# State Representation Shapes Long-Horizon Agent Learning and Transfer: An Empirical Study

arXiv: https://arxiv.org/abs/2610.15407 · HTML: https://mlsanity.com/paper/2610.15407/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 23 references that are in the map.
- First posted on arXiv 11 Oct 2026.

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

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering](https://mlsanity.com/paper/2405.15793/) (2024)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [Executable Code Actions Elicit Better LLM Agents](https://mlsanity.com/paper/2402.01030/) (2024)
- [AgentTuning: Enabling Generalized Agent Abilities for LLMs](https://mlsanity.com/paper/2310.12823/) (2023)
- [ScienceWorld: Is your Agent Smarter than a 5th Grader?](https://mlsanity.com/paper/2203.07540/) (2022)
- [HiAgent: Hierarchical Working Memory Management for Solving Long-Horizon Agent Tasks with Large Language Model](https://mlsanity.com/paper/2408.09559/) (2024)
- [Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Tasks](https://mlsanity.com/paper/2510.12635/) (2025)
- [Towards a Science of AI Agent Reliability](https://mlsanity.com/paper/2602.16666/) (2026)
- [The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Management](https://mlsanity.com/paper/2508.21433/) (2025)
- [How Is LLM Reasoning Distracted by Irrelevant Context? An Analysis Using a Controlled Benchmark](https://mlsanity.com/paper/2505.18761/) (2025)
- [Imitation Learning for Multi-turn LM Agents via On-policy Expert Corrections](https://mlsanity.com/paper/2512.14895/) (2025)
- [When Tools Fail: Benchmarking Dynamic Replanning and Anomaly Recovery in LLM Agents](https://mlsanity.com/paper/2606.05806/) (2026)
- [CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents](https://mlsanity.com/paper/2607.05378/) (2026)
- [Toward Reliable Context Compression for Long-Horizon Agents: An Empirical Study of Execution Instability](https://mlsanity.com/paper/2608.06503/) (2026)
- [Hell or High Water: Evaluating Agentic Recovery from External Failures](https://mlsanity.com/paper/2508.11027/) (2025)
- [On Training Large Language Models for Long-Horizon Tasks: An Empirical Study of Horizon Length](https://mlsanity.com/paper/2605.02572/) (2026)
- [The Devil Is in the Interface: Evaluating How Tool Architecture Shapes Coding Agent Behavior](https://mlsanity.com/paper/2608.11386/) (2026)
- [MUDDLE: Measuring Understanding of Documents under Distractor and Length Effects](https://mlsanity.com/paper/2608.29477/) (2026)

## Abstract (arXiv)

An end-to-end success rate alone cannot distinguish failure to learn an action from failure to apply it as task state changes. We study this distinction in deterministic tool workflows that provide the current artifact state and candidate action recipes. Using 1.7B, 8B, and 27B Qwen models, we compare supervised learning across workflow horizons and cross the state representation used for training with the one used at evaluation. Horizon-matched supervision yields high completion, but an H4-trained 1.7B model transfers poorly to longer workflows under full-state observations. In an H16 crossover, compact-trained 1.7B adapters complete 50/50 held-out tasks in compact view and 0/50 in full view for each of three paired training seeds; full-trained outcomes vary across seeds. A separate 8B adapter pair shows a similar asymmetry. The view dependence also appears in paired-choice tasks where both candidate calls are executable and the correct choice depends on table contents; here, the compact mask does not consult the target action. An explicit relevance cue improves full-view execution but does not close the gap for the tested checkpoint. In a separate 27B ScienceWorld study, compact-state training improves local accuracy under both views, indicating that the transfer pattern depends on the task and observation construction. Together, these experiments identify representation-sensitive transfer as a bottleneck in the controlled workflows, without isolating content reduction from other changes to the observation.