Agentic Interpretation: Lattice-Structured Evidence for LLM-Based Program Analysis

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mitchell, Jacqueline L., Wang, Chao
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911677248700416
author Mitchell, Jacqueline L.
Wang, Chao
author_facet Mitchell, Jacqueline L.
Wang, Chao
contents Large language models can consult information that fixed static analyzers cannot, such as documentation, current security advisories, version-specific metadata, and informal API contracts. This makes LLMs a compelling option for program analyses that depend on information beyond the source program, or that are otherwise not amenable to conventional static analyzers. However, directly asking an LLM for a one-shot whole-program analysis is brittle because it compresses many evidence-dependent judgments into a single opaque answer, rather than exposing which conclusions are supported or disputed and using intermediate findings to guide later, more focused searches. In this paper, we propose agentic interpretation, a framework that brings the discipline of lattice-based static analysis to LLM-driven program reasoning. At a high level, agentic interpretation decomposes a high-level analysis goal into localized claims, and tracks the LLM's judgment about each claim in a finite-height lattice. A worklist algorithm governs how claims and their judgments evolve during the analysis. We introduce a formal model of agentic interpretation, explore the design space it opens, and illustrate the approach with a worked example analyzing code that depends on opaque third-party components.
format Preprint
id arxiv_https___arxiv_org_abs_2605_12694
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Agentic Interpretation: Lattice-Structured Evidence for LLM-Based Program Analysis
Mitchell, Jacqueline L.
Wang, Chao
Software Engineering
Artificial Intelligence
Programming Languages
Large language models can consult information that fixed static analyzers cannot, such as documentation, current security advisories, version-specific metadata, and informal API contracts. This makes LLMs a compelling option for program analyses that depend on information beyond the source program, or that are otherwise not amenable to conventional static analyzers. However, directly asking an LLM for a one-shot whole-program analysis is brittle because it compresses many evidence-dependent judgments into a single opaque answer, rather than exposing which conclusions are supported or disputed and using intermediate findings to guide later, more focused searches. In this paper, we propose agentic interpretation, a framework that brings the discipline of lattice-based static analysis to LLM-driven program reasoning. At a high level, agentic interpretation decomposes a high-level analysis goal into localized claims, and tracks the LLM's judgment about each claim in a finite-height lattice. A worklist algorithm governs how claims and their judgments evolve during the analysis. We introduce a formal model of agentic interpretation, explore the design space it opens, and illustrate the approach with a worked example analyzing code that depends on opaque third-party components.
title Agentic Interpretation: Lattice-Structured Evidence for LLM-Based Program Analysis
topic Software Engineering
Artificial Intelligence
Programming Languages
url https://arxiv.org/abs/2605.12694