startMs and text segments and link to the relevant video time. Automatic captions can misspell technical words; paraphrase with that uncertainty. The study verified transcript text, not the audio or video bytes.
Comments form another path. Read a video’s comments; then use the selected comment’s repliesContinuationToken as the initial identity for the reply operation:
REPLY_TOKEN is a placeholder that must come from the returned comment. A later replies page uses the continuation token returned by that replies response; do not substitute the video URL or a search cursor. The transcript route was exercised in the September study, while the reply-token route here is based on the current operation contract and was not live-tested by that task. Channel videos, shorts, playlists and community posts have separate operations; use their returned native IDs and current web describe contract when entering those collections.