The run that was never resumed

Researching how a genre of build in public content talks about making and monetising software means watching hours of video, which is a bad way to spend a week. Transcripts turn that into something readable and searchable, so a script was written to list every video on a channel and pull the captions for each one into a local file.

It found one hundred and seventy-five videos. It saved thirty-nine transcripts. Rate limiting blocked it three times, the backoff got longer each time, and the run stopped partway through the fourth attempt and was never restarted, which means most of the catalogue was never collected at all.

There is no clever recovery in this one. The blocking was ordinary, the fix would have been ordinary too, and the honest reason it stopped is that the thirty-nine transcripts answered the question that prompted the work, so the remaining hundred and thirty-six stopped being worth the argument with the rate limiter.

It stays in the archive because the number is the interesting part. A tool that collected everything would say nothing except that collecting works. A tool that stopped at thirty-nine, and says so on its own page, is a more accurate picture of how this kind of work actually goes.

Most portfolios quietly drop the projects that stalled, which leaves a body of work where everything shipped and everything succeeded, and that is not a record of how anything is built. The abandoned ones are evidence too, as long as the count is printed next to them.