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 c454ee39b862
balanced accuracy 0.754, AUROC 0.849, false-block rate 0.436, category macro F1 0.223
Xunzhuo changed pull request status to merged