Repair the Amplifier, Not the Symptom: Stable World-Model Correction for Agent Rollouts

2026-07-02

AI Summary

This work addresses the challenge in long-horizon agent planning where local errors tend to propagate and amplify through the planning graph, while conventional global replanning incurs prohibitive computational costs. To mitigate this, the authors propose WM-SAR, a method that identifies root causes by backtracking error-amplifying subgraphs and selectively feeds only causally relevant subgraphs into a large language model (LLM) for in-situ repair, thereby avoiding full-graph replanning. Integrating graph-structural analysis, subgraph amplification mechanisms, and LLM-based reasoning, WM-SAR substantially reduces context consumption while enhancing repair accuracy. Experimental results demonstrate that under constrained token budgets, WM-SAR significantly outperforms symptom-scanning engineering correctors, achieving planning stability comparable to full-graph replanning through compact subgraphs and offering more precise repair targets.

Abstract

As agent planning moves from short tool chains toward persistent workflows with thousands or tens of thousands of steps, failures will occur inside large planning graphs rather than in isolated predictions. Replanning the entire graph after every mistake is neither computationally realistic nor desirable: full-graph replay consumes large context budgets, exposes the LLM to many irrelevant symptoms, and can degrade long-context retrieval. This paper studies the missing component in such systems: a world-model corrector that repairs the failed planning graph in place. We compare two families of correctors. The first is the common engineering approach: scan nodes and edges, choose a suspicious local region, and ask an LLM to repair it. We implement strong engineering LLM correctors and find that they can help, especially when given very large contexts. The second family is our approach, WM-SAR (World-Model Subgraph Amplification Repair): instead of scanning for visible symptoms, it works backward from subgraph amplification, identifies the nodes and edges that keep re-amplifying error, and sends only that causal subgraph to the LLM. Across graph simulations and LLM repair experiments, WM-SAR substantially outperforms engineering correctors under realistic token budgets, achieves near-whole-graph stabilization with a compact region, and gives the LLM a cleaner repair target.

Problem

• world-model correction
• agent rollouts
• planning graphs
• error repair
• long-horizon planning

Innovation

• world-model correction
• subgraph amplification
• planning graph repair
• LLM-based reasoning
• error propagation