# Trace2Policy: From Expert Behavior Traces to Self-Evolving Decision Agents

arXiv: https://arxiv.org/abs/2606.10457 · HTML: https://mlsanity.com/paper/2606.10457/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 22 of its 34 references that are in the map.
- First posted on arXiv 9 Jun 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)
- Synthetic data and self-improvement for LLMs / [Self-evolving LLM agents](https://mlsanity.com/synth/self-evolving-llm-agents/)

## Builds on (most cited first, 22 of 22)

- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Self-Refine: Iterative Refinement with Self-Feedback](https://mlsanity.com/paper/2303.17651/) (2023)
- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https://mlsanity.com/paper/2305.16291/) (2023)
- [Large Language Models as Optimizers](https://mlsanity.com/paper/2309.03409/) (2023)
- [DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines](https://mlsanity.com/paper/2310.03714/) (2023)
- [CogAgent: A Visual Language Model for GUI Agents](https://mlsanity.com/paper/2312.08914/) (2023)
- [UI-TARS: Pioneering Automated GUI Interaction with Native Agents](https://mlsanity.com/paper/2501.12326/) (2025)
- [GPT-4V(ision) is a Generalist Web Agent, if Grounded](https://mlsanity.com/paper/2401.01614/) (2024)
- [The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search](https://mlsanity.com/paper/2504.08066/) (2025)
- [LegalBench: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Models](https://mlsanity.com/paper/2308.11462/) (2023)
- [Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs](https://mlsanity.com/paper/2406.11695/) (2024)
- [TextGrad: Automatic "Differentiation" via Text](https://mlsanity.com/paper/2406.07496/) (2024)
- [DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning](https://mlsanity.com/paper/2406.11896/) (2024)
- [AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials](https://mlsanity.com/paper/2412.09605/) (2024)
- [SoK: Agentic Skills -- Beyond Tool Use in LLM Agents](https://mlsanity.com/paper/2602.20867/) (2026)
- [Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward](https://mlsanity.com/paper/2602.12430/) (2026)
- [Retroformer: Retrospective Large Language Agents with Policy Gradient Optimization](https://mlsanity.com/paper/2308.02151/) (2023)
- [A Survey on Self-Evolution of Large Language Models](https://mlsanity.com/paper/2404.14387/) (2024)
- [AgentRefine: Enhancing Agent Generalization through Refinement Tuning](https://mlsanity.com/paper/2501.01702/) (2025)
- [Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents](https://mlsanity.com/paper/2604.15877/) (2026)
- [Auditable Agents](https://mlsanity.com/paper/2604.05485/) (2026)
- [Automating Skill Acquisition through Large-Scale Mining of Open-Source Agentic Repositories: A Framework for Multi-Agent Procedural Knowledge Extraction](https://mlsanity.com/paper/2603.11808/) (2026)

## Built on by (most cited first, 1 of 1)

- [Beyond Aggregate Accuracy: Item-Level Transitions in Prompt Optimization](https://mlsanity.com/paper/2610.19084/) (2026)

## Abstract (arXiv)

Decision rules that enterprise experts apply tacitly -- in auditing, compliance, and contract review -- can be systematically recovered and improved through iterative error analysis. We present Trace2Policy, whose core mechanism -- EISR (Error-driven Iterative Skill Refinement) -- maintains a human-readable rule document as its optimization target: each round executes the rules on a validation set, clusters errors by root cause into MISSING, WRONG, or CONFLICT types, applies targeted patches, and commits only those that pass a regression gate. For this class of compliance-sensitive, skewed-base-rate decision tasks, we identify rule quality -- not model capability -- as the dominant performance lever: across five LLMs, one-shot distillation plateaus near ~70% on the deployed pool, while eight EISR rounds lift the same rules to 79.6% when compiled into deterministic Python -- zero LLM calls at inference. Execution form compounds the gain: in production, the same EISR-refined content runs 9.8~pp higher as compiled Python than as an LLM prompt, a form-and-engineering bundle the 22-day deployment matured together. Deployed for 22 days at a major logistics carrier (3,349 audit cases), the compiled pipeline outperforms the pure-LLM baseline it replaced (72.7%); on these calibrated, skewed-base-rate workloads, re-enabling LLM fallback monotonically degrades accuracy. An LLM-driven variant, Auto-EISR, reproduces this refinement at 5--10 per cycle versus ~70 expert-hours, and transfers to four public benchmarks spanning legal reasoning (LegalBench) and process-mining decisions (BPIC 2012) without re-engineering.