How it's worked out

Each stream is transcribed automatically, then an AI model reads the whole transcript and splits the stream into parts, putting each under the category above that fits it best. Anything that fits none of them is Other.

Percentages are shares of each stream's full length. Totals over several streams are weighted by length, so an eight-hour stream counts for more than a one-hour one.

The labels come from an AI reading an automatic transcript, so expect mistakes: misheard words, and judgement calls where one part turns into the next. Every part is listed with its timestamp so you can check it against the stream.