You appear to not understand what you’re reading.
He is extrapolating the results from other work to a protein with a size similar to the one Axe worked on. Specifically (10-24/2093)x(20150) = ~10-59.
Those values are between 14 and 53 orders of magnitude higher than Axe’s number(10-63 to 10-24). So Axe’s number is not within that range(which it appears you are reading Art to say), one of the reasons being that his experiment was fundamentally flawed by Axe working on an enzyme intentionally manipulated to be extra sensitive to mutation. So even the most similar number is still off by a factor of one hundred trillion.
Since the method Axe is using fundamentally is based on extrapolation from how many mutations “pass” (the rate of passing mutants feed directly into his calculation), because the enzyme he used is intentionally created to have fewer passes, this has the effect of reducing the number of sequences he estimates.
Now of course, this is not even an estimation of the amount of sequences that have that enzymatic function in sequence space(it cannot possibly be, since we know of entirely dissimilar protein enzymes with the same activity).
It is an estimation of the number of sequences that simultaneously adopt that particular protein fold and catalyzes the hydrolysis of beta lactam antibodies(as Ann Gauer also agreed here*) well enough for colonies to appear on an agar plate containing the mininum inhibitory concentration(MIC) of antibiotic. This is NOT the limit of adaptive function for an enzyme in a microorganism in the real world.
Italics are her own emphases. That structure and that enzymatic activity. Not an estimation of the true density of the function in question, much much less an estimation of any fitness-enhancing function.
And the deck was deliberately stacked by using a temperature sensitive enzyme and testing it against the MIC.
It’s trash work, all of it, and you don’t even understand it.