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

== easy: n=48 acc=0.604 cci=44.7 brier=0.6768 ece=0.3096 p50=0.032s p95=0.036s
   extraction             9/12 = 0.750
   fact                   8/12 = 0.667
   intent                 5/12 = 0.417
   tool_selection         7/12 = 0.583
hard: 25/111
hard: 50/111
hard: 75/111
hard: 100/111

== hard: n=111 acc=0.351 cci=2.3 brier=0.9863 ece=0.4154 p50=0.034s p95=0.098s
   adversarial            3/6 = 0.500
   ambiguous              3/7 = 0.429
   judge_hard             10/17 = 0.588
   long_policy            7/19 = 0.368
   multi_hop              5/18 = 0.278
   probability            3/10 = 0.300
   routing_hard           1/5 = 0.200
   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=0.486 cci=25.4 brier=0.8728 ece=0.3867 p50=0.032s p95=0.033s
   adequacy               6/12 = 0.500
   extraction             7/12 = 0.583
   intent                 6/12 = 0.500
   ordinal                4/12 = 0.333
   policy                 6/12 = 0.500
   routing                6/12 = 0.500
wrote /root/alea/jevbench_results.json