# Causal Agent Replay: Counterfactual Attribution for LLM-Agent Failures

arXiv: https://arxiv.org/abs/2606.08275 · HTML: https://mlsanity.com/paper/2606.08275/

## Key facts

- Cited by 20 papers in the map (data through 14 Oct 2026).
- Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use.
- Builds on 6 of its 13 references that are in the map.
- First posted on arXiv 6 Jun 2026.

## Placed in

- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)

## Builds on (most cited first, 6 of 6)

- [Counterfactual Multi-Agent Policy Gradients](https://mlsanity.com/paper/1705.08926/) (2017)
- [Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systems](https://mlsanity.com/paper/2505.00212/) (2025)
- [Counterfactual Credit Assignment in Model-Free Reinforcement Learning](https://mlsanity.com/paper/2011.09464/) (2020)
- [Agent Incentives: A Causal Perspective](https://mlsanity.com/paper/2102.01685/) (2021)
- [Discovering Agents](https://mlsanity.com/paper/2208.08345/) (2022)
- [Engineering Record And Replay For Deployability: Extended Technical Report](https://mlsanity.com/paper/1705.05937/) (2017)

## Built on by (most cited first, 20 of 20)

- [Autonomous Research Agents: A Survey of AI Scientists and the Verification Gap](https://mlsanity.com/paper/2608.05179/) (2026)
- [Credit Without Ground Truth: Auditing Step-Level Credit Assignment in LLM Agents Against Executed Replay](https://mlsanity.com/paper/2608.19760/) (2026)
- [Long-Horizon Agent Trajectory Attribution: A Unified Benchmark and Fine-Grained Annotation Framework](https://mlsanity.com/paper/2608.06909/) (2026)
- [Influence Is Not Authority: When Causal Guardrail Signals Make Legitimate Tool Use Look Like an Attack in Tool-Using LLM Agents](https://mlsanity.com/paper/2608.29942/) (2026)
- [What if LLMs Ate Their Words: Causal History Effects in Multi-Turn Interaction](https://mlsanity.com/paper/2609.05882/) (2026)
- [AUDITA: certified auditing and causal attribution of adverse outcomes in autonomous multi-agent systems](https://mlsanity.com/paper/2608.22160/) (2026)
- [HANSARD: A Reference Architecture for Forensic Readiness, Runtime Witnessing, and Graded Attribution in Autonomous Multi-Agent AI Systems](https://mlsanity.com/paper/2608.22512/) (2026)
- [Localizing Emergent Failures in Agentic AI: Recovering Minimal Repair Families via Counterfactual Replay](https://mlsanity.com/paper/2608.29228/) (2026)
- [Causal Attribution for Agentic Decisions: Estimators, Coupling, and a Traceability Specification](https://mlsanity.com/paper/2609.06445/) (2026)
- [Chronicle: Cut-Point Replay for Regression Testing of LLM Agents](https://mlsanity.com/paper/2609.20625/) (2026)
- [CausalLoss-Fin: Attributing Financial-Agent Loss to Decisions and Infrastructure Faults](https://mlsanity.com/paper/2609.25960/) (2026)
- [DAAF: From Failure Localization to Editable System Assets in LLM Agents](https://mlsanity.com/paper/2609.32498/) (2026)
- [ReplayLens: Auditing Agents' Use of Outcomes](https://mlsanity.com/paper/2609.34177/) (2026)
- [Same Winners, Different Success Rates: Evaluating How LLM Agents Recover from Failures](https://mlsanity.com/paper/2609.34215/) (2026)
- [Beyond Skill Evolution: Self-Evolving Context Management Policies for Long-Horizon Agent Harnesses](https://mlsanity.com/paper/2609.34649/) (2026)
- [EvoSteer: Online Self-Evolving Graph Orchestration via Reference-Anchored Credit Assignment](https://mlsanity.com/paper/2609.38661/) (2026)
- [Where MLLMs Fail and Why: Causal Task Decomposition for Capability Failure Diagnosis](https://mlsanity.com/paper/2609.38851/) (2026)
- [When Harnesses Lose the Signal: Causal Evaluation of Recovery in LLM Agents](https://mlsanity.com/paper/2610.00372/) (2026)
- [Do Tool Calls Execute as Intended? Measuring and Repairing Intent-Execution Correspondence in LLM Agents](https://mlsanity.com/paper/2610.04375/) (2026)
- [RippleCP: Measuring Counterfactual Checkpoint Advantage in Coding Agents](https://mlsanity.com/paper/2610.09088/) (2026)

## Abstract (arXiv)

When an LLM agent fails -- issues a refund it should not have, calls the wrong tool, leaks data -- existing tooling answers what happened (observability) or whether it passed (evaluation), but not which step caused the failure. The obvious heuristics are wrong: the step that executes the harmful action is usually not the step that decided on it, and LLM-judge attribution is correlational and unreliable (state-of-the-art step-level accuracy on the Who&When benchmark is about 14%). We present Causal Agent Replay (CAR), which answers the question by intervention: it models an agent run as a structural causal model, applies a do-operation to a step, and re-executes the trajectory forward under the same stochastic policy, measuring the shift in the outcome distribution. We define an intervention algebra over agent steps, a single-step contrastive estimator whose point-of-commitment rule resolves a confound specific to stochastic run-forward, and a budget-bounded Monte-Carlo Shapley estimator that splits credit across interacting steps. Every effect is reported with confidence intervals. We validate against synthetic structural causal models with planted ground truth: the contrastive estimator recovers the pivotal step, and Shapley recovers a two-step interaction (0.44, 0.45, ~0; efficiency sum 0.909 versus the analytic 0.91). CAR is open source and runs on hosted or free local models.