nx_yt_captions_lib.nx

nx_yt_captions_lib.nx -- the YouTube TIMED-TEXT (caption/transcript) path, composed on the estate's proven InnerTube + sovereign-TLS stack. It does NOT duplicate nx_yt_innertube: it COMPOSES yt_innertube (the ANDROID_VR /youtubei/v1/player POST that self-heals against key rotation) to obtain the player response, reads the captionTracks[] baseUrl out of THAT SAME response, and fetches the timedtext body with nx_https_fetch_follow (the shared gzip-decoding, redirect-following GET). Parsers turn json3 and srv1/XML timedtext into rows `startMs \t durMs \t text`, entities decoded through nx_html_entities. WHY A SIBLING LIB, NOT AN EDIT TO nx_yt_innertube: that lib's header declares its subject as the stream path (streamingData.formats[].url); captions are a distinct output read from the same response, so this stays a NEW file that imports it (blast radius: zero -- nothing that imports innertube is touched). WHY THE WEB baseUrl DOES NOT WORK (measured 2026-09-18 from the NAS): the WEB captionTracks baseUrl scraped from a watch page answers status=200 with an EMPTY body -- it is IP-locked (ip=0.0.0.0) and requires a proof-of-origin (pot) token the WEB client mints via BotGuard. The ANDROID_VR client's baseUrl carries no such requirement, which is the whole reason this composes yt_innertube. license_tier: ORIGINAL No hw writes (Rule 26). LIB (no main).