manual-selector-sweep-paper-sync 2026-07-04T04:48:53Z workspace (part 3)

workspace/runs/ctt_selector_diagnostic_sweep/command.txt

@@ -0,0 +1 @@
+python scripts/build_selector_diagnostic_sweep.py 

workspace/runs/ctt_selector_diagnostic_sweep/config.yaml

@@ -0,0 +1,20 @@
+out_dir: runs/ctt_selector_diagnostic_sweep
+selected_min: 0.4745
+proposal_oracle_min: 0.5
+selector_gap_max: 0.03
+metrics:
+  - runs/ctt_base_context_obs_learned_dominance_chartcompat_obs_utility_task_envclip_k16_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test_tau0/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_basic_tanh_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_tangent_tanh_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_tanh_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_dominance_tanh_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_perdim_trainmax_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_success_perdim_trainmax_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_tangent_perdim_trainmax_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_learned_dominance_context_tangent_success_perdim_trainmax_train_to_test/metrics.json
+  - runs/ctt_base_context_obs_dominance_perdim_trainmax_train_to_test/metrics.json

workspace/runs/ctt_selector_diagnostic_sweep/data_hash.txt

@@ -0,0 +1 @@
+2ec8fa640a2fa299e25d9a7decf951077112e2460bbb2c14deab81745b8249af

workspace/runs/ctt_selector_diagnostic_sweep/eval.log

@@ -0,0 +1,3 @@
+num_inputs=15
+families=K8 per-dim trainmax,K8 tanh,K16 env_clip
+overall_pass=False

workspace/runs/ctt_selector_diagnostic_sweep/git_hash.txt

@@ -0,0 +1 @@
+14e81292e2004925f81e0bb4810f6f144bfd87b7

workspace/runs/ctt_selector_diagnostic_sweep/metrics.json

@@ -0,0 +1,424 @@
+{
+  "best_by_family": [
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.034447199957771346,
+      "coverage": 0.4236111111111111,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.5763888888888888,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_perdim_trainmax_train_to_test",
+      "selected_success": 0.3333333333333333,
+      "selector": "context/utility_margin, bonus=1.0",
+      "selector_regret": 0.11280556699623251,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.0625,
+      "success_support_gap": 0.5138888888888888
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.005483674479356937,
+      "coverage": 0.18055555555555555,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8194444444444444,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test",
+      "selected_success": 0.3819444444444444,
+      "selector": "context/success, bonus=1.0",
+      "selector_regret": 0.23532104610300544,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.125,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.2916666666666667,
+      "calibration_ece": 0.015032666932530015,
+      "coverage": 0.6597222222222222,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.3402777777777778,
+      "family": "K16 env_clip",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 16,
+      "outcome_ptr": 0.5486111111111112,
+      "proposal_oracle_success": 0.5694444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_chartcompat_obs_utility_task_envclip_k16_train_to_test",
+      "selected_success": 0.3541666666666667,
+      "selector": "chart_compat/utility_margin, chartcompat_obs",
+      "selector_regret": 0.4511945542892338,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.24305555555555555,
+      "success_support_gap": 0.2013888888888889
+    }
+  ],
+  "data_hash": "2ec8fa640a2fa299e25d9a7decf951077112e2460bbb2c14deab81745b8249af",
+  "gates": [
+    {
+      "family": "K8 per-dim trainmax",
+      "pass": false,
+      "proposal_oracle_success": 0.2222222222222222,
+      "selected_success": 0.3333333333333333,
+      "selector": "context/utility_margin, bonus=1.0",
+      "status": "diagnostic_only",
+      "success_selector_gap": 0.0625
+    },
+    {
+      "family": "K8 tanh",
+      "pass": false,
+      "proposal_oracle_success": 0.3819444444444444,
+      "selected_success": 0.3819444444444444,
+      "selector": "context/success, bonus=1.0",
+      "status": "diagnostic_only",
+      "success_selector_gap": 0.125
+    },
+    {
+      "family": "K16 env_clip",
+      "pass": false,
+      "proposal_oracle_success": 0.5694444444444444,
+      "selected_success": 0.3541666666666667,
+      "selector": "chart_compat/utility_margin, chartcompat_obs",
+      "status": "diagnostic_only",
+      "success_selector_gap": 0.24305555555555555
+    }
+  ],
+  "input_metrics": [
+    "runs/ctt_base_context_obs_learned_dominance_chartcompat_obs_utility_task_envclip_k16_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test_tau0/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_basic_tanh_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_tangent_tanh_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_tanh_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_dominance_tanh_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_perdim_trainmax_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_success_perdim_trainmax_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_tangent_perdim_trainmax_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_learned_dominance_context_tangent_success_perdim_trainmax_train_to_test/metrics.json",
+    "runs/ctt_base_context_obs_dominance_perdim_trainmax_train_to_test/metrics.json"
+  ],
+  "num_inputs": 15,
+  "overall_pass": false,
+  "report_type": "ctt_selector_diagnostic_sweep",
+  "rows": [
+    {
+      "base_success": 0.2916666666666667,
+      "calibration_ece": 0.015032666932530015,
+      "coverage": 0.6597222222222222,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.3402777777777778,
+      "family": "K16 env_clip",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 16,
+      "outcome_ptr": 0.5486111111111112,
+      "proposal_oracle_success": 0.5694444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_chartcompat_obs_utility_task_envclip_k16_train_to_test",
+      "selected_success": 0.3541666666666667,
+      "selector": "chart_compat/utility_margin, chartcompat_obs",
+      "selector_regret": 0.4511945542892338,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.24305555555555555,
+      "success_support_gap": 0.2013888888888889
+    },
+    {
+      "base_success": 0.2916666666666667,
+      "calibration_ece": 0.16331725144301124,
+      "coverage": 0.13194444444444445,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8680555555555556,
+      "family": "K16 env_clip",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 16,
+      "outcome_ptr": 0.5486111111111112,
+      "proposal_oracle_success": 0.5694444444444444,
+      "report_type": "dominance_calibrated_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test",
+      "selected_success": 0.2777777777777778,
+      "selector": "LCB auto",
+      "selector_regret": 0.5323940152051768,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.2986111111111111,
+      "success_support_gap": 0.2013888888888889
+    },
+    {
+      "base_success": 0.2916666666666667,
+      "calibration_ece": 0.16331725144301124,
+      "coverage": 0.1111111111111111,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8888888888888888,
+      "family": "K16 env_clip",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 16,
+      "outcome_ptr": 0.5486111111111112,
+      "proposal_oracle_success": 0.5694444444444444,
+      "report_type": "dominance_calibrated_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test_tau0",
+      "selected_success": 0.2916666666666667,
+      "selector": "LCB 0",
+      "selector_regret": 0.5319753833795807,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.2986111111111111,
+      "success_support_gap": 0.2013888888888889
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.02645474340097932,
+      "coverage": 0.2569444444444444,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.7430555555555556,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_basic_tanh_train_to_test",
+      "selected_success": 0.3680555555555556,
+      "selector": "basic/utility_margin, bonus=1.0",
+      "selector_regret": 0.25259616218520226,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.13194444444444445,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.005483674479356937,
+      "coverage": 0.18055555555555555,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8194444444444444,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test",
+      "selected_success": 0.3819444444444444,
+      "selector": "context/success, bonus=1.0",
+      "selector_regret": 0.23532104610300544,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.125,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.03920034766708541,
+      "coverage": 0.18055555555555555,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8194444444444444,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_train_to_test",
+      "selected_success": 0.3402777777777778,
+      "selector": "context_tangent/success, bonus=1.0",
+      "selector_regret": 0.26619138512534035,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.14583333333333334,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.053218966408725234,
+      "coverage": 0.3125,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.6875,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_tangent_tanh_train_to_test",
+      "selected_success": 0.3472222222222222,
+      "selector": "context_tangent/utility_margin, bonus=1.0",
+      "selector_regret": 0.2437624827079061,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.13194444444444445,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.02303437704979959,
+      "coverage": 0.1111111111111111,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8888888888888888,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_tanh_train_to_test",
+      "selected_success": 0.375,
+      "selector": "context/utility_margin, bonus=1.0",
+      "selector_regret": 0.24082847581949965,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.13194444444444445,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.12239407797336463,
+      "coverage": 0.11805555555555555,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.8819444444444444,
+      "family": "K8 tanh",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.4166666666666667,
+      "proposal_oracle_success": 0.3819444444444444,
+      "report_type": "dominance_calibrated_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_dominance_tanh_train_to_test",
+      "selected_success": 0.3194444444444444,
+      "selector": "LCB auto",
+      "selector_regret": 0.3232221397451617,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.1736111111111111,
+      "success_support_gap": 0.3680555555555556
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.010301400843750107,
+      "coverage": 0.2361111111111111,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.7638888888888888,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_basic_perdim_trainmax_train_to_test",
+      "selected_success": 0.3055555555555556,
+      "selector": "basic/utility_margin, bonus=1.0",
+      "selector_regret": 0.12719054139840105,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.06944444444444445,
+      "success_support_gap": 0.5138888888888888
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.034447199957771346,
+      "coverage": 0.4236111111111111,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.5763888888888888,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_perdim_trainmax_train_to_test",
+      "selected_success": 0.3333333333333333,
+      "selector": "context/utility_margin, bonus=1.0",
+      "selector_regret": 0.11280556699623251,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.0625,
+      "success_support_gap": 0.5138888888888888
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.03148814961718582,
+      "coverage": 0.006944444444444444,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.9930555555555556,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_success_perdim_trainmax_train_to_test",
+      "selected_success": 0.3263888888888889,
+      "selector": "context/success, bonus=1.0",
+      "selector_regret": 0.11845623132669264,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.0625,
+      "success_support_gap": 0.5138888888888888
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.01449932655549304,
+      "coverage": 0.6111111111111112,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.3888888888888889,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_tangent_perdim_trainmax_train_to_test",
+      "selected_success": 0.2986111111111111,
+      "selector": "context_tangent/utility_margin, bonus=1.0",
+      "selector_regret": 0.12803097097073785,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.06944444444444445,
+      "success_support_gap": 0.5138888888888888
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.015235808250645553,
+      "coverage": 0.09027777777777778,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.9097222222222222,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "learned_dominance_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_learned_dominance_context_tangent_success_perdim_trainmax_train_to_test",
+      "selected_success": 0.2777777777777778,
+      "selector": "context_tangent/success, bonus=1.0",
+      "selector_regret": 0.16677976430704197,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.09027777777777778,
+      "success_support_gap": 0.5138888888888888
+    },
+    {
+      "base_success": 0.3333333333333333,
+      "calibration_ece": 0.11740366311592768,
+      "coverage": 0.006944444444444444,
+      "data_hash": "0c38d73dc9bc4ab756e12006671b5d12f1a23358b281e54b3f6a9ae3cbf6e9af",
+      "fallback_rate": 0.9930555555555556,
+      "family": "K8 per-dim trainmax",
+      "hidden_chart_oracle_success": 0.7291666666666666,
+      "k": 8,
+      "outcome_ptr": 0.3611111111111111,
+      "proposal_oracle_success": 0.2222222222222222,
+      "report_type": "dominance_calibrated_selector_eval",
+      "run_path": "runs/ctt_base_context_obs_dominance_perdim_trainmax_train_to_test",
+      "selected_success": 0.3263888888888889,
+      "selector": "LCB auto",
+      "selector_regret": 0.11593854168636931,
+      "split_hash": "84151df9e88c219b76751677558cc5bbbc19082bf358cea7330158e9e403e25f",
+      "success_selector_gap": 0.0625,
+      "success_support_gap": 0.5138888888888888
+    }
+  ],
+  "schema_version": 1,
+  "selection_rule": "best selected_success per diagnostic family; all candidate rows retained",
+  "split_hash": "3281cb880470bb0158a5055eeb059e90f2af4a005944132e01504c225160d8c5",
+  "thresholds": {
+    "proposal_oracle_min": 0.5,
+    "selected_min": 0.4745,
+    "selector_gap_max": 0.03
+  }
+}

workspace/runs/ctt_selector_diagnostic_sweep/metrics_by_seed.json

@@ -0,0 +1,22 @@
+{
+  "16": {
+    "base_success": 0.2916666666666667,
+    "calibration_ece": 0.11388905660618416,
+    "coverage": 0.3009259259259259,
+    "outcome_ptr": 0.5486111111111112,
+    "proposal_oracle_success": 0.5694444444444444,
+    "selected_success": 0.3078703703703704,
+    "success_selector_gap": 0.28009259259259256,
+    "success_support_gap": 0.20138888888888892
+  },
+  "8": {
+    "base_success": 0.3333333333333333,
+    "calibration_ece": 0.04109681127667372,
+    "coverage": 0.21122685185185186,
+    "outcome_ptr": 0.38888888888888884,
+    "proposal_oracle_success": 0.30208333333333337,
+    "selected_success": 0.3333333333333333,
+    "success_selector_gap": 0.10474537037037035,
+    "success_support_gap": 0.4409722222222223
+  }
+}

workspace/runs/ctt_selector_diagnostic_sweep/metrics_by_task.json

@@ -0,0 +1,32 @@
+{
+  "K16 env_clip": {
+    "base_success": 0.2916666666666667,
+    "calibration_ece": 0.11388905660618416,
+    "coverage": 0.3009259259259259,
+    "outcome_ptr": 0.5486111111111112,
+    "proposal_oracle_success": 0.5694444444444444,
+    "selected_success": 0.3078703703703704,
+    "success_selector_gap": 0.28009259259259256,
+    "success_support_gap": 0.20138888888888892
+  },
+  "K8 per-dim trainmax": {
+    "base_success": 0.3333333333333333,
+    "calibration_ece": 0.03722925805679559,
+    "coverage": 0.22916666666666663,
+    "outcome_ptr": 0.3611111111111111,
+    "proposal_oracle_success": 0.22222222222222224,
+    "selected_success": 0.31134259259259256,
+    "success_selector_gap": 0.06944444444444445,
+    "success_support_gap": 0.5138888888888888
+  },
+  "K8 tanh": {
+    "base_success": 0.3333333333333333,
+    "calibration_ece": 0.044964364496551855,
+    "coverage": 0.19328703703703706,
+    "outcome_ptr": 0.4166666666666667,
+    "proposal_oracle_success": 0.3819444444444444,
+    "selected_success": 0.3553240740740741,
+    "success_selector_gap": 0.14004629629629628,
+    "success_support_gap": 0.3680555555555556
+  }
+}

workspace/runs/ctt_selector_diagnostic_sweep/split_hash.txt

@@ -0,0 +1 @@
+3281cb880470bb0158a5055eeb059e90f2af4a005944132e01504c225160d8c5

workspace/runs/ctt_selector_diagnostic_sweep/table.tex

@@ -0,0 +1,10 @@
+% Auto-generated by scripts/build_selector_diagnostic_sweep.py
+\begin{tabular}{llrrrrrr}
+\toprule
+Family & Best selector & Base & Selected & Proposal & Coverage & Sel. gap & Support gap \\
+\midrule
+K8 per-dim trainmax & context/utility\_margin, bonus=1.0 & 0.3333 & 0.3333 & 0.2222 & 0.4236 & 0.0625 & 0.5139 \\
+K8 tanh & context/success, bonus=1.0 & 0.3333 & 0.3819 & 0.3819 & 0.1806 & 0.1250 & 0.3681 \\
+K16 env\_clip & chart\_compat/utility\_margin, chartcompat\_obs & 0.2917 & 0.3542 & 0.5694 & 0.6597 & 0.2431 & 0.2014 \\
+\bottomrule
+\end{tabular}

workspace/runs/ctt_selector_diagnostic_sweep/train.log

@@ -0,0 +1 @@
+selector sweep artifact; source selectors trained separately

workspace/runs/paper_ctt_audit/audit.json

@@ -6,7 +6,7 @@
   "forbidden_patterns": [],
   "paper": "latex/main.tex",
   "paper_inputs": {
-    "num_inputs": 38,
+    "num_inputs": 39,
     "num_missing": 0,
     "rows": [
       {
@@ -243,28 +243,35 @@
         "status": "pass"
       },
       {
-        "line": 1065,
+        "line": 1067,
+        "raw": "../runs/ctt_selector_diagnostic_sweep/table",
+        "resolved": "runs/ctt_selector_diagnostic_sweep/table.tex",
+        "run_dir": "runs/ctt_selector_diagnostic_sweep",
+        "status": "pass"
+      },
+      {
+        "line": 1089,
         "raw": "../runs/ctt_outcome_acceptance_gate/table",
         "resolved": "runs/ctt_outcome_acceptance_gate/table.tex",
         "run_dir": "runs/ctt_outcome_acceptance_gate",
         "status": "pass"
       },
       {
-        "line": 1096,
+        "line": 1120,
         "raw": "../runs/ctt_base_context_obs_learned_dominance_train_to_test/table",
         "resolved": "runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex",
         "run_dir": "runs/ctt_base_context_obs_learned_dominance_train_to_test",
         "status": "pass"
       },
       {
-        "line": 1120,
+        "line": 1144,
         "raw": "../runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test/table.tex",
         "resolved": "runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test/table.tex",
         "run_dir": "runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test",
         "status": "pass"
       },
       {
-        "line": 1227,
+        "line": 1251,
         "raw": "../runs/paper_ctt_audit/table",
         "resolved": "runs/paper_ctt_audit/table.tex",
         "run_dir": "runs/paper_ctt_audit",
@@ -300,6 +307,7 @@
       "runs/ctt_learned_dominance_val_to_test",
       "runs/ctt_outcome_acceptance_gate",
       "runs/ctt_residual_smoke_proxy",
+      "runs/ctt_selector_diagnostic_sweep",
       "runs/ctt_test_rollout_comparison",
       "runs/ctt_val_proxy_comparison",
       "runs/ctt_val_rollout_comparison",
@@ -410,6 +418,12 @@
       "path": "scripts/calibrate_dominance.py",
       "status": "pass"
     },
+    {
+      "detail": "Selector diagnostic sweep summary.",
+      "name": "selector_diagnostic_sweep",
+      "path": "scripts/build_selector_diagnostic_sweep.py",
+      "status": "pass"
+    },
     {
       "detail": "Theory section included by paper.",
       "name": "theory_tex",
@@ -726,6 +740,15 @@
       "run_dir": "runs/ctt_residual_smoke_proxy",
       "status": "pass"
     },
+    {
+      "markdown_report_policy": "warning_only_consolidated_readme",
+      "missing_advisor_contract": [
+        "report.md"
+      ],
+      "missing_required": [],
+      "run_dir": "runs/ctt_selector_diagnostic_sweep",
+      "status": "pass"
+    },
     {
       "markdown_report_policy": "warning_only_consolidated_readme",
       "missing_advisor_contract": [
@@ -776,9 +799,9 @@
   "summary": {
     "num_failures": 0,
     "num_forbidden_matches": 0,
-    "num_paper_inputs": 38,
-    "num_run_dirs_in_paper": 33,
-    "num_warnings": 33,
+    "num_paper_inputs": 39,
+    "num_run_dirs_in_paper": 34,
+    "num_warnings": 34,
     "status": "pass"
   }
 }

workspace/runs/paper_ctt_audit/metrics.json

@@ -6,7 +6,7 @@
   "forbidden_patterns": [],
   "paper": "latex/main.tex",
   "paper_inputs": {
-    "num_inputs": 38,
+    "num_inputs": 39,
     "num_missing": 0,
     "rows": [
       {
@@ -243,28 +243,35 @@
         "status": "pass"
       },
       {
-        "line": 1065,
+        "line": 1067,
+        "raw": "../runs/ctt_selector_diagnostic_sweep/table",
+        "resolved": "runs/ctt_selector_diagnostic_sweep/table.tex",
+        "run_dir": "runs/ctt_selector_diagnostic_sweep",
+        "status": "pass"
+      },
+      {
+        "line": 1089,
         "raw": "../runs/ctt_outcome_acceptance_gate/table",
         "resolved": "runs/ctt_outcome_acceptance_gate/table.tex",
         "run_dir": "runs/ctt_outcome_acceptance_gate",
         "status": "pass"
       },
       {
-        "line": 1096,
+        "line": 1120,
         "raw": "../runs/ctt_base_context_obs_learned_dominance_train_to_test/table",
         "resolved": "runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex",
         "run_dir": "runs/ctt_base_context_obs_learned_dominance_train_to_test",
         "status": "pass"
       },
       {
-        "line": 1120,
+        "line": 1144,
         "raw": "../runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test/table.tex",
         "resolved": "runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test/table.tex",
         "run_dir": "runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test",
         "status": "pass"
       },
       {
-        "line": 1227,
+        "line": 1251,
         "raw": "../runs/paper_ctt_audit/table",
         "resolved": "runs/paper_ctt_audit/table.tex",
         "run_dir": "runs/paper_ctt_audit",
@@ -300,6 +307,7 @@
       "runs/ctt_learned_dominance_val_to_test",
       "runs/ctt_outcome_acceptance_gate",
       "runs/ctt_residual_smoke_proxy",
+      "runs/ctt_selector_diagnostic_sweep",
       "runs/ctt_test_rollout_comparison",
       "runs/ctt_val_proxy_comparison",
       "runs/ctt_val_rollout_comparison",
@@ -410,6 +418,12 @@
       "path": "scripts/calibrate_dominance.py",
       "status": "pass"
     },
+    {
+      "detail": "Selector diagnostic sweep summary.",
+      "name": "selector_diagnostic_sweep",
+      "path": "scripts/build_selector_diagnostic_sweep.py",
+      "status": "pass"
+    },
     {
       "detail": "Theory section included by paper.",
       "name": "theory_tex",
@@ -726,6 +740,15 @@
       "run_dir": "runs/ctt_residual_smoke_proxy",
       "status": "pass"
     },
+    {
+      "markdown_report_policy": "warning_only_consolidated_readme",
+      "missing_advisor_contract": [
+        "report.md"
+      ],
+      "missing_required": [],
+      "run_dir": "runs/ctt_selector_diagnostic_sweep",
+      "status": "pass"
+    },
     {
       "markdown_report_policy": "warning_only_consolidated_readme",
       "missing_advisor_contract": [
@@ -776,9 +799,9 @@
   "summary": {
     "num_failures": 0,
     "num_forbidden_matches": 0,
-    "num_paper_inputs": 38,
-    "num_run_dirs_in_paper": 33,
-    "num_warnings": 33,
+    "num_paper_inputs": 39,
+    "num_run_dirs_in_paper": 34,
+    "num_warnings": 34,
     "status": "pass"
   }
 }

workspace/runs/paper_ctt_audit/table.tex

@@ -4,9 +4,9 @@
 Audit item & Count & Status \\
 \midrule
 Forbidden language matches & 0 & pass \\
-Paper inputs & 38 & pass \\
-Run dirs in paper & 33 & pass \\
-Implementation paths & 20 & pass \\
-Advisor-contract warnings & 33 & warn \\
+Paper inputs & 39 & pass \\
+Run dirs in paper & 34 & pass \\
+Implementation paths & 21 & pass \\
+Advisor-contract warnings & 34 & warn \\
 \bottomrule
 \end{tabular}