{ "arm": "lambda025", "selected_step": 282, "temperature": 1.3231350559653137, "temperatures": { "choice": 1.3231350559653137, "noul": 1.3231350559653137, "score": 1.3231350559653137 }, "fit": { "temperature": 1.3231350559653137, "inverse_temperature": 0.7557807462598249, "status": "relative-bracket-converged", "boundary_hit": false, "iterations": 38, "beta_final_bracket": [ 0.7557807462235361, 0.7557807462961137 ], "objective_at_T1": 0.44021343659702844, "objective_at_selected_T": 0.4260766624652933, "derivative_at_selected_beta": -1.8373920391834685e-12, "initial_boundary_derivatives": { "beta_0.05": -3.6546078204867998, "beta_20": 0.255508441854847 }, "derivative_at_beta1": 0.10257030038687498, "initial_boundary_objectives": [ 1.0580632285716105, 5.133197394579512 ] }, "raw": { "rows": 1600, "accuracy_fraction": [ 1511, 1800 ], "accuracy": 0.8394444444444444, "brier": 0.2232446037975029, "nll": 0.44021343659702844, "score_rps": 0.034960309670060405, "by_family": { "old/stage4_dense_table": [ 92, 100 ], "old/stage4_scope": [ 95, 100 ], "old/stage4_arithmetic": [ 79, 100 ], "old/stage4_ordinal": [ 89, 100 ], "old/stage4_natural_nli": [ 180, 200 ], "old/stage4_automaton": [ 52, 100 ], "old/stage4_boolean": [ 71, 100 ], "old/stage4_relations": [ 68, 100 ], "old/stage4_registers": [ 74, 100 ], "natural/natural_squad2_answerability": [ 185, 200 ], "natural/natural_cosmos_qa": [ 171, 200 ], "natural/natural_snli": [ 178, 200 ] }, "temperature": 1.0, "probability_source": "production-T1-normalized" }, "calibrated": { "rows": 1600, "accuracy_fraction": [ 1511, 1800 ], "accuracy": 0.8394444444444444, "brier": 0.22190586234646303, "nll": 0.4260766624652933, "score_rps": 0.03535865911571265, "by_family": { "old/stage4_dense_table": [ 92, 100 ], "old/stage4_scope": [ 95, 100 ], "old/stage4_arithmetic": [ 79, 100 ], "old/stage4_ordinal": [ 89, 100 ], "old/stage4_natural_nli": [ 180, 200 ], "old/stage4_automaton": [ 52, 100 ], "old/stage4_boolean": [ 71, 100 ], "old/stage4_relations": [ 68, 100 ], "old/stage4_registers": [ 74, 100 ], "natural/natural_squad2_answerability": [ 185, 200 ], "natural/natural_cosmos_qa": [ 171, 200 ], "natural/natural_snli": [ 178, 200 ] }, "temperature": 1.3231350559653137, "probability_source": "binary64-logits-diagnostic" }, "inputs": { "old": { "data_sha256": "5ba3bd7527da831d77cec8c6a2f260a71e10a57742e356510c719f8f30d61a70", "predictions_sha256": "7212da3cc1423e28f046003b68287ba46eb51b196a37f96ce1f826e489fb5bea", "summary_sha256": "8181fa1f1b5c0be1b74730d441ee0c437779533e0ad400c7488c555896cf6453" }, "natural": { "data_sha256": "cc75c902fe9b51c4654d6ceaeb590fe135bdef6377d45bb774e391a66bb1fa25", "predictions_sha256": "3553fb8bbae557ad6fcc2ad8962baf704525a0f9f54cda43e650019e05000a32", "summary_sha256": "380fb6686d467c19f9579ceec31ada73c4c3b6369fc3e8eb9e7599727409f7b8" } }, "protocol_sha256": "843b6c9f4db403c14e2e2e5ed1314c72f8d174b2fe66cb7a80a238d1091439c6", "weighted_math_sha256": "19c8dd94e9f078810be943bdcbf946b842a20ba77aac359a7faade5f62b19709", "model_manifest_sha256": "9755d4c393ebfba77fa48bcb7b0949323c468ec68839fe6a52674187b85cc260", "winner_lock_sha256": "b1e182dadfbbb46a26eedf423eec702c72f0ccbd36e6ccc15664dc7fd5a0779f", "both_winners_lock_sha256": "a53765cf1ce7c541a11980538a858793ee967abb166b86dd32b07de1548f66bc", "heldout_used": false, "SELECT_used_for_temperature": false, "qualification_established": false, "dev_sha256": "45b4cd46acc4be53b95c7ed8f333b3533972296a4d0557c7b8c48c9cab6ced61", "raw_predictions_sha256": "ba311f5a6bc73182651bdf5c9744f0ed95921110501ddc93dcfd20a1d06bceff", "ranking_unchanged": true, "ranking_unchanged_scope": "Positive scalar preserves mathematical within-row logit order; native float32 outputs still require the separate full public reload proof.", "calibration_objective": "0.5mean9(old family mean NLL)+0.5mean3(natural family mean NLL)", "calibration_weights": { "old_panel": [ 1, 2 ], "natural_panel": [ 1, 2 ], "each_old_family": [ 1, 18 ], "each_natural_family": [ 1, 6 ] }, "calibration_role": "original-two-panel-CAL", "uniform_1600_fit": false, "temperature_refitted": false, "bridge": { "schema": "Nox-retention-CAL-export-v1", "family": "Nox", "arm": "lambda025", "rows": 1600, "panel_order": [ "old", "natural" ], "panel_rows": { "old": 1000, "natural": 600 }, "original_temperature_sha256": "4839d6c75d9710c8db5485802a2391a9d15bb7da5d3085d6b59b853dc91bcf51", "producer_complete_sha256": "40f185c928724ceb10d1aecbe35ef36b633f5330ec9db5a19b2bfc7bd22af693", "producer_freeze_sha256": "f012c928ec060983f5a6523ed5d24c7913af32978247ff4b00317ef1e97957f2", "original_protocol_sha256": "843b6c9f4db403c14e2e2e5ed1314c72f8d174b2fe66cb7a80a238d1091439c6", "ranking_sha256": "fe4910d9eef281729279ce9c9fda7eed77fbea431910d35f8876355cf3058e6c", "sequential_B8_boundary_preserved": true, "normalization_profile_sha256": "be32858d15233e0a3fbee0e4257fb02be0b3439deee4eb9c3f61151df7b73850" } }