Filtering doesn’t actually reduce the work in this context because you still have to find the correct value before you can filter it. It’s like claiming you can pick winning lottery tickets faster by throwing out the losing ones, but you still have to check each ticket to know if it’s a winner. With private keys, unless you already have extra information (like a public key prefix that actually leaks some useful bits), every candidate has to be generated and checked in full. Filtering after the fact doesn’t magically cut down the number of computations, it just adds an unnecessary layer. You can only filter something once you’ve already done the work to find out if it matches or not so there’s no real gain.