And there's the nub of the whole problem. You all are so slavishly wedded to the concept of an ultra tight playlist that you'll start with a meager sampling of only 2,000 songs. And regardless of what various permutations of mathematical chicanery you go through to extract the "best" 400, you're stuck with inflicting a tight list of only 400 songs on us. So you test 2,000 and only play the top 20% of them. I submit that if you were to test 10,000, you could limit your playlist to only the top 10% (which means only the very highest testing songs) but we'd get to hear 1,000 different songs!
It does not work that way. Period.
Stations don't pick "400 songs" or "the top 10%" or have any other quota.
Stations test as many songs as they can afford... testing is expensive, and a large test can typically run around $50,000. From those songs, no matter how many, they play the ones that are broadly positive with minimal negatives against the target audience and all of its subsets.
So if a station comes up with 681 playable songs, that is the size of the library. It does not matter if you test 1000, 2000, or 10,000 songs because after a few years in a format, all the potential songs have been tested and the ones that always score negative are no longer tried again.
And if in a test we find that, due to changes in taste of the target demographic, there are only 542 viable songs, the playlist gets reduced. Or if we find that the same demographic likes more 80's songs, wet might find, over a several year period, 211 songs to play we were not playing before... but, of course, some of the older stuff will age off the list.
No quotas, no percentages, no desire to find just the top xx number of songs. Whatever number of tunes are positive testing get played.