I got a little nerd sniped by this. I wrote a program that takes an input (/usr/share/dict/words on Debian 12, specifically), and compresses it with gzip and Zstandard, and also expands its binary representation (by replacing each 1 bit with A and each 0 bit with B) and compresses that. The result is:

    gzip    expgz   zstd    expzstd original
    263120  421278  252449  389004  985084

So unlike in the article, expansion + compression is at least better than the original. The ratios are (smaller # better; the opposite of how most compression algorithms advertise their performance, but what the original article used): 0.27, 0.42, 0.25, 0.39. gzip and Zstandard aren't a lot different in either case. Whatever patterns that the Javascript weirdifier uses is less obvious to compression algorithms in general than just bitwise substitution.

Here's the program if you want to look for bugs: https://go.dev/play/p/wwNXVzO2TO-