An application needs to poll an asynchronous team run rather than immediately start another duplicate run. Which capability should it use?

1. Infer completion from the number of schema tables
2. Regenerate all vector embeddings
3. The framework's team-state inspection APIs
4. Recreate all agents on every poll