Theck's Simcraft Thread - WoD/6.x

Thanks for the explanation.

Some L90 stuff that's relevant to Tuesday.

Talent sim:
https://www.dropbox.com/s/4hzvxk1xlxj4n ... .html?dl=0


DP is the big winner at L90, but SW and HA aren't really that far behind. This is against a T17 Mythic boss, so bear that in mind.

Stat weights are also fairly predictable:


Stamina wins because we have a boss that's hitting way too hard. But other than that BonusArmor>Mastery>Versatility>Crit>Multistrike. Note that haste's low valuation here is because we already have nearly 50% in this gear set, so we're going over cap when we add haste to determine the stat weight. I have some sims finishing up now that should give a better idea haste's real value.

Strike that - the actors in those sims are level 100, not level 90. :/ Re-running.

So, I've been doing some more investigation of this L90 stat weight issue, because when I re-ran the stat weight calculation for different talent combos and gear sets, I got this:

Talents:
https://www.dropbox.com/s/4hzvxk1xlxj4n ... .html?dl=0





Gear:
https://www.dropbox.com/s/tl4vc7w7alxnx ... .html?dl=0



Crit config:


Haste config:


Mastery config:


Now, there's some weirdness here - I have this sim from a few days ago showing haste being awesome:


What gives?

There were a few obvious things to check, which is why I ran those extensive sims.

• First: maybe it was an EF vs. SS thing? But looking at the data from the talent sim, it's clear that haste doesn't fare any better/worse with EF (in fact, it's often much much worse!). Note that the gear set used for the talent sim is mastery-heavy, so that we're in no danger of hitting the 50% cap.
• Second: maybe it was a L75 talent issue (DP vs. SW for example)? Again, the talent sim data seems to refute this pretty handily.
• Third: maybe it was being near 50% haste? Obviously refuted here by both sims. Even in gear sets where we regem entirely for mastery or crit, haste isn't picking up in value.

The next thing I guessed was that it was an issue of local linearity. The anomalously high haste value was generated in a sim with a scale delta of only +37. That's pretty small (about 1.85% haste), and I've since increased the default scale delta to +70. So I re-ran a particular gear set (~43% haste, ~41% crit, ~32% mastery) with both EF_DP and SS_DP talent combos with the larger scale delta, and a centered interval (so +/-35 instead of +70):

EF_DP:
https://www.dropbox.com/s/ovchi3kcj17ab ... .html?dl=0

SS_DP:
https://www.dropbox.com/s/nh44sdt68af5a ... .html?dl=0


Wait, what?? How did haste climb on top again??

At this point I was pretty sure this was a local linearity issue. In other words, haste isn't performing perfectly linearly, and we're at a weird "bump" in the curve. So I ran some plot sims for haste/crit/mastery to see what was up. Here's the results:

EF_DP:
https://www.dropbox.com/s/dopqp1rpoi6g5 ... .html?dl=0


SS_DP:
https://www.dropbox.com/s/t9aa3mayjf78l ... .html?dl=0


You'll note that in both of those plots, the haste curve looks really good from 0 to +50, but then falls significantly behind. On the right-hand side, it's partly because at around +130 haste rating we hit the 50% cap, so it will clearly level off there. On the left-hand side, though, it performs about as well as crit does.

This at least tells us what was happening: between ~43% and 46% haste, haste performs really well compared to the other two (especially in the EF sim). So if we sim a character and generate stat weights that are only sampling this part of the curve, haste performs well. But everywhere else, it really only performs as well as crit (maybe a little better). Mastery, on the other hand, kicks ass just about everywhere.

So, I think we've solved our mystery. The question of why haste is performing so well right around that range isn't entirely clear, but at least we know why the stat weights were wonky.

If I'm interpreting those stat scale graphs correctly, at some breakpoints increasing haste actually negatively effects your survivability (positive slope on the graph). Could this be due to HPG push-back (as your ability cooldowns become shorter with haste, you will face decisions like wait 0.2sec for a HPG, or use Cons right away and delay the HPG until your next GCD)? I thought that this was only an issue at extreme (>50%) levels of haste, but I can't see how else Haste could make you less survivable.

Edit: Wait this can't be the cause or else we should expect similar breakpoints regardless of EF or SS, which is clearly not true (I suppose SS does cost a GCD to refresh, unlike EF, but given the stochastic nature of AS+ procs you won't be refreshing it at the same points in each simulation).

Edit the second: Hmm looking closely at the SS graph there is a decrease in haste's value at around the same point, but it's tiny compared to the EF graph; the slope just becomes a bit less negative, it never actually becomes positive as it does for EF.

Okay, totally embarrassing moment here because it makes me feel like a total newbie and may have just overlooked this info somewhere:

How do I read those graphs? Am I supposed to be reading them left to right or right to left? Am I supposed to start at 0 and work my way to whichever direction? And if I'm reading them left to right, what's with the negative values?

I understand everything else, but the graphs just make me go @.@

@ Taikishi :

The more steep is the slope, the more stat variation has an impact on TMI, whichever direction you choose.

In the 2 graph above, crit mastery (color error ><) seems better for TMI, except in range [actual stuff (or 0); haste cap], where haste is better.

Négative values show reduction in stat, and resulting TMI.

So that I can confirm I'm understanding correctly, then, because I think what was throwing me off was the deltas but a quick glance at https://code.google.com/p/simulationcra ... atsScaling made it click:

0 is our current haste/mastery/crit ratings. The deltas below are how much rating we gain/lose in that particular stat and how much it affects how likely we are to take damage spikes. The steepness I understood, it was how you concluded 43-46% was the ideal range for haste but now I think I understand. :D

This right? :)

Finkum wrote:If I'm interpreting those stat scale graphs correctly, at some breakpoints increasing haste actually negatively effects your survivability (positive slope on the graph). Could this be due to HPG push-back (as your ability cooldowns become shorter with haste, you will face decisions like wait 0.2sec for a HPG, or use Cons right away and delay the HPG until your next GCD)? I thought that this was only an issue at extreme (>50%) levels of haste, but I can't see how else Haste could make you less survivable.

Edit: Wait this can't be the cause or else we should expect similar breakpoints regardless of EF or SS, which is clearly not true (I suppose SS does cost a GCD to refresh, unlike EF, but given the stochastic nature of AS+ procs you won't be refreshing it at the same points in each simulation).

Edit the second: Hmm looking closely at the SS graph there is a decrease in haste's value at around the same point, but it's tiny compared to the EF graph; the slope just becomes a bit less negative, it never actually becomes positive as it does for EF.

Yes, there's definitely something anomalous happening there, but I don't know what it is. It certainly could be HPG push-back caused by getting near or over the haste cap, though.

This actor has 43.42% haste to start with, so it only takes 6.6*20=132 haste rating to hit the cap. The dip is occurring between 0 and 75 rating though, with the low point at 50 rating (note we're only testing every 25 rating for this - I'll re-run with finer granularity soon to investigate). However, we're also using Windsong, which grants 75 haste rating fairly frequently, which explains why we start seeing the upturn between 50 and 75 rating.

Also note that we also have Bloodlust for the first 45 seconds, and during that haste's value is very low (just SoI/SS/EF increases, basically). So that could be artificially reducing the value (if we're interested in just our survivability outside of Bloodlust).

edit: of course, it doesn't explain why we're getting such a large dip there in the first place, but *shrug*. My first guess would be EF/SS breakpoints, but I wouldn't expect those to be such a significant factor anymore. It could also just be that haste's value increases dramatically between 40% and 50% because of multiplicative effects like Divine Purpose. If I find time I'll investigate that further to try and figure out what's actually happening in that region.

Taikishi wrote:So that I can confirm I'm understanding correctly, then, because I think what was throwing me off was the deltas but a quick glance at https://code.google.com/p/simulationcra ... atsScaling made it click:

0 is our current haste/mastery/crit ratings. The deltas below are how much rating we gain/lose in that particular stat and how much it affects how likely we are to take damage spikes. The steepness I understood, it was how you concluded 43-46% was the ideal range for haste but now I think I understand. :D

This right? :)

The plot shows TMI vs. amount of stat. On the x-axis, 0 is the profile's default stats, positive values indicate we've added that much stat, and negative values indicate we've removed that much stat. Since TMI uses golf rules, we expect it to go down as we increase our survivability (i.e. add stats), so we expect the plot to have a negative slope.

A stat that increases our survivability more effectively will have a larger negative slope, because it will make TMI go down faster. From the plot, you can see that mastery makes TMI decrease the fastest, while crit and haste are slower (apart from the strange dip in the haste curve that we're discussing, of course).

Yeah, once I figured out the plots on the X-axis it started to click. I apologize for asking you all to explain that but do appreciate it! Everything else made sense then I saw the graph and was like "huh?" and couldn't find anything anywhere else in my (albeit limited) Google Searches to explain it.

I'm curious, though, especially as someone who:
a. rather enjoyed the Haste play style this expansion
b. will likely never see Mythic due to only having about 12 people in the guild

Do you suspect that Sacred Duty will ultimately be tweaked due to Haste no longer being our best stat, that Divine Bulwark could be tweaked so we're not double-dipping from Mastery, or do you think that Blizzard will make other corrections to encourage gearing for more Haste? It just seems odd we'd be given Sacred Duty after an entire expansion of Haste only to have it come out as our worst stat for TMI.

I know they're aware that haste is underperforming, and that they're not really happy about that. I've also suggested a few tweaks that could help bring haste's value up. So I wouldn't be surprised if some changes get made, but it's unlikely they'll do so at L90. It'll probably be changes on beta that will only take effect closer to WoD release.

Oh, I don't anticipate level 90 changes. Not worth their time and effort. It's level 100 I'm concerned/curious about.

theckhd wrote:...of course, it doesn't explain why we're getting such a large dip there in the first place, but *shrug*. My first guess would be EF/SS breakpoints, but I wouldn't expect those to be such a significant factor anymore. It could also just be that haste's value increases dramatically between 40% and 50% because of multiplicative effects like Divine Purpose. If I find time I'll investigate that further to try and figure out what's actually happening in that region.

Even if EF/SS haste breakpoints are still a thing, surely adding more haste outside the breakpoints shouldn't make you squishier than you otherwise would be, as the graphs imply (just double checking to make sure I'm not misunderstanding: negative slope = decreases TMI score as you stack the stat, positive slope = increases TMI score as you stack the stat). In fact I can't think of any situation where stacking more of a stat makes you less survivable (ignoring the opportunity cost of replacing it with a "better" stat).

Random hypothesis: if the boss in your sims has some sort of big nuke on a specific timer, which if in combination with a melee would spike the tank low, then at specific haste thresholds you would have a EF tick/SS bubble that falls between the nuke and the melee. Outside of those thresholds you don't get the heal in-between and so the tank spikes lower, which gives you a larger TMI value. (Although I seem to remember you addressed something like this problem a while back by adding a random delay to the boss's swing timer?)

Schroom wrote:

Note that Shining Protector cannot crit (at least, as of my last tests), which is another factor in Multistrike's low valuation.

:o nice to know, that's a shame.

It makes sense, though. Stuff generally doesn't doubledip, and the original heal that is boosted by Shining Protector CAN crit. If the original heal crits, then obviously the 30% extra is also a lot bigger.

theckhd wrote:

Fundus wrote:

theckhd wrote:Stack enough of your attunement stat - say haste - and crit and mastery will eclipse it just because of the multiplicative nature.

So perhaps this is a silly question, but could this also be part of what's happening with haste being ranked so poorly? The default gear set is fully enchanted for haste, totaling 1433 haste from gear/enchants. That's compared to 220 versatility, 622 multistrike, 666 crit, and 758 mastery. In addition we're using a weapon enchant that procs 1000 addition haste (declining by 100 haste every 2 seconds for 20 seconds). If we used a gear set that more emphasized mastery/versatility, since those tend to be the top 2 stats, how greatly would that affect stat weights?

Yes, it's entirely possible that it's partly due to the default gear setup. Or rather, I think that's exacerbating the problem. I still think that haste is going to be weaker than intended due to the swath of nerfs to different mechanics that caused haste to be strong in MoP in the first place. But stacking lots of it would obviously devalue it further. One of the sets of trials I want to run (if nobody beats me to it) is trying a few different gear sets (heavy mastery, heavy multistrike, etc.) to see how the stat weights vary. Unfortunately, stat weights are very time intensive, so I'd probably have to settle for running it on a very limited talent spread (i.e. maybe just SS_SW_HolyS for example).

What about a single balanced set that includes almost even amount of rating from all six secondaries? It might be more representative of how we'll be gearing up initially, grabbing whatever we get, and which direction we should take from there.

Finkum wrote:Random hypothesis: if the boss in your sims has some sort of big nuke on a specific timer, which if in combination with a melee would spike the tank low, then at specific haste thresholds you would have a EF tick/SS bubble that falls between the nuke and the melee. Outside of those thresholds you don't get the heal in-between and so the tank spikes lower, which gives you a larger TMI value. (Although I seem to remember you addressed something like this problem a while back by adding a random delay to the boss's swing timer?)

The boss in those sims was a T17M TMI Standard Boss with the level artificially dropped to 93. It only has auto_attack and spell_dot attacks. I'm more inclined to believe it's something affecting the rotation that we're just not thinking of.

Thels wrote:What about a single balanced set that includes almost even amount of rating from all six secondaries? It might be more representative of how we'll be gearing up initially, grabbing whatever we get, and which direction we should take from there.

The T17 gear sets have much less rating, so they're a far closer approximation to that. It's just the T16M gear set that's running into haste-cap issues.

Also, I re-ran the plot sims overnight with a 5-haste step size. Here they are:

EF_DP:
https://www.dropbox.com/s/42nxalbeahxtz ... .html?dl=0


SS_DP:
https://www.dropbox.com/s/6y2i6eev7lfj5 ... .html?dl=0


So it's a very abrupt drop-off.

Also, a side note: yesterday I tested SoT on beta, and found that they had removed the hidden, undocumented 20% multiplier (SoT procs were doing 1/5 as much for prot as for ret pretty much all expansion, thanks to a nerf in early MoP). All of our L100 Empowered Seals stuff was done with that nerf active, so I'm hoping to revisit that stuff over the next few weeks. For now I'll blithely assume that our TR rotation is still ideal and re-run just to see how much more damage that nets us. I may also add Sera_DPS entries to that list now that we have an optimized APL for it.

You may have already seen these from yesterday but also:

Celestalon wrote:As you may have noticed, we don't patch note damage/healing tuning number changes. Normally, datamining catches most of these changes, so theorycrafters know when to retest things. They only are able to do this when we make new builds. However, we've hotfixed some more tuning changes in, since the last beta build. The client should reflect these changes in tooltips, however datamining likely will not. As such, I'm providing that info here. Once we go live tomorrow, we'll use the normal hotfix blog for this sort of thing.

Paladin
Avenger’s Shield damage increased by 5%.
Consecration damage increased by 5%.
Holy Wrath damage increased by 5%.
Shield of the Righteous damage increased by 5%.

I doubt that those are enough of a change to vastly effect the outcomes of the sims, but figured they should be in there if you're re-running.

Yep, they're already coded in (though, obviously not in any data that precedes them).

In other news, I still haven't figured out what was causing that haste anomaly at 43%. The symptom is rather clear, but the cause is a little more confusing.

I ran the profile in question with varying amounts of haste rating added (0, 50, 60, 65, and 70). The 60 profile is at the bottom of the dip, and the 65 profile is the one in which it loses survivability. The sims are here:

https://www.dropbox.com/s/tgjd1p4qwalc6 ... .html?dl=0
https://www.dropbox.com/s/xierrae3b3gr4 ... .html?dl=0
https://www.dropbox.com/s/94n1yxdjmt3xc ... .html?dl=0
https://www.dropbox.com/s/lrrm9dqvj7jmq ... .html?dl=0
https://www.dropbox.com/s/fxmeo3zfcw5uy ... .html?dl=0

What's notable is the difference in HPG gain:

Code: Select all

Haste  HPG/s  CS       J      GC     AS    Cons  HW    HoW   LH
0      0.57   141.34   95.40  17.18  49.9  54.9  35.8  15.3  7.8
50     0.60   148.55  100.41  17.19  49.3  54.8  34.2  15.5  7.7
60     0.60   148.68  100.36  17.21  49.3  54.8  34.2  15.5  7.7
65     0.59   145.10   98.00  17.19  50.9  54.1  37.3  15.4  7.7
70     0.59   145.13   98.02  17.23  50.9  54.1  37.3  15.4  7.7

It's clear that the dip is caused by an anomalously large surge in HPG (and thus SotR coverage), and the loss of TMI at 65 is similarly due to a sudden drop in HPG. The culprit appears to be less CS and J casts, and more AS/HW casts (though not more GC casts).

I've combed through the sample sequences, and can confirm that both inside and outside of windsong_haste, it's following the proper CS-J-X-CS-X-J-CS-X-X rotation. The one oddity I did find is this, in the 0-haste-added parse:


What you're seeing there is that it performs

CS-J-X-CS-X-J-CS-X-J-CS-X-X-

In other words, it's getting an extra CS-X-J sequence in there, because windsong_haste (indicated by the underline) caused a short J cooldown but a long CS cooldown. I wasn't able to find any similar oddities in the +60 or +65 haste setups. That said, I don't think this explains anything - it should increase HPG relative to normal operation for the 0-haste parse, so it wouldn't explain why the 50-60 haste parses are seeing a big drop.

Next guess is that it's something happening during bloodlust. I'm going to try and re-run the plot with BL and windsong disabled (individually) to see if it's some sort of edge effect due to either of those buffs. If anyone else spots anything suspicious in those reports, let me know.

In any event, I'm feeling more certain that this is some sort of rotational anomaly we're seeing in SimC and not a real effect.

Updated L45/75/100 talent sim, with the SoT bugfix and 5% hotfixed buffs added:

https://www.dropbox.com/s/j6lwblhkwr5wa ... .html?dl=0

Glyph results. Base setup is just Divine Protection (DP_only). The rest are all divine_protection + the glyph named.

Single-target:
https://www.dropbox.com/s/rocr1vn247m7h ... .html?dl=0


Two targets:
https://www.dropbox.com/s/mq9h019i6any0 ... .html?dl=0


Three targets:
https://www.dropbox.com/s/ed8ian70me2ce ... .html?dl=0



Note that Immediate Truth and Word of Glory are basically irrelevant here because we're using SoI and not using WoG at all.