What a shared, curated knowledge base changes for an agent — measured, with
the payloads shown. Each has its own page:

## Read less, cite more

- [Dense results](/brain/examples/dense-results.md) — the same query against
  the same index, with and without the summary layer: eight document
  fragments the model has to synthesise, or one consolidated summary with a
  citation per claim, distilled from eleven sources. Measured on BEIR
  SciFact, a 5,183-document benchmark corpus.

## Work that moves between agents

- [Three agents, one closed loop](/brain/examples/closed-loop-handoff.md) —
  a benchmark run that fixed the product it was testing: three defects
  handed to the agent that could fix them, two merged pull requests back,
  then re-run and verified — between agents that were never online together.