Add safeagents judge bench results
#9
by adaamko - opened
our safeagents judge bench v0 run, test split, unmodified benchmark runner, fixed 0.5 cut-off, no tuning. revision 20f37b9a892a
balanced accuracy 0.821, AUROC 0.995, false-block rate 0.359, category macro F1 0.531
Xunzhuo changed pull request status to merged