I can see why you’d say that, and I’ve heard this argument before, but I’ve never thought that was a good method. The reason is that there is a 0% chance that Blizz has the ability to look at a true average of all clears with a given build and then essentially make corrections both for player enthusiasm and for player intent.
Player enthusiasm will hugely deform the average, because as we know, players tend to sink a lot more of their time into builds that are already powerful. And that means that those builds end up getting explored pretty thoroughly. On the other side, when looking at an average in this way, one might conclude, “oh, build X can only do GR 120, on average”, when in reality that build is kind of being sandbagged by the fact that only a small number of lower quality players are dedicating much time to it. If you then proceed to buff it by the amount you think is necessary, based on that picture, good players flock to it and it ends up becoming overpowered.
Player intent is also a problem. After all, what do you make, say, of a 7k or 8k paragon player who shows up on the leaderboard around rank 800, having cleared a GR 123 or so? Was this person just inept? Or are they maining another class, and just decided to jump in and mess around? Also: how many keys did it take them to get that clear? One? Fifty?
When you compound these two issues together, you end up with an impossibly tangled picture of what the “average” clear is. Maybe if Blizz hired a team consisting of an economist, a mathematician, and a psychologist, then they could unravel the data and form an accurate picture. Hopefully I don’t need to tell you that isn’t going to happen!
This is why I’ve always considered a clear-headed look at the uppermost outliers as the only coherent way of figuring this stuff out. If you then want to apply “discounts” for various characteristics of the build, that isn’t all that hard to do. If, say, a build is hard to play or very fishy or very single-target oriented, all of these will have fairly easy-to-account-for effects that you can factor in. And accounting for lower level gems and augments, or less paragon, is just very simple math.
A build that is hard to play (which usually means high on cooldown management and timing elements, squishy/fragile, or both) will lower the “average clear” by 2-3 GRs. This number is pretty solid simply because in many cases, if the complexity or fragility is costing you more than this, you can simply change your gearing to simplify the build or add toughness (i.e. you drop CoE in favor of something else). You can also apply this number to a player whose skills are truly poor, even if the build is not actually difficult, and compound it on top of itself. In other words, a player who is not merely average but truly bad will be 4-6 GRs behind a good player.
A build that is very fishy (i.e. derives a ton of benefit from mob density, pylons, and specific RGs) will lower the “average clear” by 5-6 GRs. I’m a little less solid on this number, but it’s based on both my own long-term observations and Wudijo’s analysis of RNG in GRs. Basically, it’s a somewhat subjective assessment of how much trouble the “average” player will go to to get a better clear. Also, note that much of this disparity goes away if you are even moderately determined.
A build that is very single target oriented will, in comparison, have an average clear only 2-3 GRs lower. Because most of the RNG factors Wudijo lists will have a smaller positive or negative effect, depending on which way you’re looking at it.
So: imagine you have a very skilled player with 10k paragon and 60k mainstat playing a not-very-difficult, high-single-target damage build, and with this they cleared 148, and that’s the best anybody has done. If you’re comparing them to a skilled player with 2k paragon and 20k mainstat, this second player will have -7 GRs from the paragon difference and 2-3 GRs from RNG factors. So, - 9-10 total. If this player is very skilled and very determined, then they’ll only have the -7 from the paragon difference. This is what you would expect to see from a buffed H90 Frenzy, for instance. If the build in question was a buffed Vile Charge, you can expect to see a larger gap.
So, I didn’t pull those numbers completely out of thin air. Here’s where they come from:
Fire Leap: Based on my own attempts at clearing 126 (not done yet, but have come close several times… 90+% sure I can do it), which with “upgrade” to 10k paragon and rank 150 gems would clock in at about 133. And before Ulmaguest left the game he said he thought the build could do 133-134. I’m pretty certain about this one.
Phys Leap: These two builds are very close in power, so same goes here.
Pro-Slam: Ulmaguest did 138, in a not-super-long push. If you get numerous high-paragon, skilled players running the build, that number will go up.
IK HOTA: Titannova, who has always been one of the best IK HOTA players out there, did 133 during the PTR when Remorseless was upgraded. At the time he had about 6k paragon. Upgrade to 10k, and 136 should be pretty doable.
Vile Charge: DP already did 135, and the rift he did it in was certainly not perfect.
LON/LOD HOTA: Based on the observed performance comparison between this build and IK HOTA.
R6 HOTA: Ditto.
H90 Frenzy: Same as Pro-Slam. Only one high-paragon, skilled player that I know of has pushed with this, and 139 is done. The number will go up if more people play it.
So, to finally get around to this part…
Yes, balancing by average, rather than by peak, would be a fine thing to do. All that really needs to be done to do that would be to reduce the buff I recommended to H90 Frenzy by about 3-4 GRs, and the buff to Pro-Slam and the HOTA builds by about 2 GRs. Leapquake and Vile Charge could stay as suggested.