Run one expert skill on up to ten windows of recordings the user may read, at once.

Takes a user's delegated access token with the insights.write scope, like a single call (readable recordings and billing as there); an API key or a browser session is refused with 403.

Each item takes the fields of a single call, including its own media_binding, except product, language and condition, which the batch shares (recordings in different languages belong in separate batches). A batch reads at most four distinct recordings (422 invalid_request on items beyond that). Every recording must be readable by the token's user before any is evaluated: one that is missing or not readable by them is a 404 binding_not_found for the whole request, and a token without media.read is a 403 naming the scope; nothing is evaluated or billed then.

The items run concurrently; the entries come back in request order, each naming its media_binding and carrying its own result or error, charge, replayed flag and model_ids, and credits_charged is their sum. A problem with the request as a whole (validation, authentication, a recording that is not readable) is the usual error response; a failure of one item is that entry's error, e.g. evaluate_timeout, expert_in_progress, media_not_ready or moments_not_ready for an item whose recording is still being processed (its field names items.{i}.media_binding, and the items on other recordings still run), or a subject, person or moment of that item the recording does not have (subject_required, moment_not_found, invalid_request naming the item's field), with the same fields a single call's error carries. A failed item is not charged and its siblings still run. If the balance cannot cover the items that need a new evaluation, those items answer insufficient_credits together; replays and requests outside the skills' scope are still answered. Identical items are evaluated once and charged once: the first carries the charge, the others are replayed.