What should an AI lab notebook remember?

A discussion on AI lab notebooks highlights the need for tools that remember more than just final results. It suggests that a useful AI research assistant should track data versions, assumptions, parameter changes, and previous attempts, not just successful runs. This detailed experiment history would preserve the path to results, preventing the AI from merely providing polished answers without context. The question is posed whether such a detailed AI assistant would be trusted or create too much overhead.