AI RESEARCH
Evaluating Memory Condensation Strategies for Coding Agents in Data-Driven Scientific Discovery
arXiv CS.LG
•
ArXi:2605.18854v1 Announce Type: new Coding agents accumulate extensive context during long-running tasks, yet fixed context windows force practitioners to choose between truncation and task failure. While numerous memory condensation strategies have been proposed, from simple sliding windows to LLM-generated summaries, no systematic comparison exists to guide strategy selection, especially in scientific discovery tasks. We evaluate eight memory condensation strategies using GPT-4o on sixty DiscoveryBench tasks spanning six scientific domains (480 total evaluations.