They did not put that in the readme. The README contains only references to test videos, videos that don't actually exist, and one small-time video (in spanish?) that seems to be an old test video, but I have a hard time seeing for sure.

The source code does contain references to copyrighted videos in the tests, tests intended to make sure that youtube-dl can download the data from videos using the extremely "token" signature scrambling that youtube employs for certain videos. You can see the test cases here:

https://gitlab.com/HacktorIT/youtube-dl/-/blob/master/youtub...

There's stuff like Justin Timberlake in there and Taylor Swift. This was immensely sloppy considering how the RIAA has won cases before.

But from what I can tell, these unit tests don't even download the video - just pull metadata.

I don’t see how Napster is more applicable than vcrs. YouTube in this case is the broadcaster, putting content out to a general viewing public, ytdl is a recording device. For Napster to work, YouTube has to be imputes with illegally providing the copy and ytdl has to be the means to facilitate the copies transport.

Napster and VCRs were both accused of being tools to infringe copyright, VCRs won the lawsuit, napster lost. A big part of that related to how napster positioned itself in terms of marketing compared to vcrs.