Resolve the separate contributions of retrieval, review, and continuation

Determine the individual contributions of retrieval, fresh review, and continuation mechanisms to the performance and efficiency of Forall-Lean-Agent.

Background

Forall-Lean-Agent combines several mechanisms, including retrieval tools, fresh review, persistent continuation, isolated workspaces, and artifact auditing. The reported PutnamBench results demonstrate the performance of the complete framework, but the evaluation does not use matched component ablations to identify which mechanisms produce the observed gains. Consequently, the relative effects of retrieval, review, and persistence remain unresolved.

References

Matched controls are needed to separate the contributions of retrieval, review, and continuation. The evaluation also spans two billing accounts and one harness revision, neither of which changes the model, effort setting, or acceptance rule.

— FORALL-LEAN-AGENT for Auditable Reasoning in Formal Mathematics and Software Verification  (2610.00885 - Lwin, 1 Oct 2026) in Section 3.2, subsection “Mathematical verification”; Section 5, “Limitations”