Please note that that is not at all what the paper shows.
First, as you can see on their figures 2 and 3, they only tested 10 deletions of amino acid regions on this one protein and the smallest version they got was still functional. As such, there is no proof that you can’t delete even more amino acids or make non-linear deletions (i.e., remove smaller stretches that don’t have to be adjacent/contiguous).
Second, they did not test any amino acid substitutions at all. Even if the smallest functional version of the protein was 301 amino acids long, they would have to try 301^19 mutated versions of the protein to completely determine which versions still function and which ones don’t. Of course, there are abbreviated strategies to explore this search space but they didn’t even try to go there.
Third, even if it was shown that the protein had to have an exact sequence and an exact length for a given function A, this still does not prove ID’s irreducible complexity because a slightly mutated version of the protein might have been perfectly fine to perform a different function B. As such, the protein could have existed for a long time to perform function B and then one day it would have been duplicated, mutated just right and thus become available to perform function A.
It’s important to clarify what the ID definition of irreducible complexity actually means - if you see it for what it is then you’ll also see that this (or any other) paper doesn’t actually provide any adequate proofs.