mlsanityData to 14 Oct 2026

Does an Agent's History Tell You When Compaction Will Hurt? A Modest, Bounded Effect on the TRACE Paired-Replay Corpus

arXiv 2610.08722 ↗First posted 6 Oct 2026cs.AI cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Many long-horizon agents compact their context on a global rule, usually a token budget, blind to what the agent was doing. We ask whether the agent's recent behaviour predicts when a compaction will hurt. TRACE's public corpus of 590 harness-triggered AppWorld compaction boundaries replays each boundary from a re-executed prefix state under the pre-compaction context and under the summary, and records the burden of the next actions: calls that error or repeat a call already made. We find that pre-boundary history predicts post-compaction harm only weakly. An internally prespecified contrast by prefix placement is a wide null, and the naive "has-written" label behind it turns out to measure trajectory phase. The best extension-protocol trigger reaches held-out AUROC 0.66 (0.64 on the replicate's own label) against a same-boundary replicate of 0.72; the best frozen, interpretable trigger avoids 21% of harmful (positive-burden) boundaries while keeping 84% of compaction opportunities, and exceeds the random-rule expectation on count but not on burden mass (a post hoc comparison). Whether the best trigger beats a token-budget rule at matched retention cannot be evaluated on the release. We state what corpora should ship to answer it.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 19 of its 28 references that are in the map. First posted on arXiv 6 Oct 2026.

Builds on

19 of 28 references.

References · 19

  1. 2023MemGPT: Towards LLMs as Operating Systemscited by 1,389
  2. 2017Selective Classification for Deep Neural Networkscited by 852
  3. 2024AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agentscited by 339
  4. 2025Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systemscited by 185
  5. 2025Scaling Long-Horizon LLM Agent via Context-Foldingcited by 148
  6. 2025ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarizationcited by 141
  7. 2025ACON: Optimizing Context Compression for Long-horizon LLM Agentscited by 136
  8. 2024Non-Determinism of "Deterministic" LLM Settingscited by 92
  9. 2025Memory as Action: Autonomous Context Curation for Long-Horizon Agentic Taskscited by 66
  10. 2023Recursively Summarizing Enables Long-Term Dialogue Memory in Large Language Modelscited by 62
  11. 2025Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inferencecited by 53
  12. 2025The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Managementcited by 48
7 more
  1. 2025Context as a Tool: Context Management for Long-Horizon SWE-Agentscited by 43
  2. 2026SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?cited by 23
  3. 2026Inside the Scaffold: A Source-Code Taxonomy of Coding Agent Architecturescited by 18
  4. 2026CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agentscited by 14
  5. 2026How to Interpret Agent Behaviorcited by 10
  6. 2026Toward Reliable Context Compression for Long-Horizon Agents: An Empirical Study of Execution Instabilitycited by 10
  7. 2025AI Agent Behavioral Sciencecited by 6

Built on by

Not cited yet.