design: candidate_3_multi_domain_budget_revised
families_observed: llama,mistral,qwen
seeds_observed: 0,1,2,3,4,5
random_direction_floor_primary: bounded_raw_abs_auc_minus_0_5_median_iqr
llama_prompted_to_locked_signed_R_mean: -0.404593
llama_prompted_to_locked_signed_R_bootstrap_ci_low: -0.691385
llama_prompted_to_locked_signed_R_bootstrap_ci_high: -0.123179
llama_prompted_to_locked_inverted_count: 4
llama_locked_to_prompted_signed_R_mean: -0.238986
llama_locked_to_prompted_signed_R_bootstrap_ci_low: -0.668176
llama_locked_to_prompted_signed_R_bootstrap_ci_high: 0.181009
llama_locked_to_prompted_inverted_count: 3
mistral_prompted_to_locked_signed_R_mean: 0.101832
mistral_prompted_to_locked_signed_R_bootstrap_ci_low: -0.294156
mistral_prompted_to_locked_signed_R_bootstrap_ci_high: 0.524779
mistral_prompted_to_locked_inverted_count: 3
mistral_locked_to_prompted_signed_R_mean: 0.093147
mistral_locked_to_prompted_signed_R_bootstrap_ci_low: -0.316819
mistral_locked_to_prompted_signed_R_bootstrap_ci_high: 0.437088
mistral_locked_to_prompted_inverted_count: 2
qwen_prompted_to_locked_signed_R_mean: -0.315208
qwen_prompted_to_locked_signed_R_bootstrap_ci_low: -0.355227
qwen_prompted_to_locked_signed_R_bootstrap_ci_high: -0.274029
qwen_prompted_to_locked_inverted_count: 6
qwen_locked_to_prompted_signed_R_mean: -0.980490
qwen_locked_to_prompted_signed_R_bootstrap_ci_low: -0.988986
qwen_locked_to_prompted_signed_R_bootstrap_ci_high: -0.971337
qwen_locked_to_prompted_inverted_count: 6
bounded_floor_locked_raw_abs_auc_minus_0_5_median_mean: 0.069355
bounded_floor_locked_raw_abs_auc_minus_0_5_median_bootstrap_ci_low: 0.042752
bounded_floor_locked_raw_abs_auc_minus_0_5_median_bootstrap_ci_high: 0.104003
bounded_floor_prompted_raw_abs_auc_minus_0_5_median_mean: 0.107881
bounded_floor_prompted_raw_abs_auc_minus_0_5_median_bootstrap_ci_low: 0.082523
bounded_floor_prompted_raw_abs_auc_minus_0_5_median_bootstrap_ci_high: 0.136398
layer_stability_curve_rows: 184
layer_stability_neighbor_R_swing_ci_high: 0.100422
layer_stability_neighbor_R_swing_ci_low: 0.069474
layer_stability_neighbor_R_swing_max: 0.731268
layer_stability_neighbor_R_swing_mean: 0.083975
layer_stability_validation_auc_range_ci_high: 0.012762
layer_stability_validation_auc_range_ci_low: 0.002968
layer_stability_validation_auc_range_mean: 0.007651
causal_patch_cross_locked_to_prompted_flip_rate_minus_null_mean: 0.037384
causal_patch_cross_locked_to_prompted_flip_rate_minus_null_bootstrap_ci_low: -0.005787
causal_patch_cross_locked_to_prompted_flip_rate_minus_null_bootstrap_ci_high: 0.088773
causal_patch_cross_locked_to_prompted_flip_rate_minus_null_p: 0.070000
causal_patch_cross_prompted_to_locked_flip_rate_minus_null_mean: -0.017361
causal_patch_cross_prompted_to_locked_flip_rate_minus_null_bootstrap_ci_low: -0.035880
causal_patch_cross_prompted_to_locked_flip_rate_minus_null_bootstrap_ci_high: 0.003009
causal_patch_cross_prompted_to_locked_flip_rate_minus_null_p: 0.956200
position_ablation_reconciled_rows: 36
position_ablation_resolved_wave2_rows: 24
position_last_instruction_prompted_to_locked_R_mean: 0.991071
position_last_instruction_prompted_to_locked_R_bootstrap_ci_low: 0.986616
position_last_instruction_prompted_to_locked_R_bootstrap_ci_high: 0.994979
position_last_instruction_prompted_to_locked_signed_R_mean: -0.991071
position_last_instruction_prompted_to_locked_signed_R_bootstrap_ci_low: -0.994985
position_last_instruction_prompted_to_locked_signed_R_bootstrap_ci_high: -0.986616
position_last_instruction_locked_to_prompted_R_mean: 1.000073
position_last_instruction_locked_to_prompted_R_bootstrap_ci_low: 0.998622
position_last_instruction_locked_to_prompted_R_bootstrap_ci_high: 1.001604
position_last_instruction_locked_to_prompted_signed_R_mean: -1.000073
position_last_instruction_locked_to_prompted_signed_R_bootstrap_ci_low: -1.001604
position_last_instruction_locked_to_prompted_signed_R_bootstrap_ci_high: -0.998623
position_mean_last8_prompted_to_locked_R_mean: 0.141451
position_mean_last8_prompted_to_locked_R_bootstrap_ci_low: 0.091721
position_mean_last8_prompted_to_locked_R_bootstrap_ci_high: 0.192470
position_mean_last8_prompted_to_locked_signed_R_mean: -0.096377
position_mean_last8_prompted_to_locked_signed_R_bootstrap_ci_low: -0.172991
position_mean_last8_prompted_to_locked_signed_R_bootstrap_ci_high: -0.017481
position_mean_last8_locked_to_prompted_R_mean: 0.425158
position_mean_last8_locked_to_prompted_R_bootstrap_ci_low: 0.235799
position_mean_last8_locked_to_prompted_R_bootstrap_ci_high: 0.629757
position_mean_last8_locked_to_prompted_signed_R_mean: -0.315788
position_mean_last8_locked_to_prompted_signed_R_bootstrap_ci_low: -0.577078
position_mean_last8_locked_to_prompted_signed_R_bootstrap_ci_high: -0.067019
stats_full_family_n_tests: 60
best_layer_mean: 4.666667
best_layer_bootstrap_ci_low: 2.888889
best_layer_bootstrap_ci_high: 7.500000
causal_patch_flip_rate_mean: 0.019167
causal_patch_flip_rate_bootstrap_ci_low: 0.008796
causal_patch_flip_rate_bootstrap_ci_high: 0.036481
causal_patch_n_mean: 600.000000
causal_patch_n_bootstrap_ci_low: 600.000000
causal_patch_n_bootstrap_ci_high: 600.000000
heldout_domain_prompted_to_locked_auc_mean: 0.408686
heldout_domain_prompted_to_locked_auc_bootstrap_ci_low: 0.302984
heldout_domain_prompted_to_locked_auc_bootstrap_ci_high: 0.518426
locked_in_dist_auc_mean: 0.998598
locked_in_dist_auc_bootstrap_ci_low: 0.996941
locked_in_dist_auc_bootstrap_ci_high: 0.999756
locked_to_prompted_R_mean: 0.627057
locked_to_prompted_R_bootstrap_ci_low: 0.462782
locked_to_prompted_R_bootstrap_ci_high: 0.786677
locked_to_prompted_auc_mean: 0.312382
locked_to_prompted_auc_bootstrap_ci_low: 0.178791
locked_to_prompted_auc_bootstrap_ci_high: 0.461478
mean_diff_cosine_mean: -0.029744
mean_diff_cosine_bootstrap_ci_low: -0.089104
mean_diff_cosine_bootstrap_ci_high: 0.032680
neutral_lora_R_mean: 0.999364
neutral_lora_R_bootstrap_ci_low: 0.998740
neutral_lora_R_bootstrap_ci_high: 0.999841
neutral_lora_auc_mean: 0.999601
neutral_lora_auc_bootstrap_ci_low: 0.999255
neutral_lora_auc_bootstrap_ci_high: 0.999897
oracle_locked_auc_mean: 0.998598
oracle_locked_auc_bootstrap_ci_low: 0.996941
oracle_locked_auc_bootstrap_ci_high: 0.999756
prompted_in_dist_auc_mean: 0.999918
prompted_in_dist_auc_bootstrap_ci_low: 0.999802
prompted_in_dist_auc_bootstrap_ci_high: 0.999991
prompted_to_locked_R_mean: 0.383326
prompted_to_locked_R_bootstrap_ci_low: 0.268920
prompted_to_locked_R_bootstrap_ci_high: 0.511524
prompted_to_locked_auc_mean: 0.397004
prompted_to_locked_auc_bootstrap_ci_low: 0.301196
prompted_to_locked_auc_bootstrap_ci_high: 0.492341
prompted_val_auc_mean: 0.999993
prompted_val_auc_bootstrap_ci_low: 0.999987
prompted_val_auc_bootstrap_ci_high: 0.999997
random_direction_R_max_mean: 47.101746
random_direction_R_max_bootstrap_ci_low: 14.311757
random_direction_R_max_bootstrap_ci_high: 103.436823
random_direction_R_mean_mean: -4.339975
random_direction_R_mean_bootstrap_ci_low: -13.750742
random_direction_R_mean_bootstrap_ci_high: 1.027813
random_direction_count_mean: 50.000000
random_direction_count_bootstrap_ci_low: 50.000000
random_direction_count_bootstrap_ci_high: 50.000000
seed_mean: 2.500000
seed_bootstrap_ci_low: 1.722222
seed_bootstrap_ci_high: 3.333333
llama_best_layer_mean: 6.833333
llama_causal_patch_flip_rate_mean: 0.036389
llama_causal_patch_n_mean: 600.000000
llama_heldout_domain_prompted_to_locked_auc_mean: 0.318960
llama_locked_in_dist_auc_mean: 0.996712
llama_locked_to_prompted_R_mean: 0.503967
llama_locked_to_prompted_auc_mean: 0.380261
llama_mean_diff_cosine_mean: -0.021980
llama_neutral_lora_R_mean: 0.998596
llama_neutral_lora_auc_mean: 0.999095
llama_oracle_locked_auc_mean: 0.996712
llama_prompted_in_dist_auc_mean: 0.999797
llama_prompted_to_locked_R_mean: 0.423772
llama_prompted_to_locked_auc_mean: 0.297696
llama_prompted_val_auc_mean: 0.999994
llama_random_direction_R_max_mean: 17.118116
llama_random_direction_R_mean_mean: -13.318810
llama_random_direction_count_mean: 50.000000
llama_seed_mean: 2.500000
mistral_best_layer_mean: 5.000000
mistral_causal_patch_flip_rate_mean: 0.016389
mistral_causal_patch_n_mean: 600.000000
mistral_heldout_domain_prompted_to_locked_auc_mean: 0.585622
mistral_locked_in_dist_auc_mean: 0.999648
mistral_locked_to_prompted_R_mean: 0.396712
mistral_locked_to_prompted_auc_mean: 0.546579
mistral_mean_diff_cosine_mean: 0.076262
mistral_neutral_lora_R_mean: 0.999497
mistral_neutral_lora_auc_mean: 0.999722
mistral_oracle_locked_auc_mean: 0.999648
mistral_prompted_in_dist_auc_mean: 0.999974
mistral_prompted_to_locked_R_mean: 0.410999
mistral_prompted_to_locked_auc_mean: 0.550917
mistral_prompted_val_auc_mean: 0.999992
mistral_random_direction_R_max_mean: 115.270291
mistral_random_direction_R_mean_mean: 1.408406
mistral_random_direction_count_mean: 50.000000
mistral_seed_mean: 2.500000
qwen_best_layer_mean: 2.166667
qwen_causal_patch_flip_rate_mean: 0.004722
qwen_causal_patch_n_mean: 600.000000
qwen_heldout_domain_prompted_to_locked_auc_mean: 0.321478
qwen_locked_in_dist_auc_mean: 0.999433
qwen_locked_to_prompted_R_mean: 0.980490
qwen_locked_to_prompted_auc_mean: 0.010307
qwen_mean_diff_cosine_mean: -0.143514
qwen_neutral_lora_R_mean: 1.000000
qwen_neutral_lora_auc_mean: 0.999985
qwen_oracle_locked_auc_mean: 0.999433
qwen_prompted_in_dist_auc_mean: 0.999985
qwen_prompted_to_locked_R_mean: 0.315208
qwen_prompted_to_locked_auc_mean: 0.342401
qwen_prompted_val_auc_mean: 0.999992
qwen_random_direction_R_max_mean: 8.916829
qwen_random_direction_R_mean_mean: -1.109521
qwen_random_direction_count_mean: 50.000000
qwen_seed_mean: 2.500000
causal_patch_positive_control_known_minus_null_mean: 0.440509
causal_patch_positive_control_known_minus_null_p: 0.000000
stats_full_family_n_tests_with_wave5: 60
wave6_signed_retention_cells_rows: 6
wave6_layer_stability_saturation_rows: 6
wave6_position_retention_bootstrap_rows: 12
causal_patch_matched_rows: 384
password_masked_oracle_rows: 12
stats_full_family_n_tests_with_wave6: 60
wave6b_matched_p2l_flip_minus_null_mean: 0.060648
wave6b_matched_p2l_flip_minus_null_p: 0.000000
wave6b_matched_l2p_flip_minus_null_mean: 0.037500
wave6b_matched_l2p_flip_minus_null_p: 0.000780
llama_masked_oracle_auc_mean: 0.991590
llama_masked_oracle_above_chance_p: 0.015625
llama_masked_minus_unmasked_oracle_mean: -0.008410
llama_masked_minus_unmasked_oracle_p: 0.062500
qwen_masked_oracle_auc_mean: 0.980640
qwen_masked_oracle_above_chance_p: 0.015625
qwen_masked_minus_unmasked_oracle_mean: -0.019325
qwen_masked_minus_unmasked_oracle_p: 0.031250