mlsanityData to 14 Oct 2026

Visible but Not Avoided: State-Dependent Re-Proposal of Rejected Candidates in Catalog-Based LLM Optimization Loops

arXiv 2610.15738 ↗First posted 11 Oct 2026cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

LLM optimization loops often show the proposer the candidates it has already tried and rejected, but leave it to the proposer to avoid them. We ask whether avoidance measured in a static probe predicts avoidance in the states an optimizer's own trajectory produces. In catalog-based loops where three open LLMs improve a GUI agent by naming one of 40 edits, a re-proposal is identified exactly by the edit's name. Clean base-state probes underpredict re-proposal in the proposers' own loops (macro-averaged loop-minus-probe rate 0.32 [0.27, 0.37]), and proposals of already evaluated harness-edit pairs consume 28-63% of valid proposer iterations. The two settings reject different edits: a probe's rejected edits are mostly ones the proposer did not choose, whereas a loop's are its own earlier choices, which remain preferred when the history is removed. Ordinary history lowers re-proposal for two proposers but raises it for the third, and the effects of failure evidence are model-specific. Stating the rejection status in one sentence lowers re-proposal in every model (-0.16 [-0.22, -0.09] relative to a neutral list of the same IDs). For catalog-based loops with exact edit identities, rejection avoidance should be measured on the rejected sets the loop produces.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 39 of its 62 references that are in the map. First posted on arXiv 11 Oct 2026.

Builds on

39 of 62 references.

References · 39

  1. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  2. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  3. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  4. 2019The Curious Case of Neural Text Degenerationcited by 2,916
  5. 2025gpt-oss-120b & gpt-oss-20b Model Cardcited by 1,386
  6. 2024RULER: What's the Real Context Size of Your Long-Context Language Models?cited by 1,346
  7. 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 1,017
  8. 2022In-context Learning and Induction Headscited by 964
  9. 2023ExpeL: LLM Agents Are Experiential Learnerscited by 876
  10. 2025AlphaEvolve: A coding agent for scientific and algorithmic discoverycited by 867
  11. 2023Large Language Models Can Be Easily Distracted by Irrelevant Contextcited by 841
  12. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 832
27 more
  1. 2023Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formattingcited by 817
  2. 2023Automatic Prompt Optimization with "Gradient Descent" and Beam Searchcited by 653
  3. 2023Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolutioncited by 551
  4. 2025GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learningcited by 542
  5. 2025Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agentscited by 356
  6. 2024Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programscited by 291
  7. 2024When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction of LLMscited by 240
  8. 2023EvoPrompt: Connecting LLMs with Evolutionary Algorithms Yields Powerful Prompt Optimizerscited by 193
  9. 2025ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolutioncited by 174
  10. 2026SkillOpt: Executive Strategy for Self-Evolving Agent Skillscited by 148
  11. 2022Learning to Break the Loop: Analyzing and Mitigating Repetitions for Neural Text Generationcited by 94
  12. 2024Can large language models explore in-context?cited by 76
  13. 2022Can Large Language Models Truly Understand Prompts? A Case Study with Negated Promptscited by 59
  14. 2025When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMscited by 53
  15. 2024Suppressing Pink Elephants with Direct Principle Feedbackcited by 15
  16. 2024Revisiting OPRO: The Limitations of Small-Scale LLMs as Optimizerscited by 15
  17. 2025No Need for Explanations: LLMs can implicitly learn from mistakes in-contextcited by 13
  18. 2025ReasonIF: Large Reasoning Models Fail to Follow Instructions During Reasoningcited by 13
  19. 2026Governance Decay: How Context Compaction Silently Erases Safety Constraints in Long-Horizon LLM Agentscited by 8
  20. 2026What Makes an LLM a Good Optimizer? A Trajectory Analysis of LLM-Guided Evolutionary Searchcited by 5
  21. 2026Omission Constraints Decay While Commission Constraints Persist in Long-Context LLM Agentscited by 5
  22. 2026Why Retrying Fails: Context Contamination in LLM Agent Pipelinescited by 4
  23. 2026Feedback That Backfires: Why Small Language Model Agents Repeat the Call They Just Watched Failcited by 4
  24. 2026Models Recall What They Violate: Constraint Adherence in Multi-Turn LLM Ideationcited by 3
  25. 2026MemArbiter: Decision-Time Memory Arbitration for Long-Horizon LLM Agentscited by 3
  26. 2026FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applicationscited by 2
  27. 2026Semantic Gravity Wells: Why Negative Constraints Backfirecited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Prior Knowledge or Search? A Study of LLM Agents in Hardware-Aware Code Optimization
  2. 2026Plans Don't Persist: Why Context Management Is Load Bearing for LLM Agents
  3. 2026SiLR: Structure-Preserving Admission and Process Reward for LLM Tool Agents
  4. 2026VERSE: Verified Self-Evolving Optimizer for Agent Harnesses
  5. 2026Reflective Prompted Policy Optimization: Trajectory-Grounded Revision and Salience Bias
  6. 2026Evolution or Illusion? Rethinking Evaluation in LLM Evolutionary Search
  7. 2026Beyond Token Savings: A Systematic Study of Context Compression in LLM Agents
  8. 2026PAIR: Prefix-Aware Internal Reward Model for Multi-Turn Agent Optimization