Proc trinkets changes

[mmolcq]

http://www.battlenet.com.cn/wow/zh/f...pic/7524042708

Summary
We have added some protections for unlucky streaks for trinkets that use the RPPM (real procs per minute) system. This change is now live.

Longer Explanation
In general, trinkets and other proc items in games have a relative high chance to proc, and they have ICDs. Theoretically, the proc change is random, but predictable. You may know that one trinket will proc randomly at first, and then proc every 45 second(or proc at other ICDs).

In 5.2, we are using RPPM system to make the proc mechanic more unpredictable and not that boring. When you are lucky enough, your trinket may proc in a row, which makes you feel great. But on the contrary, sometimes it doesnt proc at any rate.

We have made a hotfix to this RPPM system yesterday evening. Every time the trinket fails to activate, there's an increasing chance that it will activate. The time you are unlucky will be less, the time you are super unlucky will not happen. Furthermore, we have increased proc chance for these trinkets. In general, proc chance for agility and strength trinkets from Throne of Thunder has been increased by 10% and proc chance for intellect and spirit trinkets from Throne of Thunder has been increased by 5%.

Painful Theory Explanation:
As usual, if you want to use these trinkets, or you want to know how to use them better, you don't need to read the explanation below. The explanation below are for theorycrafts and geeks to understand the details of this system to design different kinds of character models and simulators.

The calculation of proc chance remains unchanged, and you can figure out the average proc interval. We are now tracking "the time since the last successful proc". This "time" is different from "the time since the last chance to proc". The maximum value of it is 1000 seconds.
Proc Chance = Normal Proc Chance * MAX(1, 1+((the time since the last successful proc/average proc interval) - 1.5) * 3)
For example, if the average proc interval of a trinket is 45 seconds, and the time since the last successful proc is 72 seconds, then you will get 1.5 times as much as the normal proc chance. If you are out of combat for a period of time, and the time since the last successful proc is 5 minutes, then you will get 16.5 times as much as the normal proc chance.

The proc chance of the trinkets below has been increased by 10%:

• Renataki's Soul Charm
• Talisman of Bloodlust
• Bad Juju
• Rune of Re-Origination
• Fabled Feather of Ji-Kun
• Primordius' Talisman of Rage
• Spark of Zandalar
• Gaze of the Twins

The proc chance of the trinkets below has been increased by 5%:

• Wushoolay's Final Choice
• Breath of the Hydra
• Unerring Vision of Lei-Shen
• Cha-Ye's Essence of Brilliance
• Horridon's Last Gasp
• Inscribed Bag of Hydra-Spawn
• Stolen Relic of Zuldazar
• Lightning-Imbued Chalice

The proc chance of Unerring Vision of Lei-Shen used by Balance Druids has been changed to 65%(was 50%).

[Xanthan]

Hopefully the US post clarifies a few points:

1) The math in that example doesn't check out. 1+((72/45)-1.5)*3 = 1.3, not 1.5.
2) What determines a trinket's average proc interval? Is it based off a trinket's RPPM rate? Is it affected by haste? Is it just a magic number assigned on a per-trinket basis like an ICD?
3) Is the 10%/5% increase in proc rates a change to the normal RPPM rates, or is it an estimate of how much more often these trinkets should proc as a result of this change?
4) Does this apply to every RPPM proc, or just trinkets? E.g. warrior 2 pc bonus, legendary meta gems, and weapon enchants are also all RPPM effects.

Lastly, as just a commentary, I don't like this change at all. This isn't just adjusting RPPM's pseudorandomness to make them more reliable. It's just shaving off the bottom of the bell curve on bad RNG without touching the top of it, which means the RPPM number is no longer an accurate measure of how many procs you should expect per minute(after adjusting for haste). I was imagining they'd find a more clever approach that would maintain the exact same number of average procs per minute and just steepen the bell curve's hump, meaning you'd see fewer unlucky stretches but also fewer lucky streaks.

[Surutcra]

Totally agree with everything you said !

I believe the average proc interval is 60s / (RPPM_rate * (1+haste)) in their mind, but the thing is that this will not be an accurate measurement of the expected time between procs once they make this change... so it's a bit unclear, obviously !

I will do the math to compute the probabylity distribution, average and variance of the time interval between procs tonight, but obviously it will still be highly random, because as you said they merely removed the long tail of the distribution, the relatively high chance to proc immediately still exists.

Since there are 2 different numbers (10% and 5%), I guess it's unrelated to the effects of the new system and just a bump to the basic RPPM rates.


Although it's better than the previous system, I also am a bit disappointed by this change. I had done a pretty thorough mathematical analysis of the RPPM system and of an alternative design that transforms the probability distributtion of the time between procs to a normal-ish one with way lower variance here : http://www.mmo-champion.com/threads/...gn-proposition

So yeah 3 main issues with the system :
- Still higly variable (can still get very lucky vs. a bit unlucky, only the very unlucky cases are removed)
- The average proc frequency is not RPPM_rate * (1+haste) anymore, it's a bit more. I will do the math to know how much after tonight's raid
- If you want to get a guaranteed proc right on pull (or rather maximize the chance you do), you need not to get a proc for 1000s which is ~17 minutes. That's a fucking long time...

[Reith]

Lastly, as just a commentary, I don't like this change at all. This isn't just adjusting RPPM's pseudorandomness to make them more reliable. It's just shaving off the bottom of the bell curve on bad RNG without touching the top of it, which means the RPPM number is no longer an accurate measure of how many procs you should expect per minute(after adjusting for haste). I was imagining they'd find a more clever approach that would maintain the exact same number of average procs per minute and just steepen the bell curve's hump, meaning you'd see fewer unlucky stretches but also fewer lucky streaks.

They don't want to nerf the good parts of the trinkets. They want to keep potential lucky streaks in. The problem wasn't the lucky streaks, it was that sometimes you'd go forever without getting a proc - and that had to be fixed.

[Kaedis]

Well, here's some notes from playing around with the formula:

- The proc chance is static until 1.5 times as long as the average proc interval.
- After that point, you start getting a continuous multiplicative increase of 100% per (AvgInterval/3.0) seconds. For example, if the average interval is 15 seconds, you gain a 100% (multiplicative!) increase to the proc chance every 15/3 = 5 seconds. If you started with a 15% proc chance, the chance would begin increasing at 22.5 seconds since last proc, and gain 15% (additive) to the proc chance every 5 seconds (on a continuous distribution, though), so at 27.5 seconds you'd have a 30% proc chance.
- The time necessary to reach 100% proc chance out of combat can be computed as (((1/ProcChance%)-1)/3.0+1.5)*AvgInterval. Since we're out of combat, however, proc chance simplifies to the maximum chance at 10 second since last procable event, which is (RPPM * Haste%)/6.0. Similarly, assuming Average Proc Interval reduces to 60 / (RPPM * Haste%), the equation simplifies (sorta) to:

120/((RPPM * Haste%)^2) + 70/(RPPM * Haste%)

Assuming 0% haste, 1 RPPM effects require 190 seconds, 2 RPPM 65 seconds, and 4 RPPM 25 seconds. Anything higher than 8 RPPM runs into issues with this formula because the expected time to 100% proc chance drops under 10 seconds, reducing the time-since-last-procable-event.

[Surutcra]

They don't want to nerf the good parts of the trinkets. They want to keep potential lucky streaks in. The problem wasn't the lucky streaks, it was that sometimes you'd go forever without getting a proc - and that had to be fixed.

Actually if you think about it rationally lucky streaks are just as big a problem as unlucky ones, though I definitely understand how you may feel differently about both. Unlucky streaks feel terrible in comparison to normal ones, but the normal ones are just as bad compared to lucky streaks ! Both lucky and unlucky situations contribute to the resulting high variance that make the value of the trinket so unreliable.

Don't get me wrong, I do like unpredictable trinkets in terms of when is it going to proc. But the RPPM system is so random that the uncomfortable uncertainty is about how many procs am I going to get from this trinket over the whole fight. You should be unable to predict when your trinket will proc during the fight (and thus, skillfull players need to react to trinket procs, not plan for them), but you should have some degree of certainty over the number of procs you're gonna get total (otherwise your DPS can vary by 10-20 % just because of luck/unluck regarding trinket procs (and that extreme randomness is not especially fun).