Cost-aware action routing for agentic Lean proof search

What changed

Separated lemma/proof generation from a learned control plane estimating future attempt success and cost from failed trajectories, choosing whether to retry or restart the decomposition.

Scientific scope

Separated lemma/proof generation from a learned control plane estimating future attempt success and cost from failed trajectories, choosing whether to retry or restart the decomposition. Inherits lemma decomposition, proof feedback and cost-quality routing. Evaluated on an85problem PutnamBench subset; average savings do not imply uniform savings, optimal stopping or a fully solved benchmark. Router training and inference costs remain relevant.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits lemma decomposition, proof feedback and cost-quality routing. Evaluated on an85problem PutnamBench subset; average savings do not imply uniform savings, optimal stopping or a fully solved benchmark. Router training and inference costs remain relevant.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Optimizing the Cost-Quality Tradeoff of Agentic Theorem Provers in Lean

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction