From 70e180c5ef5f78679f2e163ed3ee30873ee523bf Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Thu, 23 Jul 2026 08:13:02 -0500 Subject: results: pass calibrated oral-B-v2 development gate --- results/bci_v2_calibrated_dev_gate.json | 148 ++++++++++++++++++++++++++++++++ 1 file changed, 148 insertions(+) create mode 100644 results/bci_v2_calibrated_dev_gate.json (limited to 'results/bci_v2_calibrated_dev_gate.json') diff --git a/results/bci_v2_calibrated_dev_gate.json b/results/bci_v2_calibrated_dev_gate.json new file mode 100644 index 0000000..515f0e6 --- /dev/null +++ b/results/bci_v2_calibrated_dev_gate.json @@ -0,0 +1,148 @@ +{ + "calibrated_confirmation_opened": true, + "calibrated_targets_by_task_seed": { + "26": [ + 1.859693741798401, + 1.8864485919475555, + 1.9004749059677124, + 1.9156511783599854, + 1.929451322555542 + ], + "27": [ + 1.7139501571655273, + 1.7322039246559142, + 1.750077724456787, + 1.7649245738983155, + 1.783210277557373 + ], + "28": [ + 1.8872586727142333, + 1.9072148621082305, + 1.9221965074539185, + 1.9328129887580872, + 1.9453917741775513 + ] + }, + "calibration_quantiles": [ + 0.2, + 0.35, + 0.5, + 0.65, + 0.8 + ], + "challenge_success_fraction_by_task_seed": [ + 0.5078125, + 0.4875, + 0.5078125 + ], + "checks_by_task_seed": { + "26": { + "challenge_fraction_between_0p15_and_0p85": true, + "critic_contribution_value_corr_at_least_0p95": true, + "critic_expectedness_at_least_0p05": true, + "decoder_distance_corr_at_least_0p02": true, + "final_success_at_least_0p70": true, + "fixed_role_gap_at_least_0p20": true, + "learning_gain_at_least_0p10": true, + "oracle_deficit_at_most_0p10": true, + "outcome_lesion_separation_drop_at_least_0p20": true, + "plasticity_lesion_retains_at_most_half_gain": true, + "raw_residual_corr_gap_at_least_0p20": true, + "residual_soma_corr_at_most_0p10": true, + "role_cosine_at_least_0p80": true, + "sign_inversion_at_least_0p01": true, + "surrounding_event_accuracy_at_least_0p52": true, + "terminal_outcome_accuracy_at_least_0p75": true, + "terminal_role_separation_at_least_0p20": true, + "velocity_advantage_at_least_0p05": true + }, + "27": { + "challenge_fraction_between_0p15_and_0p85": true, + "critic_contribution_value_corr_at_least_0p95": true, + "critic_expectedness_at_least_0p05": true, + "decoder_distance_corr_at_least_0p02": true, + "final_success_at_least_0p70": true, + "fixed_role_gap_at_least_0p20": true, + "learning_gain_at_least_0p10": true, + "oracle_deficit_at_most_0p10": true, + "outcome_lesion_separation_drop_at_least_0p20": true, + "plasticity_lesion_retains_at_most_half_gain": true, + "raw_residual_corr_gap_at_least_0p20": true, + "residual_soma_corr_at_most_0p10": true, + "role_cosine_at_least_0p80": true, + "sign_inversion_at_least_0p01": true, + "surrounding_event_accuracy_at_least_0p52": true, + "terminal_outcome_accuracy_at_least_0p75": true, + "terminal_role_separation_at_least_0p20": true, + "velocity_advantage_at_least_0p05": true + }, + "28": { + "challenge_fraction_between_0p15_and_0p85": true, + "critic_contribution_value_corr_at_least_0p95": true, + "critic_expectedness_at_least_0p05": true, + "decoder_distance_corr_at_least_0p02": true, + "final_success_at_least_0p70": true, + "fixed_role_gap_at_least_0p20": true, + "learning_gain_at_least_0p10": true, + "oracle_deficit_at_most_0p10": true, + "outcome_lesion_separation_drop_at_least_0p20": true, + "plasticity_lesion_retains_at_most_half_gain": true, + "raw_residual_corr_gap_at_least_0p20": true, + "residual_soma_corr_at_most_0p10": true, + "role_cosine_at_least_0p80": true, + "sign_inversion_at_least_0p01": true, + "surrounding_event_accuracy_at_least_0p52": true, + "terminal_outcome_accuracy_at_least_0p75": true, + "terminal_role_separation_at_least_0p20": true, + "velocity_advantage_at_least_0p05": true + } + }, + "complete_grid": true, + "confirmation_seeds_touched": false, + "fixed_config": { + "critic_eta": 0.03, + "forward_eta": 0.1, + "gamma": 0.8, + "velocity_reward_scale": 1.0 + }, + "input_sha256": { + "base_dynamics": "d5a373314562af0daedf55baa04b975fe643236c7821be1560d2cfb4359688fc", + "common_runner": "4157a57806dedfc2eb70ec84f7a7ca1f73ab684e5d8f0b227cd80465e8e97aa5", + "confirmation_analyzer": "2536fa01f62c3b2d87e782b08075cb5948aa26bfdce71a1d8c2446d5b68bb34c", + "confirmation_runner": "5e9beba106b4b29c4ee1568052d6e8c15690dbde92917a2497a52e43e7f557d1", + "d4_gate": "636c587e47287338cdca7d9558dc3bfb99a2cec615badb23337025d85b8128ef", + "development_analyzer": "dde4a2e4d62f5f9da94357b388139c50251b6863f9c991f1a006e71eeb7953e4", + "failed_target_gate": "1df37c0b639542bbee3017da54306b0d572cbf96230428f142fafed139ac92f1", + "failed_v2_gate": "e53f46cf456d60ce4d33586ace4f4619fb1a31a58082b2b72943d4ed10cd25fd", + "old_r2_gate": "4f6f969ceae88afa2523e3472a3373522991f5ecaab69440830c07479d2d3597", + "protocol": "a175dc80fdb7218c010953d5d7e3c77fee0690d5f10fabaf9d700bb8505acde8", + "recovery_confirmation_analyzer": "29a168143266bc7078e5229183bb7ea45283973592fecd1fcd715488f5d6242b", + "recovery_development_analyzer": "2768b877132ece6d612d2a5f1e483c495b277e26fdf54fb8a716aa8de6fd67e2", + "recovery_metrics": "371cb6a7e2192d25ddbfad69c0ce7a1150aa1c21ca98e141bedb2afc2acbdbaa", + "recovery_runner": "cea0ef658c698c376b1fda0cf2f7f910ee44f7bcde153f2fbaf18e66a2d3661a", + "runner": "446ad26fc015eb79d28e0bb442c8aef71bdd1d0320910494cac20f92e3c7dc18", + "v2_dynamics": "786c5aed3091644272ae0a740913ac3cc38cdb02a8cacb6dfc8496e36836cdae", + "v2_metrics": "421da9287b92e62c742f9eeefbfdccba582c6a4ad5cf92ade4709689120fa1ee" + }, + "intact_final_by_task_seed": [ + 1.0, + 1.0, + 1.0 + ], + "protocol": "oral_b_v2_calibrated_recovery_development_v1", + "review_score_after": 7, + "review_score_before": 7, + "score_change_rule": "development validation never changes the formal score", + "source_commit": "d289a46293e0b422c195746ad8b310d4b50ded23", + "source_sha256": { + "results/bci_v2_calibrated_dev/bci_v2_calibrated_t26_m0.json": "3ad58c0d4ee6aef1835f281c60c63b7c309ff23971692ba97631fd81d43c7165", + "results/bci_v2_calibrated_dev/bci_v2_calibrated_t27_m0.json": "3c61254dbf22ba1cb550f4c35aecba2375c754ac64ad383c642ecaaff8085670", + "results/bci_v2_calibrated_dev/bci_v2_calibrated_t28_m0.json": "b09933f8776ae1941adb0db543d1f5df91e9287cb11b2536e27e3b9caafc5812" + }, + "status": "passed", + "terminal_outcome_accuracy_by_task_seed": [ + 0.9952380952380953, + 1.0, + 0.9984126984126984 + ] +} -- cgit v1.2.3