The bit that gets me is the finite set of admissible coefficients in n=4,6,8, once you know that, any search for new Gold-cube switchings in even dimension can stop immediately instead of wandering through bigger n. That seems useful for people doing computer searches around APN families, because it turns a maybe-open parameter hunt into a proof that the same trace-product ansatz just wont produce anything past 8. The normalized rank-two extension classification in dimension eight feels like it should also make the equivalence testing cheaper, since you only have to sort the eight marked switchings into two CCZ buckets instead of treating them as separate curiosities.