Hume Webhooks Batch Jobs

Operator submits a batch Expression Measurement job over 500 recordings and needs to map each result back to its source recording.

Submit via POST /v0/batch/jobs and persist the returned job id; map each input (file/url) to its own stable application-side identifier so predictions can be reattached to source rows. Do not rely on input ordering alone for identity.

A single batch job is submitted with a very large media file (or hundreds of long files) and the operator assumes it will always be accepted.

Pre-check inputs against documented per-file size/duration and per-job limits and split oversized work into multiple jobs; on a limit rejection, do not retry the identical payload. [REQUIRES-VERIFICATION] for the exact numeric caps.

After submit, the operator polls GET /v0/batch/jobs/{id} every 500ms until COMPLETED.

Poll job status with exponential backoff and rely on the terminal states (COMPLETED / FAILED). Do not hammer the status endpoint sub-second, and do not assume IN_PROGRESS means failure.