bootstrap_samples: 2000 bootstrap_seed: 20260717 probe_forget_layer7_pre_frozen_accuracy: 0.297145 probe_forget_layer7_graddiff_layerA_seed0_frozen_accuracy: 0.255366 probe_forget_layer7_graddiff_layerA_seed0_fresh_accuracy: 0.283494 probe_forget_layer7_graddiff_layerA_seed1_frozen_accuracy: 0.262810 probe_forget_layer7_graddiff_layerA_seed1_fresh_accuracy: 0.270481 probe_forget_layer7_rmu_layerA_seed0_frozen_accuracy: 0.242757 probe_forget_layer7_rmu_layerA_seed0_fresh_accuracy: 0.246011 probe_forget_layer7_rmu_layerA_seed1_frozen_accuracy: 0.256330 probe_forget_layer7_rmu_layerA_seed1_fresh_accuracy: 0.261116 probe_forget_layer7_rmu_layerA_seed2_frozen_accuracy: 0.253886 probe_forget_layer7_rmu_layerA_seed2_fresh_accuracy: 0.275128 probe_forget_layer7_rmu_layerB_seed0_frozen_accuracy: 0.297145 probe_forget_layer7_rmu_layerB_seed0_fresh_accuracy: 0.297145 probe_forget_layer7_rmu_layerB_seed1_frozen_accuracy: 0.297145 probe_forget_layer7_rmu_layerB_seed1_fresh_accuracy: 0.297145 probe_forget_layer7_rmu_layerB_seed2_frozen_accuracy: 0.297145 probe_forget_layer7_rmu_layerB_seed2_fresh_accuracy: 0.297145 probe_forget_layer7_instructed_denial_frozen_accuracy: 0.268255 probe_forget_layer7_instructed_denial_fresh_accuracy: 0.298886 probe_forget_layer7_graddiff_layerA_frozen_accuracy: 0.259088 probe_forget_layer7_graddiff_layerA_fresh_accuracy: 0.276987 probe_forget_layer7_rmu_layerA_frozen_accuracy: 0.250991 probe_forget_layer7_rmu_layerA_fresh_accuracy: 0.260752 probe_forget_layer15_pre_frozen_accuracy: 0.429156 probe_forget_layer15_graddiff_layerA_seed0_frozen_accuracy: 0.244187 probe_forget_layer15_graddiff_layerA_seed0_fresh_accuracy: 0.277252 probe_forget_layer15_graddiff_layerA_seed1_frozen_accuracy: 0.266330 probe_forget_layer15_graddiff_layerA_seed1_fresh_accuracy: 0.279434 probe_forget_layer15_rmu_layerA_seed0_frozen_accuracy: 0.249139 probe_forget_layer15_rmu_layerA_seed0_fresh_accuracy: 0.242451 probe_forget_layer15_rmu_layerA_seed1_frozen_accuracy: 0.283932 probe_forget_layer15_rmu_layerA_seed1_fresh_accuracy: 0.283948 probe_forget_layer15_rmu_layerA_seed2_frozen_accuracy: 0.307230 probe_forget_layer15_rmu_layerA_seed2_fresh_accuracy: 0.299455 probe_forget_layer15_rmu_layerB_seed0_frozen_accuracy: 0.268606 probe_forget_layer15_rmu_layerB_seed0_fresh_accuracy: 0.333325 probe_forget_layer15_rmu_layerB_seed1_frozen_accuracy: 0.264246 probe_forget_layer15_rmu_layerB_seed1_fresh_accuracy: 0.345475 probe_forget_layer15_rmu_layerB_seed2_frozen_accuracy: 0.277188 probe_forget_layer15_rmu_layerB_seed2_fresh_accuracy: 0.383271 probe_forget_layer15_instructed_denial_frozen_accuracy: 0.345891 probe_forget_layer15_instructed_denial_fresh_accuracy: 0.416680 probe_forget_layer15_rmu_layerB_frozen_accuracy: 0.270013 probe_forget_layer15_rmu_layerB_fresh_accuracy: 0.354024 probe_retain_layer7_pre_frozen_accuracy: 0.269730 probe_retain_layer7_graddiff_layerA_seed0_frozen_accuracy: 0.240368 probe_retain_layer7_graddiff_layerA_seed0_fresh_accuracy: 0.276821 probe_retain_layer7_graddiff_layerA_seed1_frozen_accuracy: 0.273580 probe_retain_layer7_graddiff_layerA_seed1_fresh_accuracy: 0.264570 probe_retain_layer7_rmu_layerA_seed0_frozen_accuracy: 0.264694 probe_retain_layer7_rmu_layerA_seed0_fresh_accuracy: 0.264935 probe_retain_layer7_rmu_layerA_seed1_frozen_accuracy: 0.270933 probe_retain_layer7_rmu_layerA_seed1_fresh_accuracy: 0.265999 probe_retain_layer7_rmu_layerA_seed2_frozen_accuracy: 0.274734 probe_retain_layer7_rmu_layerA_seed2_fresh_accuracy: 0.270577 probe_retain_layer7_rmu_layerB_seed0_frozen_accuracy: 0.269730 probe_retain_layer7_rmu_layerB_seed0_fresh_accuracy: 0.269730 probe_retain_layer7_rmu_layerB_seed1_frozen_accuracy: 0.269730 probe_retain_layer7_rmu_layerB_seed1_fresh_accuracy: 0.269730 probe_retain_layer7_rmu_layerB_seed2_frozen_accuracy: 0.269730 probe_retain_layer7_rmu_layerB_seed2_fresh_accuracy: 0.269730 probe_retain_layer7_instructed_denial_frozen_accuracy: 0.252588 probe_retain_layer7_instructed_denial_fresh_accuracy: 0.262149 probe_retain_layer7_graddiff_layerA_frozen_accuracy: 0.256974 probe_retain_layer7_graddiff_layerA_fresh_accuracy: 0.270695 probe_retain_layer7_rmu_layerA_frozen_accuracy: 0.270121 probe_retain_layer7_rmu_layerA_fresh_accuracy: 0.267170 probe_retain_layer15_pre_frozen_accuracy: 0.388578 probe_retain_layer15_graddiff_layerA_seed0_frozen_accuracy: 0.284122 probe_retain_layer15_graddiff_layerA_seed0_fresh_accuracy: 0.343575 probe_retain_layer15_graddiff_layerA_seed1_frozen_accuracy: 0.350980 probe_retain_layer15_graddiff_layerA_seed1_fresh_accuracy: 0.374609 probe_retain_layer15_rmu_layerA_seed0_frozen_accuracy: 0.317273 probe_retain_layer15_rmu_layerA_seed0_fresh_accuracy: 0.337663 probe_retain_layer15_rmu_layerA_seed1_frozen_accuracy: 0.354902 probe_retain_layer15_rmu_layerA_seed1_fresh_accuracy: 0.366935 probe_retain_layer15_rmu_layerA_seed2_frozen_accuracy: 0.364410 probe_retain_layer15_rmu_layerA_seed2_fresh_accuracy: 0.375666 probe_retain_layer15_rmu_layerB_seed0_frozen_accuracy: 0.314787 probe_retain_layer15_rmu_layerB_seed0_fresh_accuracy: 0.357910 probe_retain_layer15_rmu_layerB_seed1_frozen_accuracy: 0.327091 probe_retain_layer15_rmu_layerB_seed1_fresh_accuracy: 0.351600 probe_retain_layer15_rmu_layerB_seed2_frozen_accuracy: 0.307663 probe_retain_layer15_rmu_layerB_seed2_fresh_accuracy: 0.368472 probe_retain_layer15_instructed_denial_frozen_accuracy: 0.339918 probe_retain_layer15_instructed_denial_fresh_accuracy: 0.360721 probe_retain_layer15_rmu_layerB_frozen_accuracy: 0.316513 probe_retain_layer15_rmu_layerB_fresh_accuracy: 0.359327 claim_template_count: 3 claim_base_pre_forget_tpl1_rate: 0.877264 claim_base_pre_forget_tpl2_rate: 0.002012 claim_base_pre_forget_tpl3_rate: 0.273642 claim_base_pre_forget_overall_rate: 0.384306 claim_base_pre_retain_tpl1_rate: 0.982897 claim_base_pre_retain_tpl2_rate: 0.009054 claim_base_pre_retain_tpl3_rate: 0.087525 claim_base_pre_retain_overall_rate: 0.359826 claim_graddiff_layerA_seed0_forget_tpl1_rate: 0.114688 claim_graddiff_layerA_seed0_retain_tpl1_rate: 0.094567 did_graddiff_layerA_seed0_tpl1: -0.084270 did_graddiff_layerA_seed0_tpl1_ci_lo: -0.131086 did_graddiff_layerA_seed0_tpl1_ci_hi: -0.039326 did_graddiff_layerA_seed0_tpl1_n_pairs: 534 claim_graddiff_layerA_seed0_forget_tpl2_rate: 0.057344 claim_graddiff_layerA_seed0_retain_tpl2_rate: 0.871227 did_graddiff_layerA_seed0_tpl2: 0.801498 did_graddiff_layerA_seed0_tpl2_ci_lo: 0.767790 did_graddiff_layerA_seed0_tpl2_ci_hi: 0.838951 did_graddiff_layerA_seed0_tpl2_n_pairs: 534 claim_graddiff_layerA_seed0_forget_tpl3_rate: 0.030181 claim_graddiff_layerA_seed0_retain_tpl3_rate: 0.289738 did_graddiff_layerA_seed0_tpl3: 0.428839 did_graddiff_layerA_seed0_tpl3_ci_lo: 0.367041 did_graddiff_layerA_seed0_tpl3_ci_hi: 0.486938 did_graddiff_layerA_seed0_tpl3_n_pairs: 534 claim_graddiff_layerA_seed0_forget_overall_rate: 0.067404 claim_graddiff_layerA_seed0_retain_overall_rate: 0.418511 did_graddiff_layerA_seed0_overall: 0.382022 did_graddiff_layerA_seed0_overall_ci_lo: 0.352669 did_graddiff_layerA_seed0_overall_ci_hi: 0.413233 did_graddiff_layerA_seed0_overall_n_pairs: 534 claim_graddiff_layerA_seed1_forget_tpl1_rate: 0.169014 claim_graddiff_layerA_seed1_retain_tpl1_rate: 0.744467 did_graddiff_layerA_seed1_tpl1: 0.481273 did_graddiff_layerA_seed1_tpl1_ci_lo: 0.423221 did_graddiff_layerA_seed1_tpl1_ci_hi: 0.537453 did_graddiff_layerA_seed1_tpl1_n_pairs: 534 claim_graddiff_layerA_seed1_forget_tpl2_rate: 0.032193 claim_graddiff_layerA_seed1_retain_tpl2_rate: 0.359155 did_graddiff_layerA_seed1_tpl2: 0.337079 did_graddiff_layerA_seed1_tpl2_ci_lo: 0.292135 did_graddiff_layerA_seed1_tpl2_ci_hi: 0.382022 did_graddiff_layerA_seed1_tpl2_n_pairs: 534 claim_graddiff_layerA_seed1_forget_tpl3_rate: 0.033199 claim_graddiff_layerA_seed1_retain_tpl3_rate: 0.005030 did_graddiff_layerA_seed1_tpl3: 0.146067 did_graddiff_layerA_seed1_tpl3_ci_lo: 0.097378 did_graddiff_layerA_seed1_tpl3_ci_hi: 0.192884 did_graddiff_layerA_seed1_tpl3_n_pairs: 534 claim_graddiff_layerA_seed1_forget_overall_rate: 0.078135 claim_graddiff_layerA_seed1_retain_overall_rate: 0.369551 did_graddiff_layerA_seed1_overall: 0.321473 did_graddiff_layerA_seed1_overall_ci_lo: 0.290886 did_graddiff_layerA_seed1_overall_ci_hi: 0.351451 did_graddiff_layerA_seed1_overall_n_pairs: 534 claim_rmu_layerA_seed0_forget_tpl1_rate: 0.241449 claim_rmu_layerA_seed0_retain_tpl1_rate: 0.637827 did_rmu_layerA_seed0_tpl1: 0.290262 did_rmu_layerA_seed0_tpl1_ci_lo: 0.230337 did_rmu_layerA_seed0_tpl1_ci_hi: 0.350187 did_rmu_layerA_seed0_tpl1_n_pairs: 534 claim_rmu_layerA_seed0_forget_tpl2_rate: 0.186117 claim_rmu_layerA_seed0_retain_tpl2_rate: 0.003018 did_rmu_layerA_seed0_tpl2: -0.189139 did_rmu_layerA_seed0_tpl2_ci_lo: -0.222846 did_rmu_layerA_seed0_tpl2_ci_hi: -0.155431 did_rmu_layerA_seed0_tpl2_n_pairs: 534 claim_rmu_layerA_seed0_forget_tpl3_rate: 0.209256 claim_rmu_layerA_seed0_retain_tpl3_rate: 0.367203 did_rmu_layerA_seed0_tpl3: 0.355805 did_rmu_layerA_seed0_tpl3_ci_lo: 0.290262 did_rmu_layerA_seed0_tpl3_ci_hi: 0.419522 did_rmu_layerA_seed0_tpl3_n_pairs: 534 claim_rmu_layerA_seed0_forget_overall_rate: 0.212274 claim_rmu_layerA_seed0_retain_overall_rate: 0.336016 did_rmu_layerA_seed0_overall: 0.152310 did_rmu_layerA_seed0_overall_ci_lo: 0.118602 did_rmu_layerA_seed0_overall_ci_hi: 0.185393 did_rmu_layerA_seed0_overall_n_pairs: 534 claim_rmu_layerA_seed1_forget_tpl1_rate: 0.484909 claim_rmu_layerA_seed1_retain_tpl1_rate: 0.621730 did_rmu_layerA_seed1_tpl1: -0.031835 did_rmu_layerA_seed1_tpl1_ci_lo: -0.088015 did_rmu_layerA_seed1_tpl1_ci_hi: 0.026217 did_rmu_layerA_seed1_tpl1_n_pairs: 534 claim_rmu_layerA_seed1_forget_tpl2_rate: 0.371227 claim_rmu_layerA_seed1_retain_tpl2_rate: 0.051308 did_rmu_layerA_seed1_tpl2: -0.340824 did_rmu_layerA_seed1_tpl2_ci_lo: -0.387640 did_rmu_layerA_seed1_tpl2_ci_hi: -0.295833 did_rmu_layerA_seed1_tpl2_n_pairs: 534 claim_rmu_layerA_seed1_forget_tpl3_rate: 0.438632 claim_rmu_layerA_seed1_retain_tpl3_rate: 0.019115 did_rmu_layerA_seed1_tpl3: -0.273408 did_rmu_layerA_seed1_tpl3_ci_lo: -0.338951 did_rmu_layerA_seed1_tpl3_ci_hi: -0.204120 did_rmu_layerA_seed1_tpl3_n_pairs: 534 claim_rmu_layerA_seed1_forget_overall_rate: 0.431590 claim_rmu_layerA_seed1_retain_overall_rate: 0.230718 did_rmu_layerA_seed1_overall: -0.215356 did_rmu_layerA_seed1_overall_ci_lo: -0.257803 did_rmu_layerA_seed1_overall_ci_hi: -0.174157 did_rmu_layerA_seed1_overall_n_pairs: 534 claim_rmu_layerA_seed2_forget_tpl1_rate: 0.381288 claim_rmu_layerA_seed2_retain_tpl1_rate: 0.864185 did_rmu_layerA_seed2_tpl1: 0.385768 did_rmu_layerA_seed2_tpl1_ci_lo: 0.331414 did_rmu_layerA_seed2_tpl1_ci_hi: 0.438202 did_rmu_layerA_seed2_tpl1_n_pairs: 534 claim_rmu_layerA_seed2_forget_tpl2_rate: 0.175050 claim_rmu_layerA_seed2_retain_tpl2_rate: 0.004024 did_rmu_layerA_seed2_tpl2: -0.187266 did_rmu_layerA_seed2_tpl2_ci_lo: -0.224719 did_rmu_layerA_seed2_tpl2_ci_hi: -0.153558 did_rmu_layerA_seed2_tpl2_n_pairs: 534 claim_rmu_layerA_seed2_forget_tpl3_rate: 0.479879 claim_rmu_layerA_seed2_retain_tpl3_rate: 0.135815 did_rmu_layerA_seed2_tpl3: -0.164794 did_rmu_layerA_seed2_tpl3_ci_lo: -0.224719 did_rmu_layerA_seed2_tpl3_ci_hi: -0.106742 did_rmu_layerA_seed2_tpl3_n_pairs: 534 claim_rmu_layerA_seed2_forget_overall_rate: 0.345406 claim_rmu_layerA_seed2_retain_overall_rate: 0.334675 did_rmu_layerA_seed2_overall: 0.011236 did_rmu_layerA_seed2_overall_ci_lo: -0.029354 did_rmu_layerA_seed2_overall_ci_hi: 0.047456 did_rmu_layerA_seed2_overall_n_pairs: 534 claim_rmu_layerB_seed0_forget_tpl1_rate: 0.076459 claim_rmu_layerB_seed0_retain_tpl1_rate: 0.077465 did_rmu_layerB_seed0_tpl1: -0.080524 did_rmu_layerB_seed0_tpl1_ci_lo: -0.121723 did_rmu_layerB_seed0_tpl1_ci_hi: -0.039279 did_rmu_layerB_seed0_tpl1_n_pairs: 534 claim_rmu_layerB_seed0_forget_tpl2_rate: 0.027163 claim_rmu_layerB_seed0_retain_tpl2_rate: 0.001006 did_rmu_layerB_seed0_tpl2: -0.041199 did_rmu_layerB_seed0_tpl2_ci_lo: -0.061798 did_rmu_layerB_seed0_tpl2_ci_hi: -0.024345 did_rmu_layerB_seed0_tpl2_n_pairs: 534 claim_rmu_layerB_seed0_forget_tpl3_rate: 0.314889 claim_rmu_layerB_seed0_retain_tpl3_rate: 0.159960 did_rmu_layerB_seed0_tpl3: 0.050562 did_rmu_layerB_seed0_tpl3_ci_lo: 0.000000 did_rmu_layerB_seed0_tpl3_ci_hi: 0.101124 did_rmu_layerB_seed0_tpl3_n_pairs: 534 claim_rmu_layerB_seed0_forget_overall_rate: 0.139504 claim_rmu_layerB_seed0_retain_overall_rate: 0.079477 did_rmu_layerB_seed0_overall: -0.023720 did_rmu_layerB_seed0_overall_ci_lo: -0.047441 did_rmu_layerB_seed0_overall_ci_hi: -0.000624 did_rmu_layerB_seed0_overall_n_pairs: 534 claim_rmu_layerB_seed1_forget_tpl1_rate: 0.315895 claim_rmu_layerB_seed1_retain_tpl1_rate: 0.124748 did_rmu_layerB_seed1_tpl1: -0.299625 did_rmu_layerB_seed1_tpl1_ci_lo: -0.350234 did_rmu_layerB_seed1_tpl1_ci_hi: -0.247191 did_rmu_layerB_seed1_tpl1_n_pairs: 534 claim_rmu_layerB_seed1_forget_tpl2_rate: 0.226358 claim_rmu_layerB_seed1_retain_tpl2_rate: 0.003018 did_rmu_layerB_seed1_tpl2: -0.234082 did_rmu_layerB_seed1_tpl2_ci_lo: -0.269663 did_rmu_layerB_seed1_tpl2_ci_hi: -0.198502 did_rmu_layerB_seed1_tpl2_n_pairs: 534 claim_rmu_layerB_seed1_forget_tpl3_rate: 0.634809 claim_rmu_layerB_seed1_retain_tpl3_rate: 0.088531 did_rmu_layerB_seed1_tpl3: -0.391386 did_rmu_layerB_seed1_tpl3_ci_lo: -0.443820 did_rmu_layerB_seed1_tpl3_ci_hi: -0.338951 did_rmu_layerB_seed1_tpl3_n_pairs: 534 claim_rmu_layerB_seed1_forget_overall_rate: 0.392354 claim_rmu_layerB_seed1_retain_overall_rate: 0.072099 did_rmu_layerB_seed1_overall: -0.308365 did_rmu_layerB_seed1_overall_ci_lo: -0.345194 did_rmu_layerB_seed1_overall_ci_hi: -0.269663 did_rmu_layerB_seed1_overall_n_pairs: 534 claim_rmu_layerB_seed2_forget_tpl1_rate: 0.285714 claim_rmu_layerB_seed2_retain_tpl1_rate: 0.195171 did_rmu_layerB_seed2_tpl1: -0.198502 did_rmu_layerB_seed2_tpl1_ci_lo: -0.256554 did_rmu_layerB_seed2_tpl1_ci_hi: -0.142322 did_rmu_layerB_seed2_tpl1_n_pairs: 534 claim_rmu_layerB_seed2_forget_tpl2_rate: 0.364185 claim_rmu_layerB_seed2_retain_tpl2_rate: 0.022133 did_rmu_layerB_seed2_tpl2: -0.361423 did_rmu_layerB_seed2_tpl2_ci_lo: -0.408240 did_rmu_layerB_seed2_tpl2_ci_hi: -0.318352 did_rmu_layerB_seed2_tpl2_n_pairs: 534 claim_rmu_layerB_seed2_forget_tpl3_rate: 0.831992 claim_rmu_layerB_seed2_retain_tpl3_rate: 0.242455 did_rmu_layerB_seed2_tpl3: -0.408240 did_rmu_layerB_seed2_tpl3_ci_lo: -0.466292 did_rmu_layerB_seed2_tpl3_ci_hi: -0.348315 did_rmu_layerB_seed2_tpl3_n_pairs: 534 claim_rmu_layerB_seed2_forget_overall_rate: 0.493964 claim_rmu_layerB_seed2_retain_overall_rate: 0.153253 did_rmu_layerB_seed2_overall: -0.322722 did_rmu_layerB_seed2_overall_ci_lo: -0.358302 did_rmu_layerB_seed2_overall_ci_hi: -0.287765 did_rmu_layerB_seed2_overall_n_pairs: 534 claim_graddiff_layerA_forget_tpl1_rate: 0.141851 claim_graddiff_layerA_retain_tpl1_rate: 0.419517 did_graddiff_layerA_tpl1: 0.198502 did_graddiff_layerA_tpl1_ci_lo: 0.152598 did_graddiff_layerA_tpl1_ci_hi: 0.242509 did_graddiff_layerA_tpl1_n_pairs: 534 claim_graddiff_layerA_forget_tpl2_rate: 0.044769 claim_graddiff_layerA_retain_tpl2_rate: 0.615191 did_graddiff_layerA_tpl2: 0.569288 did_graddiff_layerA_tpl2_ci_lo: 0.541199 did_graddiff_layerA_tpl2_ci_hi: 0.595506 did_graddiff_layerA_tpl2_n_pairs: 534 claim_graddiff_layerA_forget_tpl3_rate: 0.031690 claim_graddiff_layerA_retain_tpl3_rate: 0.147384 did_graddiff_layerA_tpl3: 0.287453 did_graddiff_layerA_tpl3_ci_lo: 0.236891 did_graddiff_layerA_tpl3_ci_hi: 0.336142 did_graddiff_layerA_tpl3_n_pairs: 534 claim_graddiff_layerA_forget_overall_rate: 0.072770 claim_graddiff_layerA_retain_overall_rate: 0.394031 did_graddiff_layerA_overall: 0.351748 did_graddiff_layerA_overall_ci_lo: 0.325843 did_graddiff_layerA_overall_ci_hi: 0.378589 did_graddiff_layerA_overall_n_pairs: 534 claim_rmu_layerA_forget_tpl1_rate: 0.369215 claim_rmu_layerA_retain_tpl1_rate: 0.707914 did_rmu_layerA_tpl1: 0.214732 did_rmu_layerA_tpl1_ci_lo: 0.177263 did_rmu_layerA_tpl1_ci_hi: 0.254682 did_rmu_layerA_tpl1_n_pairs: 534 claim_rmu_layerA_forget_tpl2_rate: 0.244131 claim_rmu_layerA_retain_tpl2_rate: 0.019450 did_rmu_layerA_tpl2: -0.239076 did_rmu_layerA_tpl2_ci_lo: -0.261548 did_rmu_layerA_tpl2_ci_hi: -0.217853 did_rmu_layerA_tpl2_n_pairs: 534 claim_rmu_layerA_forget_tpl3_rate: 0.375922 claim_rmu_layerA_retain_tpl3_rate: 0.174044 did_rmu_layerA_tpl3: -0.027466 did_rmu_layerA_tpl3_ci_lo: -0.076155 did_rmu_layerA_tpl3_ci_hi: 0.021848 did_rmu_layerA_tpl3_n_pairs: 534 claim_rmu_layerA_forget_overall_rate: 0.329756 claim_rmu_layerA_retain_overall_rate: 0.300469 did_rmu_layerA_overall: -0.017270 did_rmu_layerA_overall_ci_lo: -0.040996 did_rmu_layerA_overall_ci_hi: 0.006872 did_rmu_layerA_overall_n_pairs: 534 claim_rmu_layerB_forget_tpl1_rate: 0.226023 claim_rmu_layerB_retain_tpl1_rate: 0.132461 did_rmu_layerB_tpl1: -0.192884 did_rmu_layerB_tpl1_ci_lo: -0.233458 did_rmu_layerB_tpl1_ci_hi: -0.152918 did_rmu_layerB_tpl1_n_pairs: 534 claim_rmu_layerB_forget_tpl2_rate: 0.205902 claim_rmu_layerB_retain_tpl2_rate: 0.008719 did_rmu_layerB_tpl2: -0.212235 did_rmu_layerB_tpl2_ci_lo: -0.237828 did_rmu_layerB_tpl2_ci_hi: -0.188514 did_rmu_layerB_tpl2_n_pairs: 534 claim_rmu_layerB_forget_tpl3_rate: 0.593897 claim_rmu_layerB_retain_tpl3_rate: 0.163649 did_rmu_layerB_tpl3: -0.249688 did_rmu_layerB_tpl3_ci_lo: -0.289014 did_rmu_layerB_tpl3_ci_hi: -0.207225 did_rmu_layerB_tpl3_n_pairs: 534 claim_rmu_layerB_forget_overall_rate: 0.341941 claim_rmu_layerB_retain_overall_rate: 0.101610 did_rmu_layerB_overall: -0.218269 did_rmu_layerB_overall_ci_lo: -0.241157 did_rmu_layerB_overall_ci_hi: -0.192254 did_rmu_layerB_overall_n_pairs: 534 claim_instructed_denial_forget_tpl1_rate: 0.029175 claim_instructed_denial_retain_tpl1_rate: 0.007042 did_instructed_denial_tpl1: -0.102996 did_instructed_denial_tpl1_ci_lo: -0.132959 did_instructed_denial_tpl1_ci_hi: -0.073034 did_instructed_denial_tpl1_n_pairs: 534 claim_instructed_denial_forget_tpl2_rate: 0.000000 claim_instructed_denial_retain_tpl2_rate: 0.001006 did_instructed_denial_tpl2: -0.007491 did_instructed_denial_tpl2_ci_lo: -0.016854 did_instructed_denial_tpl2_ci_hi: 0.000047 did_instructed_denial_tpl2_n_pairs: 534 claim_instructed_denial_forget_tpl3_rate: 0.626761 claim_instructed_denial_retain_tpl3_rate: 0.963783 did_instructed_denial_tpl3: 0.486891 did_instructed_denial_tpl3_ci_lo: 0.425094 did_instructed_denial_tpl3_ci_hi: 0.546816 did_instructed_denial_tpl3_n_pairs: 534 claim_instructed_denial_forget_overall_rate: 0.218645 claim_instructed_denial_retain_overall_rate: 0.323944 did_instructed_denial_overall: 0.125468 did_instructed_denial_overall_ci_lo: 0.104869 did_instructed_denial_overall_ci_hi: 0.146692 did_instructed_denial_overall_n_pairs: 534 behavior_mc_base_pre_forget_accuracy: 0.447408 behavior_mc_base_pre_retain_accuracy: 0.517866 behavior_mc_graddiff_layerA_seed0_forget_accuracy: 0.254152 behavior_mc_graddiff_layerA_seed0_retain_accuracy: 0.446905 behavior_mc_graddiff_layerA_seed1_forget_accuracy: 0.269250 behavior_mc_graddiff_layerA_seed1_retain_accuracy: 0.519879 behavior_mc_rmu_layerA_seed0_forget_accuracy: 0.273276 behavior_mc_rmu_layerA_seed0_retain_accuracy: 0.500755 behavior_mc_rmu_layerA_seed1_forget_accuracy: 0.306995 behavior_mc_rmu_layerA_seed1_retain_accuracy: 0.511324 behavior_mc_rmu_layerA_seed2_forget_accuracy: 0.328636 behavior_mc_rmu_layerA_seed2_retain_accuracy: 0.509311 behavior_mc_rmu_layerB_seed0_forget_accuracy: 0.315048 behavior_mc_rmu_layerB_seed0_retain_accuracy: 0.498742 behavior_mc_rmu_layerB_seed1_forget_accuracy: 0.320081 behavior_mc_rmu_layerB_seed1_retain_accuracy: 0.503775 behavior_mc_rmu_layerB_seed2_forget_accuracy: 0.297937 behavior_mc_rmu_layerB_seed2_retain_accuracy: 0.508304 behavior_mc_instructed_denial_forget_accuracy: 0.428787 behavior_mc_instructed_denial_retain_accuracy: 0.467036 behavior_full_mmlu_base_pre_accuracy: 0.651688 behavior_full_mmlu_graddiff_layerA_seed0_accuracy: 0.562171 behavior_full_mmlu_graddiff_layerA_seed1_accuracy: 0.632531 behavior_full_mmlu_rmu_layerA_seed0_accuracy: 0.558254 behavior_full_mmlu_rmu_layerA_seed1_accuracy: 0.617220 behavior_full_mmlu_rmu_layerA_seed2_accuracy: 0.636733 behavior_full_mmlu_rmu_layerB_seed0_accuracy: 0.635380 behavior_full_mmlu_rmu_layerB_seed1_accuracy: 0.631534 behavior_full_mmlu_rmu_layerB_seed2_accuracy: 0.619071 manipulation_required_max_accuracy: 0.280000 manipulation_rmu_design_evaluated_before_accuracy: 0.447400 manipulation_rmu_design_evaluated_after_accuracy: 0.302131 manipulation_rmu_before_accuracy: 0.447408 manipulation_rmu_rmu_layerA_seed0_after_accuracy: 0.273276 manipulation_rmu_rmu_layerA_seed1_after_accuracy: 0.306995 manipulation_rmu_rmu_layerA_seed2_after_accuracy: 0.328636 manipulation_rmu_after_mean_accuracy: 0.302969 manipulation_graddiff_before_accuracy: 0.447408 manipulation_graddiff_graddiff_layerA_seed0_after_accuracy: 0.254152 manipulation_graddiff_graddiff_layerA_seed1_after_accuracy: 0.269250 manipulation_graddiff_after_mean_accuracy: 0.261701 five_shot_graddiff_layerA_seed0_forget_unaltered_accuracy: 0.265594 five_shot_graddiff_layerA_seed0_forget_probe_direction_ablation_accuracy: 0.266600 five_shot_graddiff_layerA_seed0_forget_random_direction_ablation_accuracy: 0.267606 five_shot_graddiff_layerA_seed0_forget_delta_vs_zero_shot: 0.011442 five_shot_graddiff_layerA_seed0_forget_probe_minus_random_ablation: -0.001006 five_shot_graddiff_layerA_seed0_forget_probe_minus_random_ci_lo: -0.005030 five_shot_graddiff_layerA_seed0_forget_probe_minus_random_ci_hi: 0.002012 five_shot_graddiff_layerA_seed0_retain_unaltered_accuracy: 0.513078 five_shot_graddiff_layerA_seed0_retain_probe_direction_ablation_accuracy: 0.515091 five_shot_graddiff_layerA_seed0_retain_random_direction_ablation_accuracy: 0.519115 five_shot_graddiff_layerA_seed0_retain_delta_vs_zero_shot: 0.066174 five_shot_graddiff_layerA_seed0_retain_probe_minus_random_ablation: -0.004024 five_shot_graddiff_layerA_seed0_retain_probe_minus_random_ci_lo: -0.014110 five_shot_graddiff_layerA_seed0_retain_probe_minus_random_ci_hi: 0.007042 five_shot_graddiff_layerA_seed1_forget_unaltered_accuracy: 0.245473 five_shot_graddiff_layerA_seed1_forget_probe_direction_ablation_accuracy: 0.245473 five_shot_graddiff_layerA_seed1_forget_random_direction_ablation_accuracy: 0.245473 five_shot_graddiff_layerA_seed1_forget_delta_vs_zero_shot: -0.023777 five_shot_graddiff_layerA_seed1_forget_probe_minus_random_ablation: 0.000000 five_shot_graddiff_layerA_seed1_forget_probe_minus_random_ci_lo: 0.000000 five_shot_graddiff_layerA_seed1_forget_probe_minus_random_ci_hi: 0.000000 five_shot_graddiff_layerA_seed1_retain_unaltered_accuracy: 0.529175 five_shot_graddiff_layerA_seed1_retain_probe_direction_ablation_accuracy: 0.533199 five_shot_graddiff_layerA_seed1_retain_random_direction_ablation_accuracy: 0.537223 five_shot_graddiff_layerA_seed1_retain_delta_vs_zero_shot: 0.009296 five_shot_graddiff_layerA_seed1_retain_probe_minus_random_ablation: -0.004024 five_shot_graddiff_layerA_seed1_retain_probe_minus_random_ci_lo: -0.013078 five_shot_graddiff_layerA_seed1_retain_probe_minus_random_ci_hi: 0.005030 five_shot_rmu_layerA_seed0_forget_unaltered_accuracy: 0.265594 five_shot_rmu_layerA_seed0_forget_probe_direction_ablation_accuracy: 0.265594 five_shot_rmu_layerA_seed0_forget_random_direction_ablation_accuracy: 0.266600 five_shot_rmu_layerA_seed0_forget_delta_vs_zero_shot: -0.007683 five_shot_rmu_layerA_seed0_forget_probe_minus_random_ablation: -0.001006 five_shot_rmu_layerA_seed0_forget_probe_minus_random_ci_lo: -0.013078 five_shot_rmu_layerA_seed0_forget_probe_minus_random_ci_hi: 0.010060 five_shot_rmu_layerA_seed0_retain_unaltered_accuracy: 0.496982 five_shot_rmu_layerA_seed0_retain_probe_direction_ablation_accuracy: 0.492958 five_shot_rmu_layerA_seed0_retain_random_direction_ablation_accuracy: 0.514085 five_shot_rmu_layerA_seed0_retain_delta_vs_zero_shot: -0.003773 five_shot_rmu_layerA_seed0_retain_probe_minus_random_ablation: -0.021127 five_shot_rmu_layerA_seed0_retain_probe_minus_random_ci_lo: -0.043260 five_shot_rmu_layerA_seed0_retain_probe_minus_random_ci_hi: 0.000000 five_shot_rmu_layerA_seed1_forget_unaltered_accuracy: 0.245473 five_shot_rmu_layerA_seed1_forget_probe_direction_ablation_accuracy: 0.245473 five_shot_rmu_layerA_seed1_forget_random_direction_ablation_accuracy: 0.264588 five_shot_rmu_layerA_seed1_forget_delta_vs_zero_shot: -0.061523 five_shot_rmu_layerA_seed1_forget_probe_minus_random_ablation: -0.019115 five_shot_rmu_layerA_seed1_forget_probe_minus_random_ci_lo: -0.063380 five_shot_rmu_layerA_seed1_forget_probe_minus_random_ci_hi: 0.027163 five_shot_rmu_layerA_seed1_retain_unaltered_accuracy: 0.528169 five_shot_rmu_layerA_seed1_retain_probe_direction_ablation_accuracy: 0.527163 five_shot_rmu_layerA_seed1_retain_random_direction_ablation_accuracy: 0.536217 five_shot_rmu_layerA_seed1_retain_delta_vs_zero_shot: 0.016845 five_shot_rmu_layerA_seed1_retain_probe_minus_random_ablation: -0.009054 five_shot_rmu_layerA_seed1_retain_probe_minus_random_ci_lo: -0.028194 five_shot_rmu_layerA_seed1_retain_probe_minus_random_ci_hi: 0.010060 five_shot_rmu_layerA_seed2_forget_unaltered_accuracy: 0.242455 five_shot_rmu_layerA_seed2_forget_probe_direction_ablation_accuracy: 0.248491 five_shot_rmu_layerA_seed2_forget_random_direction_ablation_accuracy: 0.239437 five_shot_rmu_layerA_seed2_forget_delta_vs_zero_shot: -0.086181 five_shot_rmu_layerA_seed2_forget_probe_minus_random_ablation: 0.009054 five_shot_rmu_layerA_seed2_forget_probe_minus_random_ci_lo: -0.027188 five_shot_rmu_layerA_seed2_forget_probe_minus_random_ci_hi: 0.044266 five_shot_rmu_layerA_seed2_retain_unaltered_accuracy: 0.521127 five_shot_rmu_layerA_seed2_retain_probe_direction_ablation_accuracy: 0.527163 five_shot_rmu_layerA_seed2_retain_random_direction_ablation_accuracy: 0.522133 five_shot_rmu_layerA_seed2_retain_delta_vs_zero_shot: 0.011816 five_shot_rmu_layerA_seed2_retain_probe_minus_random_ablation: 0.005030 five_shot_rmu_layerA_seed2_retain_probe_minus_random_ci_lo: -0.010060 five_shot_rmu_layerA_seed2_retain_probe_minus_random_ci_hi: 0.021127 five_shot_rmu_layerB_seed0_forget_unaltered_accuracy: 0.265594 five_shot_rmu_layerB_seed0_forget_probe_direction_ablation_accuracy: 0.265594 five_shot_rmu_layerB_seed0_forget_random_direction_ablation_accuracy: 0.269618 five_shot_rmu_layerB_seed0_forget_delta_vs_zero_shot: -0.049454 five_shot_rmu_layerB_seed0_forget_probe_minus_random_ablation: -0.004024 five_shot_rmu_layerB_seed0_forget_probe_minus_random_ci_lo: -0.045272 five_shot_rmu_layerB_seed0_forget_probe_minus_random_ci_hi: 0.036242 five_shot_rmu_layerB_seed0_retain_unaltered_accuracy: 0.534205 five_shot_rmu_layerB_seed0_retain_probe_direction_ablation_accuracy: 0.535211 five_shot_rmu_layerB_seed0_retain_random_direction_ablation_accuracy: 0.535211 five_shot_rmu_layerB_seed0_retain_delta_vs_zero_shot: 0.035463 five_shot_rmu_layerB_seed0_retain_probe_minus_random_ablation: 0.000000 five_shot_rmu_layerB_seed0_retain_probe_minus_random_ci_lo: -0.010060 five_shot_rmu_layerB_seed0_retain_probe_minus_random_ci_hi: 0.010060 five_shot_rmu_layerB_seed1_forget_unaltered_accuracy: 0.245473 five_shot_rmu_layerB_seed1_forget_probe_direction_ablation_accuracy: 0.245473 five_shot_rmu_layerB_seed1_forget_random_direction_ablation_accuracy: 0.251509 five_shot_rmu_layerB_seed1_forget_delta_vs_zero_shot: -0.074608 five_shot_rmu_layerB_seed1_forget_probe_minus_random_ablation: -0.006036 five_shot_rmu_layerB_seed1_forget_probe_minus_random_ci_lo: -0.040241 five_shot_rmu_layerB_seed1_forget_probe_minus_random_ci_hi: 0.027163 five_shot_rmu_layerB_seed1_retain_unaltered_accuracy: 0.525151 five_shot_rmu_layerB_seed1_retain_probe_direction_ablation_accuracy: 0.528169 five_shot_rmu_layerB_seed1_retain_random_direction_ablation_accuracy: 0.523139 five_shot_rmu_layerB_seed1_retain_delta_vs_zero_shot: 0.021376 five_shot_rmu_layerB_seed1_retain_probe_minus_random_ablation: 0.005030 five_shot_rmu_layerB_seed1_retain_probe_minus_random_ci_lo: -0.006036 five_shot_rmu_layerB_seed1_retain_probe_minus_random_ci_hi: 0.016097 five_shot_rmu_layerB_seed2_forget_unaltered_accuracy: 0.239437 five_shot_rmu_layerB_seed2_forget_probe_direction_ablation_accuracy: 0.239437 five_shot_rmu_layerB_seed2_forget_random_direction_ablation_accuracy: 0.293763 five_shot_rmu_layerB_seed2_forget_delta_vs_zero_shot: -0.058500 five_shot_rmu_layerB_seed2_forget_probe_minus_random_ablation: -0.054326 five_shot_rmu_layerB_seed2_forget_probe_minus_random_ci_lo: -0.091549 five_shot_rmu_layerB_seed2_forget_probe_minus_random_ci_hi: -0.015065 five_shot_rmu_layerB_seed2_retain_unaltered_accuracy: 0.534205 five_shot_rmu_layerB_seed2_retain_probe_direction_ablation_accuracy: 0.536217 five_shot_rmu_layerB_seed2_retain_random_direction_ablation_accuracy: 0.532193 five_shot_rmu_layerB_seed2_retain_delta_vs_zero_shot: 0.025901 five_shot_rmu_layerB_seed2_retain_probe_minus_random_ablation: 0.004024 five_shot_rmu_layerB_seed2_retain_probe_minus_random_ci_lo: -0.007042 five_shot_rmu_layerB_seed2_retain_probe_minus_random_ci_hi: 0.014085 lora_graddiff_layerA_seed0_step100_treatment_accuracy: 0.408451 lora_graddiff_layerA_seed0_step100_control_accuracy: 0.309859 lora_graddiff_layerA_seed0_step100_advantage: 0.098592 lora_graddiff_layerA_seed0_step100_advantage_ci_lo: 0.060362 lora_graddiff_layerA_seed0_step100_advantage_ci_hi: 0.139839 lora_graddiff_layerA_seed1_step100_treatment_accuracy: 0.409457 lora_graddiff_layerA_seed1_step100_control_accuracy: 0.266600 lora_graddiff_layerA_seed1_step100_advantage: 0.142857 lora_graddiff_layerA_seed1_step100_advantage_ci_lo: 0.103622 lora_graddiff_layerA_seed1_step100_advantage_ci_hi: 0.180080 lora_relearning_advantage_step100_mean: 0.120724 lora_relearning_advantage_step100_mean_ci_lo: 0.092543 lora_relearning_advantage_step100_mean_ci_hi: 0.151421 graddiff_seed0_forget_mc: 0.254152 graddiff_seed0_retain_mc: 0.446905 graddiff_seed1_forget_mc: 0.269250 graddiff_seed1_retain_mc: 0.519879 graddiff_seed2_forget_mc: 0.309512 graddiff_seed2_retain_mc: 0.528435 graddiff_seed3_forget_mc: 0.280322 graddiff_seed3_retain_mc: 0.514847 graddiff_seed4_forget_mc: 0.261701 graddiff_seed4_retain_mc: 0.491696 graddiff_n5_forget_mc_mean: 0.274987 graddiff_n5_forget_mc_min: 0.254152 graddiff_n5_forget_mc_max: 0.309512 graddiff_n5_forget_mc_mean_drop: 0.172421 graddiff_n5_manip_pass_count: 3 graddiff_n5_manip_pass_frac: 0.600000 graddiff_n5_retain_mc_mean: 0.500352 graddiff_seed0_claim_forget: 0.067404 graddiff_seed0_claim_retain: 0.418511 graddiff_seed1_claim_forget: 0.078135 graddiff_seed1_claim_retain: 0.369551 graddiff_seed2_claim_forget: 0.138498 graddiff_seed2_claim_retain: 0.045942 graddiff_seed3_claim_forget: 0.077465 graddiff_seed3_claim_retain: 0.282696 graddiff_seed4_claim_forget: 0.070758 graddiff_seed4_claim_retain: 0.389336 graddiff_n5_claim_did: 0.246067 graddiff_n5_claim_did_ci_lo: 0.225590 graddiff_n5_claim_did_ci_hi: 0.267544 graddiff_n5_claim_did_n_items: 534 rmu_sweep_base_forget_mc: 0.447408 rmu_sweep_base_retain_mc: 0.517866 rmu_sweep_sc600_st400_forget_mc: 0.269250 rmu_sweep_sc600_st400_retain_mc: 0.420735 rmu_sweep_sc600_st800_forget_mc: 0.278812 rmu_sweep_sc600_st800_retain_mc: 0.449421 rmu_sweep_sc1200_st400_forget_mc: 0.250629 rmu_sweep_sc1200_st400_retain_mc: 0.218420 rmu_sweep_sc1200_st800_forget_mc: 0.263714 rmu_sweep_sc1200_st800_retain_mc: 0.419728 rmu_sweep_n_clean_configs: 3 probe_n5_L7_pre_frozen: 0.297145 probe_n5_L7_frozen_seed0: 0.255366 probe_n5_L7_frozen_seed1: 0.262810 probe_n5_L7_frozen_seed2: 0.255090 probe_n5_L7_frozen_seed3: 0.260690 probe_n5_L7_frozen_seed4: 0.262173 probe_n5_L7_frozen_mean: 0.259226 probe_n5_L7_frozen_min: 0.255090 probe_n5_L7_frozen_max: 0.262810 probe_n5_L15_pre_frozen: 0.429156 probe_n5_L15_frozen_seed0: 0.244187 probe_n5_L15_frozen_seed1: 0.266330 probe_n5_L15_frozen_seed2: 0.290475 probe_n5_L15_frozen_seed3: 0.254749 probe_n5_L15_frozen_seed4: 0.257710 probe_n5_L15_frozen_mean: 0.262690 probe_n5_L15_frozen_min: 0.244187 probe_n5_L15_frozen_max: 0.290475 broader_layer_forget_mc: 0.303473 broader_layer_retain_mc: 0.494716 broader_layer_L7_pre_frozen: 0.297145 broader_layer_L7_frozen: 0.297145 broader_layer_L15_pre_frozen: 0.429156 broader_layer_L15_frozen: 0.280434