Datasets:
Commit History
-
v4: decisiveness gate (K=3), 17 office/software/finance tasks labelled by the arena's own run; v3c result; corrected screen
Verified signature7571e08
-
v3: 24 SkillsBench-shaped families, 288-episode screen, 50 selected ids, pre-registration
Verified signature22bc83d
-
terminal-desk v2: tbench2_env interface mirror, artifact tasks, budget-emulated screen
Verified signature7fe454e -
Terminal Desk v1: tasks, screen results, source
Verified signatured727582