unresolved_states": 517,
        "by_players": {
          "7": {
            "states": 404,
            "teacher_agreement": 0.9653465346534653,
            "teacher_ev_regret": 0.0005245397586633664
          },
          "3": {
            "states": 570,
            "teacher_agreement": 0.9421052631578948,
            "teacher_ev_regret": 0.0016378837719298247
          },
          "6": {
            "states": 474,
            "teacher_agreement": 0.9345991561181435,
            "teacher_ev_regret": 0.0016482067510548524
          },
          "4": {
            "states": 588,
            "teacher_agreement": 0.9625850340136054,
            "teacher_ev_regret": 0.0008196149553571429
          },
          "2": {
            "states": 751,
            "teacher_agreement": 0.96271637816245,
            "teacher_ev_regret": 0.0008223085365346206
          },
          "1": {
            "states": 759,
            "teacher_agreement": 0.9525691699604744,
            "teacher_ev_regret": 0.0009504321275161915
          },
          "5": {
            "states": 550,
            "teacher_agreement": 0.96,
            "teacher_ev_regret": 0.0007164417613636364
          }
        },
        "by_stratum": {
          "general": {
            "states": 2048,
            "teacher_agreement": 0.9638671875,
            "teacher_ev_regret": 0.000675438124557217
          },
          "depleted": {
            "states": 2048,
            "teacher_agreement": 0.9453125,
            "teacher_ev_regret": 0.0013468597615376508
          }
        },
        "by_teacher": {
          "monte_carlo": {
            "states": 3589,
            "teacher_agreement": 0.9543048202842017,
            "teacher_ev_regret": 0.0010268667608491224
          },
          "exact": {
            "states": 507,
            "teacher_agreement": 0.9566074950690335,
            "teacher_ev_regret": 0.0008998841539147721
          }
        }
      },
      "selection_eligible": true
    }
  ],
  "stopped_for_budget": false,
  "updates": 73728,
  "baseline": {
    "states": 8192,
    "teacher_agreement": 0.9495849609375,
    "teacher_ev_regret": 0.0014074805906066532,
    "hit_bust_brier": 0.0007162610006696003,
    "dealer_brier": 0.0015329088272374996,
    "action_brier": 0.07573528661987305,
    "teacher_action_ece": 0.013433056640624967
  },
  "baseline_selection": {
    "action_brier": 0.0818870382032546,
    "teacher_agreement": 0.947998046875,
    "teacher_ev_regret": 0.0015162492694547767,
    "hit_bust_brier": 0.0007207577826748892,
    "dealer_brier": 0.0015160600000445174,
    "teacher_action_ece": 0.015856055714206178,
    "resolved_agreement": 0.9793238334730372,
    "unresolved_agreement": 0.7311411992263056,
    "resolved_states": 3579,
    "unresolved_states": 517,
    "by_players": {
      "7": {
        "states": 404,
        "teacher_agreement": 0.9579207920792079,
        "teacher_ev_regret": 0.001230069906404703
      },
      "3": {
        "states": 570,
        "teacher_agreement": 0.9280701754385965,
        "teacher_ev_regret": 0.002407997532894737
      },
      "6": {
        "states": 474,
        "teacher_agreement": 0.9409282700421941,
        "teacher_ev_regret": 0.0015104269679588607
      },
      "4": {
        "states": 588,
        "teacher_agreement": 0.9523809523809523,
        "teacher_ev_regret": 0.0013508699378188774
      },
      "2": {
        "states": 751,
        "teacher_agreement": 0.952063914780293,
        "teacher_ev_regret": 0.0018087862017310253
      },
      "1": {
        "states": 759,
        "teacher_agreement": 0.9486166007905138,
        "teacher_ev_regret": 0.0012070137206017996
      },
      "5": {
        "states": 550,
        "teacher_agreement": 0.9563636363636364,
        "teacher_ev_regret": 0.0010114080255681818
      }
    },
    "by_stratum": {
      "general": {
        "states": 2048,
        "teacher_agreement": 0.958984375,
        "teacher_ev_regret": 0.0009934053901861305
      },
      "depleted": {
        "states": 2048,
        "teacher_agreement": 0.93701171875,
        "teacher_ev_regret": 0.0020390931487234234
      }
    },
    "by_teacher": {
      "monte_carlo": {
        "states": 3589,
        "teacher_agreement": 0.9492894956812482,
        "teacher_ev_regret": 0.0015255898013025913
      },
      "exact": {
        "states": 507,
        "teacher_agreement": 0.9388560157790927,
        "teacher_ev_regret": 0.0014501286209305047
      }
    }
  },
  "test_deferred": false,
  "general_margin": 0.0005,
  "test": {
    "states": 8192,
    "teacher_agreement": 0.958984375,
    "teacher_ev_regret": 0.0009072807931029825,
    "hit_bust_brier": 0.00028900535657547276,
    "dealer_brier": 0.0009029593136866748,
    "action_brier": 0.061715743963623046,
    "teacher_action_ece": 0.012601574707031124
  },
  "temperature_by_options": {
    "choice:3-5": 2.1134890398366477,
    "noul:2": 1.0,
    "choice:6-10": 1.059253725177289,
    "choice:2": 2.23872113856834
  },
  "states": {
    "train": 65536,
    "selection": 4096,
    "calibration": 2048,
    "test": 8192,
    "validation": 2048
  },
  "dataset_hash": "0ecbe37e7489a78a7570dfa31fd67b0f61535a9156926e1243b4320c8c884678",
  "selected": {
    "epoch": 3,
    "complete_epoch": true,
    "updates": 73728,
    "mean_loss": 0.5832477171116298,
    "selection": {
      "action_brier": 0.07964431880238707,
      "teacher_agreement": 0.95458984375,
      "teacher_ev_regret": 0.0010111489430474339,
      "hit_bust_brier": 0.00029111769938248327,
      "dealer_brier": 0.0009628682438364339,
      "teacher_action_ece": 0.03631929756629491,
      "resolved_agreement": 0.9879854708019,
      "unresolved_agreement": 0.723404255319149,
      "resolved_states": 3579,
      "unresolved_states": 517,
      "by_players": {
        "7": {
          "states": 404,
          "teacher_agreement": 0.9653465346534653,
          "teacher_ev_regret": 0.0005245397586633664
        },
        "3": {
          "states": 570,
          "teacher_agreement": 0.9421052631578948,
          "teacher_ev_regret": 0.0016378837719298247
        },
        "6": {
          "states": 474,
          "teacher_agreement": 0.9345991561181435,
          "teacher_ev_regret": 0.0016482067510548524
        },
        "4": {
          "states": 588,
          "teacher_agreement": 0.9625850340136054,
          "teacher_ev_regret": 0.0008196149553571429
        },
        "2": {
          "states": 751,
          "teacher_agreement": 0.96271637816245,
          "teacher_ev_regret": 0.0008223085365346206
        },
        "1": {
          "states": 759,
          "teacher_agreement": 0.9525691699604744,
          "teacher_ev_regret": 0.0009504321275161915
        },
        "5": {
          "states": 550,
          "teacher_agreement": 0.96,
          "teacher_ev_regret": 0.0007164417613636364
        }
      },
      "by_stratum": {
        "general": {
          "states": 2048,
          "teacher_agreement": 0.9638671875,
          "teacher_ev_regret": 0.000675438124557217
        },
        "depleted": {
          "states": 2048,
          "teacher_agreement": 0.9453125,
          "teacher_ev_regret": 0.0013468597615376508
        }
      },
      "by_teacher": {
        "monte_carlo": {
          "states": 3589,
          "teacher_agreement": 0.9543048202842017,
          "teacher_ev_regret": 0.0010268667608491224
        },
        "exact": {
          "states": 507,
          "teacher_agreement": 0.9566074950690335,
          "teacher_ev_regret": 0.0008998841539147721
        }
      }
    },
    "selection_eligible": true
  },
  "elapsed_seconds": 11041.982143932022,
  "gpu_peak_allocated_gb": 8.453748736,
  "limitations": "Reference scope is recorded in reference_method. Monte Carlo fallbacks use basic continuation and heuristic stopping. Teacher agreement does not establish global optimality or profit. Candidate is not auto-activated.",
  "test_was_deferred": true,
  "test_inference": "Serving SDK, after global checkpoint selection was frozen",
  "selection_freeze_sha256": "e0e00bcef8a035507186161eff1fb5d7e30d08767f651331e1b8936adfdf6e2a",
  "release_evaluation": {
    "test": {
      "states": 16384,
      "teacher_agreement": 0.95947265625,
      "teacher_ev_regret": 0.0010142500523316195,
      "hit_bust_brier": 0.0002707670232742885,
      "dealer_brier": 0.0008787809315606064,
      "action_brier": 0.0608471745513916,
      "teacher_action_ece": 0.011771899414062644
    },
    "baseline": {
      "states": 16384,
      "teacher_agreement": 0.9283447265625,
      "teacher_ev_regret": 0.0032042316146959245,
      "hit_bust_brier": 0.0016498112438239993,
      "dealer_brier": 0.0025270695927741118,
      "action_brier": 0.10846779101623535,
      "teacher_action_ece": 0.01973449707031245
    },
    "baseline_label": "Broad-training baseline",
    "selection_states": 8192,
    "dataset_sha256": "1f261a9e6f27778e237680295810e3b979ed4efa7edd3dedfe3e5b9eb45cc018",
    "selection_sha256": "1ae28756bbe9ce00bbb0a85c9cfd0d56285257c71cbc243de9002e9269c450f6",
    "inference_mode": "sdk",
    "scope": "Fresh release test; original training and calibration records are preserved separately."
  }
}