For literature surveys or project note management in a materials laboratory, it may be worth testing a structure that creates topic-specific folders and accumulates conditions, results, and remarks extracted from each paper at the file level. This would allow the agent to continue across literature collection or hypothesis organization that spans several days without losing prior context.
This approach may be useful especially when organizing conflicting measurements or missing conditions, since the file system can be used as a kind of living research notebook. However, the schema of materials data changes frequently, so it is important to balance a human-readable notebook with a machine-readable structured format.