I have a hypothesis. I think that Tesla is choosing exactly 20% of the beta community to be the first public wave of each release.
TeslaFi has a little less than 4% of Tesla's FSD beta community. If I randomly select 4% of group X and call it group A, and then also randomly select 20% of group X and call it group B, the percentage of group A that are also in group B will not be exactly 20%. It will be some number near 20%, and the variance will be randomly distributed.
That's exactly what we are seeing. Each new release shows up on Teslafi as a number somewhere around 20%. Last time it was about 18%. This time it is a little under 22%. (512 out of 2366). I haven't kept careful track of the earlier releases, but it seems that it has always been "around" 20%.
Teslafi's 4% is not randomly selected, it is a self-selected group of enthusiasts. And we don't know that Tesla's 20% is randomly selected either - they may have specific criteria that they use to select the first wave.
But as long as the criteria used to select each group are stochastically independent (no correlation between them), the variance from 20% for the intersecting groups will still be randomly distributed.
So if Tesla is selecting exactly 1 out of every 5 beta participants for the first wave, it will never show up as exactly 20% on TeslaFi or TeslaScope. It will always be some percentage near 20%, just as we have been seeing.