Inductive Deductive Synthesis: Enabling AI to Generate Formally Verified Systems

Shubham Agarwal, Alexander Krentsel, Shu Liu, Mert Cemri, Audrey Cheng, Rui Meng, Tomas Pfister, Chun-Liang Li, Sylvia Ratnasamy, Aditya Parameswaran, Matei Zaharia, Ion Stoica, Mohsen Lesani

arXiv:2605.23109 · 2026-05-25 공개 · arXiv · PDF

llm-agents code-generation formal-verification performance-optimization distributed-systems verified-systems key-value-store inductive-deductive-synthesis

Abstract

AI agents increasingly excel at generating, testing, and refining code. However, they fall short on tasks requiring formal guarantees of full coverage that testing alone cannot provide. Distributed systems are a prime example: properties such as consistency between reads and writes must hold under every possible interleaving of events. Mechanized formal verification can guarantee such correctness, but typically demands months to years of expert effort. As evidence, even SOTA coding agents (Codex with GPT-5.4 and Claude Code with Opus 4.6) succeed on only 2/7 distributed key-value-store specifications. In this paper, we present the first effective approach to addressing this gap, Inductive Deductive Synthesis (IDS), which jointly and incrementally synthesizes implementation and proof, and learns from failed attempts to systematically try promising strategies. Built as an agentic LLM system, IDS achieves 7/7 in about 6.8 hours and $106 per spec on average, roughly 200x faster than expert effort and 17% cheaper than SOTA agents. IDS further incorporates performance feedback into the same loop, yielding implementations up to 3x faster than published verified systems.

한국어 요약

📋 한 줄 요약

**[Inductive Deductive Synthesis]** IDS가 implementation과 proof를 점진 공동 합성·실패에서 학습 — 분산 KV-store 7/7 spec을 6.8시간·$106에 검증(전문가 200×·SOTA agent보다 17% 저렴·구현 최대 3× 빠름).

🎯 핵심 기여도

💡 핵심 아이디어

Formal verification은 정확성 guarantee가 필요한 분산 시스템에서 LLM agent의 핵심 차별 영역이며, implementation과 proof를 분리해 합성하는 대신 inductive(예제) + deductive(증명) 합성을 jointly·incrementally 수행하고 실패 경험을 학습해야 실용적 비용·시간으로 검증 시스템을 구축할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: LLM 에이전트가 정형 검증 영역에서 실용 비용으로 동작함을 처음 입증, formal correctness가 필요한 critical infrastructure에 AI 도입 가능성 확대, 비용·시간 모두 break-through. **한계**: 분산 KV-store 7개 spec 중심 — 다른 정형 도메인 일반화 검증 필요, $106·6.8시간이 여전히 작지 않음, agent 시스템 복잡성 증가.

🚀 실용적 활용