List Datasets

Filter by project ID. For a real project, orphan (no-project) datasets are included only when include_orphans=True. Omitted or the 'default' sentinel always includes orphans.

When true, include the caller's own failed datasets (status='failed'). Defaults to false so the UI never renders broken-but-clickable records that 404 on preview/analyze. The result set is always scoped to the authenticated user — this flag does NOT grant cross-user visibility, so passing true is safe for any authenticated caller. Failed rows remain in the database for support / debugging; their original upload is preserved at raw_s3_key.

When true and project_id names a real project, also include the caller's own orphan (no-project) datasets alongside the project's own rows. Ignored when project_id is omitted or is the 'default' sentinel, where orphans are always included regardless. Defaults to false so an ordinary project-scoped list stays scoped to the project instead of being swamped by every one of the caller's orphans; set this explicitly from surfaces that deliberately want orphans folded in, e.g. the MLE agent's dataset picker (ENG-6077).

Maximum datasets to return, newest first, after collapsing to the latest version per name. Defaults to and is bounded at 200 (oversize requests are rejected, matching /felix/training-jobs and /projects/{id}/evaluation-runs). Callers that need the full set page with limit/offset (ENG-7085): the frontend datasets tab and the sandbox dataset picker both loop until a short page rather than issuing one unbounded read.

Number of leading datasets to skip.