n_finetunes_ok: 62.000000 n_seeds_per_arm: 8.000000 ceiling_acc_mean: 0.925500 weak_floor_p0.10: 0.900000 weak_floor_p0.20: 0.800000 exploitability_accuracy: 0.712000 organic_weak_train_accuracy: 0.500000 organic_weak_val_accuracy: 0.500000 strong_zero_shot_val_accuracy: 0.500000 organic_weak_val_pred_nonentailment_frac: 1.000000 strong_zero_shot_val_pred_nonentailment_frac: 1.000000 gpu_recorded_hours: 2.072014 manip_random_uniform_p0.20_spread: 0.019781 manip_systematic_p0.20_spread: 0.163760 manip_genre_blind_p0.20_spread: 0.014385 systematic_local_error_rate_mean_p0.10: 0.184375 blind_per_genre_local_error_rate_mean_p0.10: 0.124687 blind_partition_error_rate_mean_p0.10: 0.166667 mean_acc_random_uniform_p0.10: 0.920250 pgr_random_uniform_p0.10: 0.794118 mean_acc_systematic_3of5_genres_p0.10: 0.926750 pgr_systematic_3of5_genres_p0.10: 1.049020 mean_acc_genre_blind_clustered_p0.10: 0.918000 pgr_genre_blind_clustered_p0.10: 0.705882 systematic_local_error_rate_mean_p0.20: 0.353438 blind_per_genre_local_error_rate_mean_p0.20: 0.224062 blind_partition_error_rate_mean_p0.20: 0.333333 mean_acc_random_uniform_p0.20: 0.917000 pgr_random_uniform_p0.20: 0.932271 mean_acc_systematic_3of5_genres_p0.20: 0.916000 pgr_systematic_3of5_genres_p0.20: 0.924303 mean_acc_genre_blind_clustered_p0.20: 0.915750 pgr_genre_blind_clustered_p0.20: 0.922311 positive_control_mean_acc_p0.35: 0.865000 positive_control_pgr_p0.35: 0.780399 positive_control_weak_floor_p0.35: 0.650000 positive_control_mean_acc_p0.45: 0.642333 positive_control_pgr_p0.45: 0.245894 positive_control_weak_floor_p0.45: 0.550000 positive_control_low_rate_mean_pgr: 0.863194 positive_control_high_rate_mean_pgr: 0.513147 positive_control_sensitivity_drop: 1.000000 gap_systematic_minus_blind_p0.10: 0.008750 gap_systematic_minus_blind_p0.20: 0.000250 gap_systematic_minus_random_p0.10: 0.006500 gap_systematic_minus_random_p0.20: -0.001000 systematic_per_genre_delta_corrupted_minus_clean_mean_p0.10: -0.000208 systematic_per_genre_delta_corrupted_minus_clean_mean_p0.20: -0.004062 clean_label_per_genre_acc_fiction: 0.923125 clean_label_per_genre_acc_government: 0.933750 clean_label_per_genre_acc_slate: 0.918750 clean_label_per_genre_acc_telephone: 0.913750 clean_label_per_genre_acc_travel: 0.938125 final_loss_min: 0.001752 final_loss_max: 0.273028 final_loss_ratio: 155.835182 final_loss_accuracy_pearson: -0.161554 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_statistic: 0.008750 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_p: 0.046875 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_holm_p: 0.187500 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_ci_lo: 0.002750 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_ci_hi: 0.014378 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.10_statistic: 0.343137 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.10_p: 0.046875 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.10_holm_p: 0.187500 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.10_ci_lo: 0.107843 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.10_ci_hi: 0.563848 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_statistic: 0.008750 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_p: 0.007199 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_ci_lo: 0.002500 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.10_ci_hi: 0.015000 stats_item_systematic_minus_genre-blind_pgr_difference_p0.10_statistic: 0.343137 stats_item_systematic_minus_genre-blind_pgr_difference_p0.10_p: 0.007199 stats_item_systematic_minus_genre-blind_pgr_difference_p0.10_ci_lo: 0.098039 stats_item_systematic_minus_genre-blind_pgr_difference_p0.10_ci_hi: 0.588235 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.10_statistic: 0.006500 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.10_p: 0.062500 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.10_holm_p: 0.187500 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.10_ci_lo: 0.001375 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.10_ci_hi: 0.012000 stats_seed_systematic_minus_random_pgr_difference_p0.10_statistic: 0.254902 stats_seed_systematic_minus_random_pgr_difference_p0.10_p: 0.062500 stats_seed_systematic_minus_random_pgr_difference_p0.10_holm_p: 0.187500 stats_seed_systematic_minus_random_pgr_difference_p0.10_ci_lo: 0.053922 stats_seed_systematic_minus_random_pgr_difference_p0.10_ci_hi: 0.470588 stats_item_systematic_minus_random_raw_accuracy_difference_p0.10_statistic: 0.006500 stats_item_systematic_minus_random_raw_accuracy_difference_p0.10_p: 0.038996 stats_item_systematic_minus_random_raw_accuracy_difference_p0.10_ci_lo: 0.000500 stats_item_systematic_minus_random_raw_accuracy_difference_p0.10_ci_hi: 0.012625 stats_item_systematic_minus_random_pgr_difference_p0.10_statistic: 0.254902 stats_item_systematic_minus_random_pgr_difference_p0.10_p: 0.038996 stats_item_systematic_minus_random_pgr_difference_p0.10_ci_lo: 0.019608 stats_item_systematic_minus_random_pgr_difference_p0.10_ci_hi: 0.495098 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_statistic: 0.000250 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_p: 0.937500 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_holm_p: 1.000000 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_ci_lo: -0.005500 stats_seed_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_ci_hi: 0.006875 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.20_statistic: 0.001992 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.20_p: 0.937500 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.20_holm_p: 1.000000 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.20_ci_lo: -0.043825 stats_seed_systematic_minus_genre-blind_pgr_difference_p0.20_ci_hi: 0.054781 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_statistic: 0.000250 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_p: 0.955704 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_ci_lo: -0.003875 stats_item_systematic_minus_genre-blind_raw_accuracy_difference_p0.20_ci_hi: 0.004375 stats_item_systematic_minus_genre-blind_pgr_difference_p0.20_statistic: 0.001992 stats_item_systematic_minus_genre-blind_pgr_difference_p0.20_p: 0.955704 stats_item_systematic_minus_genre-blind_pgr_difference_p0.20_ci_lo: -0.030876 stats_item_systematic_minus_genre-blind_pgr_difference_p0.20_ci_hi: 0.034861 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.20_statistic: -0.001000 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.20_p: 0.726562 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.20_holm_p: 1.000000 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.20_ci_lo: -0.005125 stats_seed_systematic_minus_random_raw_accuracy_difference_p0.20_ci_hi: 0.003000 stats_seed_systematic_minus_random_pgr_difference_p0.20_statistic: -0.007968 stats_seed_systematic_minus_random_pgr_difference_p0.20_p: 0.726562 stats_seed_systematic_minus_random_pgr_difference_p0.20_holm_p: 1.000000 stats_seed_systematic_minus_random_pgr_difference_p0.20_ci_lo: -0.040837 stats_seed_systematic_minus_random_pgr_difference_p0.20_ci_hi: 0.023904 stats_item_systematic_minus_random_raw_accuracy_difference_p0.20_statistic: -0.001000 stats_item_systematic_minus_random_raw_accuracy_difference_p0.20_p: 0.721628 stats_item_systematic_minus_random_raw_accuracy_difference_p0.20_ci_lo: -0.005875 stats_item_systematic_minus_random_raw_accuracy_difference_p0.20_ci_hi: 0.003750 stats_item_systematic_minus_random_pgr_difference_p0.20_statistic: -0.007968 stats_item_systematic_minus_random_pgr_difference_p0.20_p: 0.721628 stats_item_systematic_minus_random_pgr_difference_p0.20_ci_lo: -0.046813 stats_item_systematic_minus_random_pgr_difference_p0.20_ci_hi: 0.029880 stats_seed_positive_control_pgr_p0.45_minus_random_p0.20_pgr_difference_p0.45_statistic: -0.659816 stats_seed_positive_control_pgr_p0.45_minus_random_p0.20_pgr_difference_p0.45_p: 0.250000 stats_seed_positive_control_pgr_p0.45_minus_random_p0.20_pgr_difference_p0.45_ci_lo: -1.046806 stats_seed_positive_control_pgr_p0.45_minus_random_p0.20_pgr_difference_p0.45_ci_hi: -0.133612