How do I enable introspection and long-term learning?
Give the factory enough durable evidence to study its own behavior and improve future plans, dispatches, and environments.
Questions the system should repeatedly ask
- Which models are good for what?
- What jobs should run on macOS, Linux, local accelerators, or cloud infrastructure?
- What is the value of building a test suite?
- How much planning is necessary for a good result?
- Should a plan target a specific model, a model tier, or any capable worker?
Evidence to preserve
Plans, searches, tool calls, failed paths, review comments, runtimes, costs, and outcomes should remain available for later analysis.
Living record
This page will track how Yesod turns retained execution history into better future behavior.