auroc_DeepSeek_cross: 0.745583
auroc_DeepSeek_other: 0.732693
auroc_DeepSeek_same: 0.598782
auroc_DeepSeek_self: 0.458008
auroc_Gemma_cross: 0.735063
auroc_Gemma_other: 0.810384
auroc_Gemma_same: 0.853828
auroc_Gemma_self: 0.890462
auroc_Llama_cross: 0.763594
auroc_Llama_other: 0.556478
auroc_Llama_same: 0.689290
auroc_Llama_self: 0.875326
auroc_Mistral_cross: 0.707239
auroc_Mistral_other: 0.651042
auroc_Mistral_same: 0.729397
auroc_Mistral_self: 0.815755
auroc_Qwen_cross: 0.685685
auroc_Qwen_other: 0.827799
auroc_Qwen_same: 0.852336
auroc_Qwen_self: 0.872396
auroc_authored_pooled_cross: 0.772311
auroc_authored_pooled_other: 0.715423
auroc_authored_pooled_same: 0.793986
auroc_authored_pooled_self: 0.866246
auroc_pooled_cross: 0.726605
auroc_pooled_other: 0.671510
auroc_pooled_same: 0.736251
auroc_pooled_self: 0.794785
authored_n_DeepSeek: 57
authored_n_Gemma: 66
authored_n_Llama: 63
authored_n_Mistral: 36
authored_n_Qwen: 74
calibration_auc_deepseek_coder_6_7b: 0.528400
calibration_auc_deepseek_r1_8b: 0.735000
calibration_auc_gemma2_2b: 0.762200
calibration_auc_gemma2_9b: 0.832200
calibration_auc_huggingface_co_fdtn_ai_Foundation_Sec_8B_Reasoning_Q8_0_GGUF_latest: 0.500400
calibration_auc_llama3_1_8b: 0.770600
calibration_auc_mistral_7b: 0.853400
calibration_auc_qwen2_5_7b: 0.804400
calibration_auc_qwen2_5_coder_7b: 0.824600
calibration_auc_zephyr_7b: 0.639000
capability_adjusted_ci_hi: 0.020057
capability_adjusted_ci_lo: -0.045874
capability_adjusted_deficit: -0.013904
capability_adjusted_n_pairs: 50
capability_adjusted_p: 0.403596
capadj_deficit_se: 0.017309
capadj_equivalent: True
capadj_tost_ci_hi: 0.014841
capadj_tost_ci_lo: -0.041940
capadj_tost_p: 0.012987
family_DeepSeek_same_ci_hi: 0.262742
family_DeepSeek_same_ci_lo: 0.128910
family_DeepSeek_same_deficit: 0.196289
family_DeepSeek_same_p: 0.000999
family_DeepSeek_same_p_holm: 0.004995
family_Gemma_same_ci_hi: 0.054675
family_Gemma_same_ci_lo: -0.015684
family_Gemma_same_deficit: 0.018500
family_Gemma_same_p: 0.478521
family_Gemma_same_p_holm: 0.687313
family_Llama_same_ci_hi: 0.217992
family_Llama_same_ci_lo: 0.089535
family_Llama_same_deficit: 0.151964
family_Llama_same_p: 0.001998
family_Llama_same_p_holm: 0.007992
family_Mistral_same_ci_hi: 0.105227
family_Mistral_same_ci_lo: -0.002865
family_Mistral_same_deficit: 0.049533
family_Mistral_same_p: 0.212787
family_Mistral_same_p_holm: 0.638362
family_Qwen_same_ci_hi: 0.014868
family_Qwen_same_ci_lo: -0.075013
family_Qwen_same_deficit: -0.026964
family_Qwen_same_p: 0.343656
family_Qwen_same_p_holm: 0.687313
h1_authored_matched_ci_hi: 0.062327
h1_authored_matched_ci_lo: -0.023918
h1_authored_matched_deficit: 0.012553
h1_authored_matched_n: 592
h1_authored_matched_p: 0.426573
h1_authored_pooled_perrow_ci_hi: 0.061080
h1_authored_pooled_perrow_ci_lo: -0.012417
h1_authored_pooled_perrow_deficit: -0.021674
h1_authored_pooled_perrow_n: 1184
h1_authored_pooled_perrow_p: 0.187812
h1_deficit_k1: -0.050911
h1_deficit_k2: 0.005390
h1_deficit_k4: 0.067905
h1_deficit_k8: 0.110197
h1_deficit_se: 0.019159
h1_equivalent: True
h1_matched_ci_hi: 0.043743
h1_matched_ci_lo: -0.031023
h1_matched_deficit: 0.011355
h1_matched_n: 960
h1_matched_p: 0.718282
h1_mde_auroc: 0.053676
h1_pooled_perrow_ci_hi: 0.048924
h1_pooled_perrow_ci_lo: -0.011461
h1_pooled_perrow_deficit: -0.009646
h1_pooled_perrow_n: 1920
h1_pooled_perrow_p: 0.234765
h1_tost_ci_hi: 0.039229
h1_tost_ci_lo: -0.023457
h1_tost_margin: 0.050000
h1_tost_p: 0.013986
h2_ci_hi: 0.319223
h2_ci_lo: -0.261703
h2_mde_rho: 0.429609
h2_n_pairs: 50
h2_p: 0.776224
h2_rho_se: 0.153345
h2_spearman_rho: 0.040631
n_calibration_rows: 1000
n_monitor_rows: 14400
n_pair_aurocs: 50
n_seeds: 3
naive_unequal_h1_other_ci_hi: 0.254248
naive_unequal_h1_other_ci_lo: 0.181876
naive_unequal_h1_other_deficit: 0.217400
naive_unequal_h1_other_n: 960
naive_unequal_h1_other_p: 0.000999
naive_unequal_h1_same_ci_hi: 0.142407
naive_unequal_h1_same_ci_lo: 0.076159
naive_unequal_h1_same_deficit: 0.110197
naive_unequal_h1_same_n: 960
naive_unequal_h1_same_p: 0.000999
naive_unequal_h1_self_ci_hi: 0.122529
naive_unequal_h1_self_ci_lo: 0.064668
naive_unequal_h1_self_deficit: 0.094125
naive_unequal_h1_self_n: 960
naive_unequal_h1_self_p: 0.000999
placebo_deficit_shrinkage_authored_minus_placebo: -0.108832
placebo_same_ci_hi: 0.150261
placebo_same_ci_lo: 0.090476
placebo_same_deficit: 0.120187
placebo_same_p: 0.000999
self_vs_other_ci_hi: -0.088120
self_vs_other_ci_lo: -0.161517
self_vs_other_deficit_other_minus_self: -0.123275
self_vs_other_p: 0.000999
solution_source_reference_rate_DeepSeek: 0.406250
solution_source_reference_rate_Gemma: 0.312500
solution_source_reference_rate_Llama: 0.343750
solution_source_reference_rate_Mistral: 0.625000
solution_source_reference_rate_Qwen: 0.229167