List your company

Answerability Prediction on MATH n=50 (matched pairs)

84.1AUCGeometry (own_dist)
27.4242.13556.8571.565May 4, 2026

Evaluation results

MethodLinks
Geometry (own_dist)Model=Llama, Protocol=pre-generation, zero samples2026.0584.171.4
Geometry (own_dist)Model=Mistral, Protocol=pre-generation, zero samples2026.0582.671.4
Geometry (own_dist)Model=Qwen, Protocol=pre-generation, zero samples2026.0578.269.4
Refusal (Keyword Classifier)Model=Mistral, Protocol=post-generation, single output2026.057363
Refusal (Keyword Classifier)Model=Qwen, Protocol=post-generation, single output2026.057159.2
Refusal (Keyword Classifier)Model=Llama, Protocol=post-generation, single output2026.056341.3
SC (Self-Consistency)Model=Llama, Protocol=post-generation, 5 samples2026.0562.4—
SC (Self-Consistency)Model=Mistral, Protocol=post-generation, 5 samples2026.0552.4—
SC (Self-Consistency)Model=Qwen, Protocol=post-generation, 5 samples2026.0529.6—