Page through a job's finished items (JSONL)

Streams one page of FINISHED items (status succeeded, failed, cancelled or skipped) in seq order as JSON Lines: one BatchResultLine object per line, each terminated by \n, no array, no trailing pagination object. Works while the job is still running (returns what has finished so far). Unfinished items are never listed.

Pagination is in headers: if X-Next-Cursor is present, request again with cursor=<that value> (the Link header already contains that URL with your limit and status preserved). No X-Next-Cursor means this page is the last one right now; on a running job, later calls may return more items after the last seq you saw. A stream that ends early without the header you expected on a full page indicates a dropped connection — retry the same cursor.

Bodies: result.content is inlined from the job's stitched result file, which exists only after the job is terminal, so result is absent on a running job. Caps: 524,288 bytes (512 KiB) per item and 2,097,152 bytes (2 MiB) per page; see BatchResultBody for truncated / omitted / unavailable.

Retention: results are readable until results_expire_at (72 h after submission by default). After that this endpoint answers 410 ERR::REQUEST::BEYOND_RETENTION even if rows still exist; the job object itself stays readable via GET /v1/batch/{batchID}.