get_job_request

Read back the request a document operation ran under. Cost class: fast.

What a finished run was asked for — the page selection, the schema, the targets — so a caller can show the settings that produced a result, rebuild the call that made it, or submit it again. Job.source names the document; this names everything else.

Its own route rather than a field on Job: an extract schema or a ground target list is large, and GET /v1/jobs returns up to 200 rows.

Retention is the same as the result's — the request is read from the stored payload, so it dies with it rather than outliving it.

Args: session: Database session. caller: The authenticated tenant. job_id: The job whose request to read.

Returns: The stored request, tagged with request_type (always equal to the job's kind).

Raises: PlatformError: job_not_found when no such job exists for this key; invalid_request for a kind that carries no single document request; result_expired once the payload is gone.