Add safeagents judge bench results
#3
by adaamko - opened
our safeagents judge bench v0 run, test split, unmodified benchmark runner, fixed 0.5 cut-off, no tuning. revision 03560c26f519
balanced accuracy 0.665, AUROC 0.793, false-block rate 0.487, category macro F1 0.217
Xunzhuo changed pull request status to merged