results: pass calibrated oral-B-v2 development gate

1 files changed, 461 insertions, 0 deletions

diff --git a/results/bci_v2_calibrated_dev/bci_v2_calibrated_t26_m0.json b/results/bci_v2_calibrated_dev/bci_v2_calibrated_t26_m0.json
new file mode 100644
index 0000000..b2fc4f3
--- /dev/null
+++ b/results/bci_v2_calibrated_dev/bci_v2_calibrated_t26_m0.json
@@ -0,0 +1,461 @@
+{
+  "args": {
+    "model_seed": 0,
+    "outdir": "results/bci_v2_calibrated_dev",
+    "task_seed": 26
+  },
+  "assays": {
+    "challenge": {
+      "active_state_episode_steps_by_mode": {
+        "acute_critic_lesion": 12882,
+        "acute_outcome_lesion": 12882,
+        "intact": 12882
+      },
+      "calibration": {
+        "active_state_episode_steps": 14336,
+        "episodes": 512,
+        "maximum_cursor_summary": {
+          "maximum": 1.9664530754089355,
+          "median": 1.9004390239715576,
+          "minimum": 1.6664398908615112
+        },
+        "quantiles": [
+          0.2,
+          0.35,
+          0.5,
+          0.65,
+          0.8
+        ],
+        "seed": 550026,
+        "targets": [
+          1.859693741798401,
+          1.8864485919475555,
+          1.9004749059677124,
+          1.9156511783599854,
+          1.929451322555542
+        ],
+        "uses_outcome_labels": false
+      },
+      "episodes_per_target": 128,
+      "maximum_steps_per_episode": 28,
+      "selection_over_evaluation_outcomes": false,
+      "targets": [
+        1.859693741798401,
+        1.8864485919475555,
+        1.9004749059677124,
+        1.9156511783599854,
+        1.929451322555542
+      ],
+      "trajectory_seeds": {
+        "1": 586000,
+        "2": 586001,
+        "3": 586002,
+        "4": 586003,
+        "5": 586004
+      }
+    },
+    "performance_evaluation_episodes": 256,
+    "performance_evaluation_seed": 540026
+  },
+  "conditions": {
+    "critic_training_lesion": {
+      "cost": {
+        "active_state_episode_steps": 25088,
+        "cursor_scalar_observations": 12544,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 6272,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.0,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 7168,
+      "final_success": 0.0,
+      "late_success": 0.0,
+      "learning_gain": 0.0,
+      "role_cosine_after_training": 0.9942126274108887,
+      "training_wall_s": 0.21116646006703377
+    },
+    "fixed_role": {
+      "cost": {
+        "active_state_episode_steps": 25088,
+        "cursor_scalar_observations": 12544,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 6272,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.06177661940455437,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 7168,
+      "final_success": 0.0,
+      "late_success": 0.0,
+      "learning_gain": 0.0,
+      "role_cosine_after_training": -0.1393236517906189,
+      "training_wall_s": 0.21275918185710907
+    },
+    "intact": {
+      "cost": {
+        "active_state_episode_steps": 16204,
+        "cursor_scalar_observations": 8554,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 4277,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.5207661390304565,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.015625,
+        0.203125,
+        0.6875,
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 263,
+      "final_success": 1.0,
+      "late_success": 1.0,
+      "learning_gain": 1.0,
+      "role_cosine_after_training": 0.9405146241188049,
+      "training_wall_s": 0.2710344232618809
+    },
+    "oracle_role": {
+      "cost": {
+        "active_state_episode_steps": 15936,
+        "cursor_scalar_observations": 8436,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 4218,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.45849862694740295,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.03125,
+        0.265625,
+        0.796875,
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 267,
+      "final_success": 1.0,
+      "late_success": 1.0,
+      "learning_gain": 1.0,
+      "role_cosine_after_training": 1.0000001192092896,
+      "training_wall_s": 0.2009308896958828
+    },
+    "outcome_training_lesion": {
+      "cost": {
+        "active_state_episode_steps": 19102,
+        "cursor_scalar_observations": 9788,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 4894,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.18997597694396973,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.015625,
+        0.15625,
+        0.25,
+        0.671875,
+        0.8125,
+        0.96875,
+        1.0,
+        1.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 641,
+      "final_success": 1.0,
+      "late_success": 0.9895833333333334,
+      "learning_gain": 0.9895833333333334,
+      "role_cosine_after_training": 0.9652196764945984,
+      "training_wall_s": 0.22571271285414696
+    },
+    "plasticity_lesion": {
+      "cost": {
+        "active_state_episode_steps": 25088,
+        "cursor_scalar_observations": 12544,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 6272,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.06284985691308975,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 7168,
+      "final_success": 0.0,
+      "late_success": 0.0,
+      "learning_gain": 0.0,
+      "role_cosine_after_training": 0.9942126870155334,
+      "training_wall_s": 0.20297767594456673
+    }
+  },
+  "config": {
+    "context_ar": 0.8,
+    "context_dim": 16,
+    "coupling_scale": 1.0,
+    "critic_eta": 0.03,
+    "days": 14,
+    "eligibility_decay": 0.8,
+    "episodes_per_day": 64,
+    "feedback": "performance_velocity",
+    "forward_eta": 0.1,
+    "gamma": 0.8,
+    "inertia": 0.65,
+    "kappa": 0.0,
+    "n_background": 30,
+    "n_minus": 5,
+    "n_plus": 5,
+    "perturb_every": 4,
+    "perturb_sigma": 0.03,
+    "predictor_eta": 0.2,
+    "process_noise": 0.12,
+    "steps_per_episode": 28,
+    "target": 0.8,
+    "terminal_reward": 1.0,
+    "vectorizer_eta": 0.03,
+    "velocity_reward_scale": 1.0
+  },
+  "finite": true,
+  "hardware": {
+    "device": "cpu",
+    "platform": "Linux-5.15.0-161-generic-x86_64-with-glibc2.35",
+    "threads": 1,
+    "torch_version": "2.10.0+cu128"
+  },
+  "peak_rss_mib": 962.34375,
+  "protocol": {
+    "calibration_uses_outcome_labels": false,
+    "confirmation_seeds_touched": false,
+    "failed_target_gate_sha256": "1df37c0b639542bbee3017da54306b0d572cbf96230428f142fafed139ac92f1",
+    "fixed_config": {
+      "critic_eta": 0.03,
+      "forward_eta": 0.1,
+      "gamma": 0.8,
+      "velocity_reward_scale": 1.0
+    },
+    "hyperparameter_selection": false,
+    "name": "oral_b_v2_calibrated_recovery_development_v1",
+    "protocol_sha256": "a175dc80fdb7218c010953d5d7e3c77fee0690d5f10fabaf9d700bb8505acde8",
+    "selection_split": "development_validation"
+  },
+  "provenance": {
+    "git_commit": "d289a46293e0b422c195746ad8b310d4b50ded23",
+    "git_tracked_dirty": false,
+    "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"
+    },
+    "tracked_inputs": {
+      "base_dynamics": true,
+      "common_runner": true,
+      "confirmation_analyzer": true,
+      "confirmation_runner": true,
+      "d4_gate": true,
+      "development_analyzer": true,
+      "failed_target_gate": true,
+      "failed_v2_gate": true,
+      "old_r2_gate": true,
+      "protocol": true,
+      "recovery_confirmation_analyzer": true,
+      "recovery_development_analyzer": true,
+      "recovery_metrics": true,
+      "recovery_runner": true,
+      "runner": true,
+      "v2_dynamics": true,
+      "v2_metrics": true
+    }
+  },
+  "schema_version": 4,
+  "signatures": {
+    "acute_outcome_lesion_outcome_balanced_acc": 0.5002197802197802,
+    "acute_outcome_lesion_role_aligned_separation": 0.0005471409294360297,
+    "causal_role_sign_inversion_index": 0.03878189930912869,
+    "challenge_episodes": 640,
+    "challenge_failure_count": 315,
+    "challenge_success_count": 325,
+    "challenge_success_fraction": 0.5078125,
+    "critic_contribution_value_prediction_corr": 0.9999999999995302,
+    "decoder_distance_residual_corr": 0.11469715057185319,
+    "mean_abs_raw_soma_corr": 0.9992015754075411,
+    "mean_abs_residual_soma_corr": 0.0563586791161443,
+    "mean_critic_expectedness_contribution": 0.24124016763074801,
+    "nonterminal_training_events": 15308,
+    "raw_minus_residual_abs_soma_corr": 0.9428428962913967,
+    "role_aligned_error_cv_corr": 0.32966432771863374,
+    "role_aligned_velocity_cv_corr": 0.9988614599944285,
+    "surrounding_event_decoder_balanced_acc": 0.5539961311201976,
+    "target_success_fraction": {
+      "1.859693741798401": 0.828125,
+      "1.8864485919475555": 0.71875,
+      "1.9004749059677124": 0.4453125,
+      "1.9156511783599854": 0.328125,
+      "1.929451322555542": 0.21875
+    },
+    "terminal_outcome_separation_drop_under_acute_lesion": 0.3956960119743165,
+    "terminal_previous_soma_outcome_balanced_acc": 0.7272771672771673,
+    "terminal_residual_minus_previous_soma_acc": 0.26796092796092796,
+    "terminal_residual_outcome_balanced_acc": 0.9952380952380953,
+    "terminal_role_aligned_outcome_separation": 0.3962431529037525,
+    "terminal_training_events": 896,
+    "velocity_minus_error_abs_cv_corr": 0.6691971322757948
+  },
+  "split": "development_validation",
+  "wall_s": 2.583215858787298,
+  "warmup": {
+    "critic_training_lesion": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602600,
+      "role_cosine_after_warmup": 0.9944602251052856,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": true
+    },
+    "fixed_role": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602600,
+      "role_cosine_after_warmup": -0.1393236517906189,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": false
+    },
+    "intact": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602600,
+      "role_cosine_after_warmup": 0.9944602251052856,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": true
+    },
+    "oracle_role": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602600,
+      "role_cosine_after_warmup": 1.0000001192092896,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": false
+    },
+    "outcome_training_lesion": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602600,
+      "role_cosine_after_warmup": 0.9944602251052856,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": true
+    },
+    "plasticity_lesion": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602600,
+      "role_cosine_after_warmup": 0.9944602251052856,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": true
+    }
+  }
+}