List the sentence-sized chunks a long take was generated from.

VoiceLabs (long-form phase 1). This is what the studio renders the "re-roll this sentence" picker from: each entry carries the text that was spoken and where it sits in the audio, so the user can hear a chunk before replacing it.

409, not 404, for a take that predates chunk tracking: the generation exists, it simply has no per-chunk record — the text was chunked, mixed and the boundaries discarded. Same degradation shape as a pre-captions capture.