LatentSift: Policy-State Filtering for Token-Efficient Verification of Software Engineering Agents

arXiv 2609.36371 ↗First posted 28 Sep 2026cs.SE cs.AIcited by 0

On the map: LLM reasoning → Reasoning for agents, planning and code

also in 1 more

LLM agents and tool use → Software engineering coding agents

Abstract (arXiv)

Test-time scaling improves software engineering agents by generating multiple candidate trajectories and selecting the best one. Verifying and selecting among these long interactions can consume as many tokens as generation itself. Existing hybrid workflows first apply an LLM-based execution-free (EF) verifier to filter candidates before running tests, which adds another model pass over every trajectory. We introduce LatentSift, a token-free and execution-free filter that replaces this first stage with hidden states the policy already produces while generating the candidates. It represents each candidate through its reasoning, observation, and function-call states, compares them with positive and negative banks of such states collected from successful and unsuccessful trajectories during policy training, and fuses the resulting distance scores with a learned linear score to retain promising candidates for the execution-based stages. On SWE-bench Verified, across three agents and two policy sizes, LatentSift cuts EF-verifier tokens by 66.6--81.0% and total verification tokens, which include test generation, by 49.1--62.1% at K=16, while hybrid Best@16 matches or improves on each agent's reference workflow, rising from 59.26% to 60.06% on DeepSWE-Preview.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 52 of its 57 references that are in the map. First posted on arXiv 28 Sep 2026.

Builds on

52 of 57 references.

Closest ancestors · 9

Same direction, newest first

Other references · 43

2021Training Verifiers to Solve Math Word Problemscited by 9,743
2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
2023Let's Verify Step by Stepcited by 3,598
2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,155
2022Language Models (Mostly) Know What They Knowcited by 1,783

31 more

2024SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineeringcited by 1,628
2025s1: Simple test-time scalingcited by 1,311
2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,259
2022Scaling Laws for Reward Model Overoptimizationcited by 1,056
2023Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotationscited by 895
2022Solving math word problems with process- and outcome-based feedbackcited by 871
2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 733
2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 673
2023The Internal State of an LLM Knows When It's Lyingcited by 644
2024RewardBench: Evaluating Reward Models for Language Modelingcited by 480
2025The Lessons of Developing Process Reward Models in Mathematical Reasoningcited by 402
2024Training Software Engineering Agents and Verifiers with SWE-Gymcited by 300
2024Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Modelscited by 282
2024Semantic Entropy Probes: Robust and Cheap Hallucination Detection in LLMscited by 246
2024LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinationscited by 230
2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verificationcited by 154
2024SWE-Search: Enhancing Software Agents with Monte Carlo Tree Search and Iterative Refinementcited by 124
2023Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMscited by 111
2024Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoningcited by 105
2023Let's reward step by step: Step-Level reward model as the Navigators for Reasoningcited by 96
2025CWM: An Open-Weights LLM for Research on Code Generation with World Modelscited by 87
2025Process Reward Models for LLM Agents: Practical Framework and Directionscited by 81
2024Process Reward Model with Q-Value Rankingscited by 78
2025AgentPRM: Process Reward Models for LLM Agents via Step-Wise Promise and Progresscited by 69
2025Scaling Test-time Compute for LLM Agentscited by 59
2025Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distributioncited by 24
2026Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probingcited by 13
2025Can LLMs Predict Their Own Failures? Self-Awareness via Internal Circuitscited by 12
2025Mining Intrinsic Rewards from LLM Hidden States for Efficient Best-of-N Samplingcited by 9
2026SWE-Shepherd: Advancing PRMs for Reinforcing Code Agentscited by 4
2025Word Salad Chopper: Reasoning Models Waste A Ton Of Decoding Budget On Useless Repetitions, Self-Knowinglycited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way