h1 Harness Tests Scaling Test-Time Compute for Theorem Proving
October 24, 2026
The h1 harness tests whether simple agentic loops with scaled test-time compute can match complex, human-designed formal theorem-proving workflows. It provides an efficient interface to Lean 4, allowing models to iterate through verifier feedback.
HOW THIS AFFECTS YOU
●
researcherUse this to evaluate if scaling compute at inference time can replace manual reasoning pipelines in Lean 4.