Grep the networks · one capture window · as of 2026-07-21
Read this first — what this is, and what it is not
This is the transcript of a single capture window across three cable channels (CNN 45,936 · MS NOW 46,526 · Fox News 48,057). It is not comprehensive — not every network, not the whole day, just the stretch of tape the desk logged.
Every line is machine-transcribed from what was broadcast, and timestamped in UTC. Transcription is automated, so a line is a close record of the audio, not a certified quotation.
Each line is tagged program, ad, or promo (79,069 program · 33,378 ad · 2,500 promo) so paid time is distinguishable from program time; a line outside any classified block is marked untagged.
It is a machine-readable record of what was broadcast in the window, offered for search. A record to search, not a verdict.
Capture window: 2026-07-18 15:42 UTC – 2026-07-21 15:42 UTC. Full method: the Method.
154,621spoken lines captured across three channels in one window, each tagged program, ad, or promo. Type a word to search the tape; the count is a count of lines, never a claim about a network.
Most of the desk’s beat reads coverage against coverage. This rail is simpler and more mechanical: it is the tape itself, made searchable. Everything the three channels said in one window — program, ad, and promo alike — is here as spoken lines you can grep. The desk holds the record; what a reader looks for in it is their own.
[OUTPUT] Transcript Corpus: 154,621 spoken lines logged across 9 channels in a single capture window — each with a channel, a UTC timestamp, and a program / ad / promo tag, searchable client-side and published verbatim at /transcripts.json. One window, not a network’s full record; a queryable record, never a verdict. probability mass ≠ 1.0.
