Goal
Avoid wasting long benchmark wall-clock time by letting research and synthesis continue while experiments run in the background.
Motivation
A blocking benchmark loop is fine for simple automation, but it is bad for a research agent that could be doing useful work during waits:
- synthesizing paper notes
- expanding hypotheses
- turning ideas into lane summaries
Proposed mechanism
- make experiment execution backgroundable
- return control to the agent immediately
- trigger a follow-up turn when the run completes
- spend the waiting interval on knowledge-base maintenance or research
Why it matters
This is not an LLM idea. It is a research-operations idea that increases time-to-insight by reducing idle time, so it belongs in meta.