Re-run the crew on a failed job
Replay the original request on a failed job.
Returns the reset proposal. Poll for completion the same way as the initial submit.
Authorizations
The access token received from the authorization server in the OAuth 2.0 flow.
Path Parameters
Feature engineering job ID.
Response
Successful Response
The crew's proposal returned to the user for review.
Lifecycle of an AutoML feature engineering job.
running, pending_approval, failed, approved, materializing, complete Structured recipe describing the training table the crew designed.
The CREATE TABLE AS SELECT query the crew produced.
Result of running the CTAS query with a row limit during validation.
Kept intentionally lean. Large agents produce more than 10 KB of JSON when this grows with per-column null rates and sample rows. That often exceeds output-token limits. Run the SQL directly to inspect rows. Only persist what's cheap and structurally useful.

