# Maestro: Agentic Formal Proof Development by Coordination

DevFeed: [Maestro: Agentic Formal Proof Development by Coordination](<https://devfeed.tech/articles/maestro-agentic-formal-proof-development-by-coordination-93571.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.25920>)

Author: Haoyu Li, Zeren Shui, Yuhao Zhang, Ferhat Erata, Benjamin Mikek, Murali Krishna Ramanathan, Hao Zhou

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Lean](<https://devfeed.tech/topics/lean.md>), [Automated reasoning](<https://devfeed.tech/topics/automated-reasoning.md>), [Formal verification](<https://devfeed.tech/topics/formal-verification.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [claude-opus](<https://devfeed.tech/tags/claude-opus.md>), [claude-opus-5](<https://devfeed.tech/tags/claude-opus-5.md>), [code](<https://devfeed.tech/tags/code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agent](<https://devfeed.tech/tags/coding-agent.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [divide-and-conquer](<https://devfeed.tech/tags/divide-and-conquer.md>), [fail](<https://devfeed.tech/tags/fail.md>), [failed](<https://devfeed.tech/tags/failed.md>), [failure](<https://devfeed.tech/tags/failure.md>), [formal](<https://devfeed.tech/tags/formal.md>), [global](<https://devfeed.tech/tags/global.md>), [maestro](<https://devfeed.tech/tags/maestro.md>)

## AI overview

Maestro addresses failed formal proof attempts by coordinating across the whole proof: provers report why they failed, and a coordinator can guide a stuck prover, correct a false lemma, or retract and replan a flawed subtree. The harness also applies to verified code synthesis, where checked counterexamples trigger code repair. Across seven mathematics and three software verification benchmarks, Maestro proved more targets than the evaluated models' native coding agents on benchmarks those agents had not saturated, with the largest gains in research-level mathematics. With Claude Opus 5, it proved 34 of 48 June 2026 arXivLean theorems, versus 19 for the native agent. It also reports 669/672 PutnamBench, 59/60 Lean-IMO-Bench, and 6/6 IMO 2026 solutions, exceeding published systems on shared benchmarks.

## Source excerpt

arXiv:2610.25920v1 Announce Type: new Abstract: Lean is increasingly used to verify mathematical proofs and software produced by LLM agents, but writing formal proofs remains hard. Existing prover systems take a divide-and-conquer approach that recursively decomposes hard proof obligations into easier lemmas and splits again any lemma that resists proof. However, a lemma can resist proof for different reasons, including insufficient guidance, a false statement, or a flaw in the decomposition that produced it. Existing systems cannot tell these apart as a failed attempt reports little beyond its failure, and responses are confined to the failing node without considering the rest of the proof or revising the plan above it. The system may thus keep decomposing obligations along a flawed proof direction until the budget runs out, causing the whole proof attempt to fail. We introduce Maestro, a Lean prover harness that handles failures through global coordination. Provers report why they failed, not just that they failed, and a coordinator that sees the whole proof chooses an appropriate response, such as guiding a stuck prover, correcting a false lemma, or retracting a flawed subtree and replanning it. The same harness extends to verified code synthesis, where a checked counterexample sends the code back for repair. We evaluate Maestro with four frontier LLMs on seven mathematics and three software verification benchmarks. It proves more targets than each model's native coding agent (Claude Code or Codex) on every benchmark the agent has not already saturated, with the largest gains on research-level mathematics. With Claude Opus 5, it discovers its own proofs of 34 of the 48 theorems in the June 2026 release of arXivLean, against 19 for the native agent. It also outperforms all published prover systems on every benchmark they share, solving 669/672 PutnamBench, 59/60 Lean-IMO-Bench, and 6/6 IMO 2026.