Nice post. Relatedly, see also malus.sh and this talk by the people that made it (both of which I posted in this lemmy community here).

A couple of minor corrections to your text:

Blanchard's account is that he never looked at the existing source code directly.

Blanchard doesn't say that he never looked at the existing code; on the contrary, he has been the maintainer (and primary contributor) to it for over a decade so he is probably the person who is most familiar with the pre-Claude version's implementation details. Rather, he says that he didn't prompt Claude with the source code while reimplementing it. iirc he does not acknowledge that it is extremely likely that multiple prior versions of it were included in Claude's training corpus (which is non-public, so this can only be conclusively verified easily by Anthropic).

The GPL's conditions are triggered only by distribution. If you distribute modified code, or offer it as a networked service, you must make the source available under the same terms.

The GPL does not require you to offer GPL-licensed source code when using the program to provide a network service; because it is solely a copyright license, the GPL's obligations are only triggered by distribution. (It's the AGPL which goes beyond copyright and imposes these obligations on people running a program as a network service...)