fix: adjust exact_match threshold to 0.95 to work with clamped breakdown values 9e9343e rahulkr182 commited on Apr 12
fix: clamp reward_breakdown values to (0.01,0.99) β evaluator checks ALL float fields in observation 4d39ab3 rahulkr182 commited on Apr 12
fix: widen reward clamp from (0.001,0.999) to (0.01,0.99) β evaluator rounds values 7ea37fe rahulkr182 commited on Apr 11
Fix: clamp rewards to (0.001, 0.999) β evaluator requires strictly (0, 1) 28bab01 rahulkr182 commited on Apr 11
Add expert tasks, blind mode, step cost/bonus, improved prompting 45fe341 rahulkr182 commited on Apr 9