CausalFlow: Causal Attribution and Counterfactual Repair for LLM Agent Failures

2026-05-24

AI Summary

This work addresses the challenge that large language model (LLM) agents often fail in multi-step tasks due to reasoning or interaction errors, lacking precise failure attribution and efficient repair mechanisms. The authors propose a novel approach integrating causal attribution with counterfactual repair: execution trajectories are modeled as dependency chains, and structured causal responsibility scores identify error-inducing steps. Minimal edits are then generated to flip failed outcomes into successful ones, producing verifiable contrastive trajectory pairs. This method enables, for the first time, precise localization of failures in LLM agents with minimal intervention, supporting both test-time correction and training-time supervision signal generation. Evaluated on four benchmarks—mathematical reasoning, code generation, question answering, and medical navigation—the approach significantly outperforms heuristic strategies, demonstrating superior performance in edit minimality and causal consistency, thereby enhancing agent reliability and learning efficiency.

Abstract

Large language model (LLM) agents frequently fail on multi-step tasks involving reasoning, tool use, and environment interaction. While such failures are typically logged or retried heuristically, they contain structured signals about where execution broke down. We introduce CausalFlow, an interventional framework that converts failed agent traces into minimal counterfactual repairs and reusable supervision. CausalFlow models execution traces as sequential chains of dependent steps and computes Causal Responsibility Scores(CRS) via step-level counterfactual intervention to identify failure-inducing steps. For these steps, we generate minimally edited repairs that flip the final outcome to success, producing validated contrastive pairs of the form (wrong step, corrected step). CausalFlow supports two complementary uses: targeted test-time repair that recovers from failures with minimal behavioral drift, and training-time supervision suitable for offline preference optimization or reward modeling. Across four benchmarks spanning mathematical reasoning, code generation, question answering, and medical browsing, CausalFlow converts failed executions into validated minimal repairs with high minimality and causal-consensus scores, and demonstrates that causal attribution is necessary for reliable improvement across diverse agent tasks, outperforming heuristic refinement in complex retrieval settings while producing more localized repairs throughout. These results demonstrate that interventional analysis over structured execution traces provides a principled and scalable mechanism for transforming agent failures into reliability gains and learning-ready supervision.

Problem

• LLM agent failures
• causal attribution
• counterfactual repair
• execution traces
• multi-step tasks

Innovation

• Causal Attribution
• Counterfactual Repair
• Causal Responsibility Score
• LLM Agent Failure
• Interventional Analysis