I've now answered this in the writeup (point 11).

This is a really funny comment. In setting the world record for Li's 39-round collision[1] (still unbroken, and one of our favorite papers), he also set some records in sha-224, reaching 40 rounds in that one. Of course, saying sha-224 is "87% of the way" to sha-256 is correct in a sense, and that's why his record is slightly larger in reduced-round full-schedule collisions on that metric, 40 rounds for sha-224 and only 39 in sha-256. At the same time, the fact that he reached only 39/40 rounds on those shows the difficulty of getting through the full 64 rounds, which is what our paper does with a slightly relaxed schedule adherence.

[1] https://eprint.iacr.org/2024/349.pdf