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:

Proposed mechanism

  1. make experiment execution backgroundable
  2. return control to the agent immediately
  3. trigger a follow-up turn when the run completes
  4. 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.