State of Set Balance Report

General Discussion
21

Hi Argos, first I want to make clear – there is no agenda other than pursuit of the truth. Rage and I only wish to put helpful information in front of the community.

The 10 richest representing the wealth of 500 random people example is too loose of an example to be a fair one due to population size being in the billions.

I recommend you go through the same journey and see what happens when you start averaging the Top 1000, with and without the top+bottom 10%. Do this for a popular and unpopular set. The results will contradict the top clears to the point where the report is deemed unhelpful, then not credible. Top clears are highly influential, they are the proof that a build is performing within acceptable limits, they cause nerfs, they heavily impact the decisions of players.

Rage and I recognized that the average metric deteriorated very quickly even before the Top 100, so a stop was put there, though we draw and sort our data from the Top 200 in various eras.

What is the point of all of this anyways? To keep entertained while helping others. Take this example, which thousands of players rely on to learn about the best builds:

The Maxroll people organize their own tier list season-after-season, it impacts players decisions – especially those who cannot afford the time to setup all builds. And while Maxroll doesn’t claim their tier list as an absolute truth, people trust and use it regularly. But what is it based on? Player opinions, collaborative spreadsheets like ours, a proprietary Maxroll algorithm, secret sauce? Who knows right?

Here we have created a parallel list. We’ve exposed all of the data behind it, the calculations, the methods with explanations. Our results are similar to Maxroll’s, but not exactly the same. Our list was derived AFTER the patch, theirs coincided with its release.

I’d say our work product was inspired by this Maxroll tier list. Without structuring the averages as we’ve done it — the tier list would look very different. Maxroll’s Tier list is helpful for competitive players, as is ours. Form your opinions using the Average Top 100 since the Top 10 is bothering you – that’s why the column is there.

I hope you’re still interested in telling us the specific barb set results you disagree with and why?

4 Likes
22

If more people played you could do a top 100. Now, in NA, maybe 3 or 4 builds could hold up to top 100 scrutiny. Most of the others are smaller niche communities. Im pretty consistent with the stats.

I have 3k paragon on the nose. I havent pushed a lot yet as I was trying to gain paragons, but now I’m at 135/135/131/131 with my top 4 builds. Keeping in mind I haven’t played reakor in 800+ paragons that’s not terrible. btw - you really find out how squishy it is after playing other top builds. YOUCH.

2 Likes
23

Yeah, I still find it insane that some of the high paragon guys play with no BoM and no Esoteric.

24

It’s almost unplayable to me now lol. Gonna do up a leapquake build whilest I re-adjust my playstyles.

25

Right! That’s basically an old issue which a lot of people just didn’t notice.

A better solution is to check the top clear distributions. Not sure exactly how this works. As long as we acquire enough data we will get a curve which almost independent of the number of data. I remember someone did that before? Pro khan? Right?

1 Like
26

Pro did this analysis a while back, which was interesting, but which also kind of fell into the “averages trap”…

1 Like
27

Thanks for the link. looks like “#players” distribution data is not used.

When # of players increases, the potential limit also increases. If the top clears satisfies Gauss Distribution (adjusted tier cleared, “#players” ), we may be able to get a model and see the difference.

That’s just pretty intuitive ideas…

28

i dont understand the adjusted clears. Do you also subtract paragon in cases that the clear was a much higher paragon?

29

Unfortunately this method does not work when setting up a tier list. I used to be pretty active in the competitive fighting game community, and anytime tier list discussion came up, there used to be an argument over how to determine tier placement.

Over time it became apparent that only top play could reliably determine tier placement. It makes no sense to do anything less because then standards are all over the place. Some would complain that a really hard to use character was top tier when most couldn’t achieve that kind of result. But the truth was that in good hands they were very strong, and once that was known most players attempted to get to that point.

You could say the same of Zunimassa. It’s theoretically a strong build, but almost no one is able to reach its full potential. Does this mean it shouldn’t be as high as it is? No, because that is simply obscuring the truth.

You cannot use regular old averages when in a competitive environment. Anything less than full potential will mean you lose or push aside real data. You can make a different section with different criteria if you want, but to get true and accurate measures you need to use top play. That’s just the way it is.

3 Likes
30

well yah, seems like common sense to me.

though we likely dont have enough parses to even know if half the sets are being played to their full potential.

1 Like
31

Just thanks for your work (You and Rage).

Your main thread is the most interesting thread since a long time.

2 Likes
32

Hi Bravata, straight from the appendix, those ‘Marks’ examples are super helpful for quickly understanding how it works. There are two types of Adjustments that we process: Paragon and Time.

For the paragon adjustments, we assume that a player at 5000 paragon will have 35,000 mainstat. This is just a ballpark figure, and the actual amount will of course vary from case to case. But this is a good average, that is quite close to the mark in the majority of cases.

Next, we calculate how much less (or more) mainstat the player of the clear in question would have, based on their paragon. For instance, we presume that a 4000 paragon player has 1000 less paragon, and thus 5000 less mainstat, than a 5000 paragon player. So that player would have 30,000 mainstat.

Then, we calculate the mainstat difference between the player and the 5000 paragon benchmark. So in this example, that would be 30000 / 35000 = 0.8571. Then we take the base 1.17 log of this amount (since mobs grow 17% tougher per GR tier). So, base 1.17 log of 0.8571 = -0.98, or in other words, a player with 4000 paragon is getting roughly a -1 tier penalty (we round to the nearest tenth) compared to a player with 5000 paragon. So we add +1 tier to this player to even them out.

The general paragon marks for bonus/penalty tiers are:

  • 0: +8
  • 325: +7
  • 725: +6
  • 1200: +5
  • 1750: +4
  • 2400: +3
  • 3100: +2
  • 4000: +1
  • 5000: 0
  • 6200: -1
  • 7600: -2
  • 9200: -3
  • 11100: -4
  • 13250: -5
  • 16000: -6
  • 19000: -7
  • 20000: -7.3 (and that’s the end of the scaling, since paragon stops at 20k).

There is also a Time Adjustment bonus:

So: we compare the time taken vs the max time of 15 minutes. For instance, a 12:00 clear, or 720 seconds, vs 15:00, or 900 seconds. 720 / 900 = 0.8.

And again, we take the base 1.17 log, which in this case would be -1.42. So, we correct for this by giving a +1.4 tier bonus for this 12 minute clear. There is obviously no negative version of this correction, since if you exceed 15 minutes you fail the rift and aren’t on the leaderboard.

The general time marks for bonus tiers are:

  • 15:00: 0
  • 12:50: +1
  • 11:00: +2
  • 9:20: +3
  • 8:00: +4
  • 6:50: +5
  • 5:50: +6
  • 5:00: +7
  • 4:15: +8
1 Like
33

Yeah. Here’s a concrete example from D3 that makes it clear what would actually happen if you used that “averages” method:

Let’s compare two builds.

“Build #1” has peak adjusted clear 151.1, avg of top ten 150.5, and 100th place 147.7.

“Build #2” has peak adjusted clear 145.0, avg of top ten 140.6, 100th place 125.7.

The actual full data pool for each build is 400-600 clears. So, let’s do something like what’s been suggested, and drop the top 10% and bottom 10% off both pools, then average the rest. Doing that, we get 146.3 for Build #1, and 120.8 for Build #2.

So, Build #2 should get a buff of 146.3 - 120.8 = 25.5 tiers, in order to achieve parity with Build #1, right?

Wrong. If we did that, Build #2 would be taking down GR150 at 0 paragon, probably in less than 10 minutes, and quite possibly in as little as 5 minutes.

How do I know that? Well, because Build #1 and Build #2 are the same build, Masquerade Bone Spear Necro, before and after it got nerfed. And, we know exactly how much that nerf was worth, because it’s purely mathematical, taking the build down from a 3111x multiplier to a 671x multiplier, which equates to -9.8 tiers.

So, the first 10 tiers or so of our 25.5 tier buff take the build exactly back to where it was before it got nerfed, which was one of the most powerful builds ever seen in the game, taking down 150 around 4500 paragon. And THEN we’re stacking another 15.5 tier buff on top of that!

And that is enough of a buff to take both the paragon requirement for 150 down from 4500 to 0, and the time requirement down to about 5:00.

So, that’s a case where we KNOW, basically with complete certainty, what would happen if we were to adopt this balancing method (an epic disaster). And the exact same thing would happen if we tried to use that method to balance LoD DH, Trag’oul Necro, Raiment Monk, etc.

4 Likes
34

Design has always go by bigger risk, inclined backdraws or narrowed reaction frames bringing more rewards. So it’s not surprising that Raekor is one of the top builds by fitting at least one. Not for the feint of heart for sure.

35

I loved it when reakor was the only build i was playing. Then I got to the point where I needed to get some paragon so I made a monk. Super tanky, most utility, highest quality of life, tons of damage. (Whereas reakor just has tons of damage) After playing that for a while I lost the feel of reakor. I know I can get it back if I play a few days, but at this point do I really want to?

36

The one thing that really bothers me about Raekor is the jailer bug. Without that, I don’t think I’d actually die very often.

37

Similar bug applies to Monks’ Dashing Strike and Demon Hunters’ Vault. You can dash around with Jailer imprisonment still on you. At random occurances DS sometimes miscast twice in a row for no reason or simply refuse to be activated in the heat of battle when you try to diaengage. I think it’s due a collision issue on the engine or it stems from action priority of attacking a nearby target in a crowd taking over to ignore it somehow. Compared to DS and Charge, Vault is much more precise.

On a similar case if you happen to hover over a corpse chest or a weapon rack as you finish casting DS, Vault or Charge, you will be halted in your tracks too. Action priority of clicking the corpse expects you to press m1 but you want to cast another spell by holding that another button down. Best option is having Force Move (mwheel flicks) to shake it off so you can keep casting it.

38

Yeah, that doesn’t work with this bug. Nothing does. Even becoming CC immune (i.e. casting WOTB) won’t free you from the Jail.

39

I find Vault to be a poor movement skill due to inaccuracies caused by map collisions and it being influenced by attack speed. I’ll take Dash double spending charges over Vault’s clumsiness any day.

1 Like
40

You know the sad part about this? The devs will just ignore it.

Edit post 2.7.4 patch notes: I was right, sadly.