results: retain target-ladder recovery failure

1 files changed, 430 insertions, 0 deletions

diff --git a/results/bci_v2_recovery_dev/bci_v2_recovery_t23_m0.json b/results/bci_v2_recovery_dev/bci_v2_recovery_t23_m0.json
new file mode 100644
index 0000000..41aca9f
--- /dev/null
+++ b/results/bci_v2_recovery_dev/bci_v2_recovery_t23_m0.json
@@ -0,0 +1,430 @@
+{
+  "args": {
+    "model_seed": 0,
+    "outdir": "results/bci_v2_recovery_dev",
+    "task_seed": 23
+  },
+  "assays": {
+    "challenge": {
+      "active_state_episode_steps_by_mode": {
+        "acute_critic_lesion": 10799,
+        "acute_outcome_lesion": 10799,
+        "intact": 10799
+      },
+      "episodes_per_target": 128,
+      "maximum_steps_per_episode": 28,
+      "selection_over_targets": false,
+      "targets": [
+        1.55,
+        1.6,
+        1.65,
+        1.7,
+        1.75,
+        1.8
+      ],
+      "trajectory_seeds": {
+        "1.55": 493000,
+        "1.6": 493001,
+        "1.65": 493002,
+        "1.7": 493003,
+        "1.75": 493004,
+        "1.8": 493005
+      }
+    },
+    "performance_evaluation_episodes": 256,
+    "performance_evaluation_seed": 460023
+  },
+  "conditions": {
+    "critic_training_lesion": {
+      "cost": {
+        "active_state_episode_steps": 25069,
+        "cursor_scalar_observations": 12536,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 6268,
+        "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,
+        0.015625,
+        0.0,
+        0.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 7168,
+      "final_success": 0.0,
+      "late_success": 0.005208333333333333,
+      "learning_gain": 0.005208333333333333,
+      "role_cosine_after_training": 0.9950354099273682,
+      "training_wall_s": 0.21576084941625595
+    },
+    "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": 0.061243072152137756,
+      "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": 0.2373625412583351
+    },
+    "intact": {
+      "cost": {
+        "active_state_episode_steps": 14590,
+        "cursor_scalar_observations": 7848,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 3924,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.4425785541534424,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.046875,
+        0.140625,
+        0.703125,
+        1.0,
+        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": 0.9832371473312378,
+      "training_wall_s": 0.2712929956614971
+    },
+    "oracle_role": {
+      "cost": {
+        "active_state_episode_steps": 14626,
+        "cursor_scalar_observations": 7862,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 3931,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.45056235790252686,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.046875,
+        0.140625,
+        0.703125,
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 262,
+      "final_success": 1.0,
+      "late_success": 1.0,
+      "learning_gain": 1.0,
+      "role_cosine_after_training": 1.0000001192092896,
+      "training_wall_s": 0.1980770006775856
+    },
+    "outcome_training_lesion": {
+      "cost": {
+        "active_state_episode_steps": 18170,
+        "cursor_scalar_observations": 9358,
+        "maximum_state_episode_steps": 25088,
+        "reverse_mode_calls": 0,
+        "role_probe_examples": 4679,
+        "task_loss_queries": 0,
+        "terminal_outcome_observations": 896
+      },
+      "critic_l2_after_training": 0.1713862419128418,
+      "daily_success": [
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.0,
+        0.046875,
+        0.0625,
+        0.21875,
+        0.59375,
+        0.828125,
+        0.90625,
+        0.96875,
+        1.0,
+        0.984375
+      ],
+      "early_success": 0.0,
+      "evaluation_active_state_episode_steps": 1534,
+      "final_success": 0.98828125,
+      "late_success": 0.984375,
+      "learning_gain": 0.984375,
+      "role_cosine_after_training": 0.9750589728355408,
+      "training_wall_s": 0.22877522557973862
+    },
+    "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": 0.06210753321647644,
+      "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": 0.9950442314147949,
+      "training_wall_s": 0.20400389283895493
+    }
+  },
+  "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": 998.17578125,
+  "protocol": {
+    "confirmation_seeds_touched": false,
+    "d4_gate_sha256": "636c587e47287338cdca7d9558dc3bfb99a2cec615badb23337025d85b8128ef",
+    "failed_v2_gate_sha256": "e53f46cf456d60ce4d33586ace4f4619fb1a31a58082b2b72943d4ed10cd25fd",
+    "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_cold_start_recovery_development_v1",
+    "old_r2_gate_sha256": "4f6f969ceae88afa2523e3472a3373522991f5ecaab69440830c07479d2d3597",
+    "protocol_sha256": "0621b47367074cb99ee74b332192878e76acbca11678b5238e1312adc2a2bcd9",
+    "selection_split": "development_validation"
+  },
+  "provenance": {
+    "git_commit": "dae172b1425c646f0441609c28fc5156b50f820d",
+    "git_tracked_dirty": false,
+    "input_sha256": {
+      "base_dynamics": "d5a373314562af0daedf55baa04b975fe643236c7821be1560d2cfb4359688fc",
+      "confirmation_analyzer": "29a168143266bc7078e5229183bb7ea45283973592fecd1fcd715488f5d6242b",
+      "confirmation_runner": "0582ae7a0c173d2fef842ea57d75b77b23cb6f70391f7013ad27ec29e0fedfdf",
+      "d4_gate": "636c587e47287338cdca7d9558dc3bfb99a2cec615badb23337025d85b8128ef",
+      "development_analyzer": "2768b877132ece6d612d2a5f1e483c495b277e26fdf54fb8a716aa8de6fd67e2",
+      "failed_v2_gate": "e53f46cf456d60ce4d33586ace4f4619fb1a31a58082b2b72943d4ed10cd25fd",
+      "old_r2_gate": "4f6f969ceae88afa2523e3472a3373522991f5ecaab69440830c07479d2d3597",
+      "protocol": "0621b47367074cb99ee74b332192878e76acbca11678b5238e1312adc2a2bcd9",
+      "recovery_metrics": "371cb6a7e2192d25ddbfad69c0ce7a1150aa1c21ca98e141bedb2afc2acbdbaa",
+      "runner": "cea0ef658c698c376b1fda0cf2f7f910ee44f7bcde153f2fbaf18e66a2d3661a",
+      "v2_dynamics": "786c5aed3091644272ae0a740913ac3cc38cdb02a8cacb6dfc8496e36836cdae",
+      "v2_metrics": "421da9287b92e62c742f9eeefbfdccba582c6a4ad5cf92ade4709689120fa1ee"
+    },
+    "tracked_inputs": {
+      "base_dynamics": true,
+      "confirmation_analyzer": true,
+      "confirmation_runner": true,
+      "d4_gate": true,
+      "development_analyzer": true,
+      "failed_v2_gate": true,
+      "old_r2_gate": true,
+      "protocol": true,
+      "recovery_metrics": true,
+      "runner": true,
+      "v2_dynamics": true,
+      "v2_metrics": true
+    }
+  },
+  "schema_version": 3,
+  "signatures": {
+    "acute_outcome_lesion_outcome_balanced_acc": 0.5541654667379767,
+    "acute_outcome_lesion_role_aligned_separation": 0.015703842176951255,
+    "causal_role_sign_inversion_index": 0.04036224680362202,
+    "challenge_episodes": 768,
+    "challenge_failure_count": 223,
+    "challenge_success_count": 545,
+    "challenge_success_fraction": 0.7096354166666666,
+    "critic_contribution_value_prediction_corr": 0.9999999999993424,
+    "decoder_distance_residual_corr": 0.11840537575539738,
+    "mean_abs_raw_soma_corr": 0.9990800248304881,
+    "mean_abs_residual_soma_corr": 0.05415501113040725,
+    "mean_critic_expectedness_contribution": 0.3046008384890047,
+    "nonterminal_training_events": 13694,
+    "raw_minus_residual_abs_soma_corr": 0.9449250137000809,
+    "role_aligned_error_cv_corr": 0.35695106277832594,
+    "role_aligned_velocity_cv_corr": 0.9984821724524531,
+    "surrounding_event_decoder_balanced_acc": 0.551090861356458,
+    "target_success_fraction": {
+      "1.55": 1.0,
+      "1.6": 0.984375,
+      "1.65": 0.96875,
+      "1.7": 0.7578125,
+      "1.75": 0.421875,
+      "1.8": 0.125
+    },
+    "terminal_outcome_separation_drop_under_acute_lesion": 0.40689127104423845,
+    "terminal_previous_soma_outcome_balanced_acc": 0.5775167647179824,
+    "terminal_residual_minus_previous_soma_acc": 0.4224832352820176,
+    "terminal_residual_outcome_balanced_acc": 1.0,
+    "terminal_role_aligned_outcome_separation": 0.4225951132211897,
+    "terminal_training_events": 896,
+    "velocity_minus_error_abs_cv_corr": 0.6415311096741272
+  },
+  "split": "development",
+  "wall_s": 2.6419225111603737,
+  "warmup": {
+    "critic_training_lesion": {
+      "batches": 100,
+      "examples": 6400,
+      "instruction_present": false,
+      "predictor_max_abs_error": 2.384185791015625e-07,
+      "rng_seed": 602300,
+      "role_cosine_after_warmup": 0.9931626915931702,
+      "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": 602300,
+      "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": 602300,
+      "role_cosine_after_warmup": 0.9931626915931702,
+      "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": 602300,
+      "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": 602300,
+      "role_cosine_after_warmup": 0.9931626915931702,
+      "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": 602300,
+      "role_cosine_after_warmup": 0.9931626915931702,
+      "role_cursor_scalar_observations": 12800,
+      "role_update_enabled": true
+    }
+  }
+}