Jamberry Proposal
Basis of this proposal is that Jas' idea has some value, but is a bit complicated for new TCers and also has far too much spread from more rare to more common artifacts.
Values for my proposal were chosen from general observation in qph and other servers' help channels that generally, at late game people's common mainstat artis will be better than their rare ones, but not dramatically. In general, though some artis may be more common, there is only so good an arti can get, and % gain in quality per resin (and therefore, per # of that artifact farmed and rolled) decreases the better players' artis get.
For example, ATK goblets are a bit over 4x (simplified to exactly 4x bc I'm lazy) as common as Anemo goblets.
At AR50, you might have gotten 5 Anemo goblets and 20 ATK goblets ever. If you need to gear for example both your Jean and your (4verm for argument's sake, since mixed sets Xiao could prefer ATK) Xiao, and choose Anemo goblets, your Jean will be using a top 40% (2nd best out of 5) Anemo goblet. If you were to choose ATK goblets, your Jean could be using a top 10% ATK goblet.
At AR55, that could be 20 Anemo goblets and 80 ATK, for a top 10% Anemo goblet and a top 3% ATK goblet.
At AR59, staring into the void ahead and questioning the choices that brought you here, that could be 100 Anemo goblets and 400 ATK, for a top 2% Anemo goblet and a top .5% ATK goblet.
Realistically speaking, a top 2% Anemo goblet is much much closer to a top .5% ATK goblet than a top 40% to a top 10%. Thus, direct correlation between rarity of mainstat and useful sub count does not make sense as a means of modeling artifact quality. The best solution should come from some compromise between practical data and the need for standards to be easily implementable.