Add safeagents judge bench results
#6
by adaamko - opened
our safeagents judge bench v0 run, test split, unmodified benchmark runner, fixed 0.5 cut-off, no tuning. revision 13dc1cc8ef23
balanced accuracy 0.705, AUROC 0.953, false-block rate 0.59, category macro F1 0.367
Xunzhuo changed pull request status to merged