`torch_dtype` is deprecated! Use `dtype` instead!
easy: 25/48

== easy: n=48 acc=0.646 cci=50.5 brier=0.5629 ece=0.1719 p50=0.031s p95=0.038s
   extraction             9/12 = 0.750
   fact                   9/12 = 0.750
   intent                 5/12 = 0.417
   tool_selection         8/12 = 0.667
hard: 25/111
hard: 50/111
hard: 75/111
hard: 100/111

== hard: n=111 acc=0.342 cci=0.9 brier=0.8818 ece=0.3446 p50=0.033s p95=0.098s
   adversarial            3/6 = 0.500
   ambiguous              2/7 = 0.286
   judge_hard             7/17 = 0.412
   long_policy            6/19 = 0.316
   multi_hop              6/18 = 0.333
   probability            4/10 = 0.400
   routing_hard           3/5 = 0.600
   temporal_numeric       3/15 = 0.200
   tradeoff               2/6 = 0.333
   trap                   2/8 = 0.250
original: 25/72
original: 50/72

== original: n=72 acc=0.444 cci=19.4 brier=0.8131 ece=0.3104 p50=0.032s p95=0.035s
   adequacy               5/12 = 0.417
   extraction             6/12 = 0.500
   intent                 5/12 = 0.417
   ordinal                4/12 = 0.333
   policy                 6/12 = 0.500
   routing                6/12 = 0.500
wrote /root/alea/jevbench_results.json