# Verified Critical Step Optimization for LLM Agents

arXiv: https://arxiv.org/abs/2602.03412 · HTML: https://mlsanity.com/paper/2602.03412/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use.
- Builds on 25 of its 43 references that are in the map.
- First posted on arXiv 3 Feb 2026.

## Placed in

- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)

## Builds on (most cited first, 25 of 25)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://mlsanity.com/paper/2403.13372/) (2024)
- [GAIA: a benchmark for General AI Assistants](https://mlsanity.com/paper/2311.12983/) (2023)
- [Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning](https://mlsanity.com/paper/2506.01939/) (2025)
- [Group-in-Group Policy Optimization for LLM Agent Training](https://mlsanity.com/paper/2505.10978/) (2025)
- [Reasoning with Exploration: An Entropy Perspective](https://mlsanity.com/paper/2506.14758/) (2025)
- [WebSailor: Navigating Super-human Reasoning for Web Agent](https://mlsanity.com/paper/2507.02592/) (2025)
- [R-Zero: Self-Evolving Reasoning LLM from Zero Data](https://mlsanity.com/paper/2508.05004/) (2025)
- [Trial and Error: Exploration-Based Trajectory Optimization for LLM Agents](https://mlsanity.com/paper/2403.02502/) (2024)
- [WebDancer: Towards Autonomous Information Seeking Agency](https://mlsanity.com/paper/2505.22648/) (2025)
- [WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning](https://mlsanity.com/paper/2505.16421/) (2025)
- [WebShaper: Agentically Data Synthesizing via Information-Seeking Formalization](https://mlsanity.com/paper/2507.15061/) (2025)
- [xbench: Tracking Agents Productivity Scaling with Profession-Aligned Real-World Evaluations](https://mlsanity.com/paper/2506.13651/) (2025)
- [Watch Every Step! LLM Agent Learning via Iterative Step-Level Process Refinement](https://mlsanity.com/paper/2406.11176/) (2024)
- [Process Reward Models for LLM Agents: Practical Framework and Directions](https://mlsanity.com/paper/2502.10325/) (2025)
- [MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling](https://mlsanity.com/paper/2511.11793/) (2025)
- [Harnessing Uncertainty: Entropy-Modulated Policy Gradients for Long-Horizon LLM Agents](https://mlsanity.com/paper/2509.09265/) (2025)
- [Cognitive Kernel-Pro: A Framework for Deep Research Agents and Agent Foundation Models Training](https://mlsanity.com/paper/2508.00414/) (2025)
- [Guided Self-Evolving LLMs with Minimal Human Supervision](https://mlsanity.com/paper/2512.02472/) (2025)
- [Towards Self-Improvement of LLMs via MCTS: Leveraging Stepwise Knowledge with Curriculum Preference Learning](https://mlsanity.com/paper/2410.06508/) (2024)
- [Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification](https://mlsanity.com/paper/2601.15808/) (2026)
- [WebCoT: Enhancing Web Agent Reasoning by Reconstructing Chain-of-Thought in Reflection, Branching, and Rollback](https://mlsanity.com/paper/2505.20013/) (2025)
- [Teaching Your Models to Understand Code via Focal Preference Alignment](https://mlsanity.com/paper/2503.02783/) (2025)
- [Co-Evolving Agents: Learning from Failures as Hard Negatives](https://mlsanity.com/paper/2511.22254/) (2025)

## Built on by (most cited first, 14 of 14)

- [How to Allocate, How to Learn? Dynamic Rollout Allocation and Advantage Modulation for Policy Optimization](https://mlsanity.com/paper/2602.19208/) (2026)
- [Exact Is Easier: Credit Assignment for Cooperative LLM Agents](https://mlsanity.com/paper/2603.06859/) (2026)
- [SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks](https://mlsanity.com/paper/2604.08865/) (2026)
- [Credit Without Ground Truth: Auditing Step-Level Credit Assignment in LLM Agents Against Executed Replay](https://mlsanity.com/paper/2608.19760/) (2026)
- [Measure Twice, Click Once: Co-evolving Proposer and Visual Critic via Reinforcement Learning for GUI Grounding](https://mlsanity.com/paper/2604.21268/) (2026)
- [ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment](https://mlsanity.com/paper/2608.05102/) (2026)
- [SciResearcher: Scaling Deep Research Agents for Frontier Scientific Reasoning](https://mlsanity.com/paper/2605.01489/) (2026)
- [Mitigating Context Interference for Reliable and Efficient Search Agents](https://mlsanity.com/paper/2608.10743/) (2026)
- [Placing Puzzle Pieces Where They Matter: A Question Augmentation Framework for Reinforcement Learning](https://mlsanity.com/paper/2604.15830/) (2026)
- [CRISP: Critical Step Perception for Training Efficient Deep Search Agents](https://mlsanity.com/paper/2608.01867/) (2026)
- [Behavior Quotient Learning for Low-Rank Adaptation of LLM Agents](https://mlsanity.com/paper/2609.12896/) (2026)
- [Critical-State RL: Diagnosing Trainable States for Multi-Turn Tool Use](https://mlsanity.com/paper/2609.24985/) (2026)
- [The Tasteful Agent: Measuring and Improving Taste in Long-Horizon Tasks](https://mlsanity.com/paper/2609.25804/) (2026)
- [DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter](https://mlsanity.com/paper/2610.17666/) (2026)

## Abstract (arXiv)

As large language model agents tackle increasingly complex long-horizon tasks, effective post-training becomes critical. Prior work faces fundamental challenges: outcome-only rewards fail to precisely attribute credit to intermediate steps, estimated step-level rewards introduce systematic noise, and Monte Carlo sampling approaches for step reward estimation incur prohibitive computational cost. Inspired by findings that only a small fraction of high-entropy tokens drive effective RL for reasoning, we propose Critical Step Optimization (CSO), which focuses preference learning on verified critical steps, decision points where alternate actions demonstrably flip task outcomes from failure to success. Crucially, our method starts from failed policy trajectories rather than expert demonstrations, directly targeting the policy model's weaknesses. We use a process reward model (PRM) to identify candidate critical steps, leverage expert models to propose high-quality alternatives, then continue execution from these alternatives using the policy model itself until task completion. Only alternatives that the policy successfully executes to correct outcomes are verified and used as DPO training data, ensuring both quality and policy reachability. This yields fine-grained, verifiable supervision at critical decisions while avoiding trajectory-level coarseness and step-level noise. Experiments on GAIA-Text-103 and XBench-DeepSearch show that CSO achieves 37% and 26% relative improvement over the SFT baseline and substantially outperforms other post-training methods, while requiring supervision at only 16% of trajectory steps. This demonstrates the effectiveness of selective verification-based learning for agent post-training.