Retry Last Failed Alignment Project
Retry a failed alignment for the authenticated user.
Query Parameters:
alignment_id(optional): Retry this specific failed alignment. If omitted, the endpoint picks the user’s most recently FAILED alignment that has not already been retried.
Request Body (optional):
base_model(optional): Override the failed alignment’s base model when starting the retry. If omitted (or the body is omitted entirely), the failed row’sbase_modelis reused.
The endpoint reuses the failed row’s synthetic_data_id, name, and
description to start a new alignment — the same path
/create-alignment takes, just with the inputs pulled from the prior
failed row instead of a request body. base_model can be overridden via
the request body; otherwise it’s reused as well.
Once a retry is created, the original failed row is stamped with
metadata.retried_by = <new_alignment_id> so it won’t be retried again,
whether picked implicitly or referenced explicitly.
Returns:
alignment_id: ID of the newly created alignment task.status: AlwaysPROCESSING.
Raises:
404: No failed alignment exists for this user, or the explicitalignment_idis not owned by the user.400: The explicitalignment_idis not in FAILED state, or the base model no longer supports alignment.409: The failure is not retryable — already retried, failed during synthetic data generation, has nosynthetic_data_idto reuse, or the source synthetic data is no longer completed/accessible.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Query Parameters
Body
Optional request body for /retry-last-failed.
alignment_id targets a specific failed alignment. When omitted, the
most recently failed (unretried) alignment is used.
base_model overrides the failed alignment's original base model. When
omitted (or the body is omitted entirely), the failed row's base_model is
reused as before.
Response
Successful Response