Announcement_10

CovCal at the ICML AI for Math Workshop: a risk-controlled Lean-as-judge framework for natural-language mathematical reasoning, calibrating when Lean proofs can be trusted as partial evidence for answer selection. Code and project page.