SimulationCraft for Warlocks (Panda Edition) - Page 12

jason1975 · 07/18/12, 11:37 PM

Originally Posted by Seiru

Just wanted to point out GC's comments on Warlock SimC numbers here:
Beta Class Balance Analysis - Forums - World of Warcraft

Specifically:

We aren’t able to debug your simulations, but will say that those very much do not reflect the numbers we’re seeing. We’re seeing Destruction slightly above Demonology (1-3% difference) on a single target, and Affliction slightly (3-5% difference) above Destruction. Those are gaps that we expect to tune tighter by ship. I’m not sure if your simulation of Affliction/Demonology is too high, or if your simulation of Destruction is too low, or if our results are off. Providing statistically significant logs of these performances would be very helpful.

In terms of generating statistically significant logs, this is one area where my knowledge of Simc is very limited. I know how to generate a combat log, but it always says "iterations 1" so I'm a little afraid I'm not doing it right. Can anyone take GC up on his offer and generate the logs and post them? Or give me the info on how to do it? Thx . . .

• Zakalwe · 07/18/12, 11:51 PM

He's talking about logs of in-game performances.

• Zakalwe · 07/19/12, 12:03 AM

The biggest challenge in trying to keep the simulation accurate during this beta has been the fact that Blizzard removed most spell power coefficients from the client's spell data. The coefficients used to be available for us to extract, but for most warlock spells they no longer are - there's only a textual representation of them in the tooltip itself, which very often ends up being inaccurate because Blizzard don't update the tooltip text when they tune their numbers.

Case in point: I just discovered that shadow bolt's spell power coefficient is 125%, not the 150% claimed by the tooltip. Another example is drain soul, whose tooltip has always claimed a coefficient of 200%, but which has had a 150% coefficient for a long while, and which in the latest patch has been nerfed by a large amount without the tooltip changing from the original 200%.

The shadow bolt coefficient should have a minor impact on demo's numbers, while affliction will obviously lose a significant amount as soon as the DS nerf is in. With those adjustments, we may not be quite so far off from GC's internal numbers.

Honestly, I'm pretty confident in the actual mechanics of the sim at this point, but if someone wants to go over the spell power coefficients and actually test them in-game, that would be great. The ones we use in the simulator are included in the HTML report in the details for each spell.

FlyingCow · 07/19/12, 2:28 PM

I tested nearly every source of our damage in Beta Patch 15851. The results can be found here: Mists of Pandaria: All Specs

I found several variations compared to supposed values:

Shadow Bolt's spell power coefficient is 125% (150% in the tooltip)
Unstable Afflictions's spell power coefficient is 40% per tick (20% in the tooltip)
Drain Soul's spell power coefficient is 150% (200% in the tooltip)
Immolation Aura's spell power coefficient is 25% (30% in the tooltip)
Shivarra deals 40% more autoattack damage then Succubus (supposed to be 20% more damage)
Wrathguard deals 50% more autoattack damage then Felguard (as a result of dualwielding and 50% offhand damage)
Wrathguard's special attack (Mortal Cleave) deals 50% weapon damage (150% in the tooltip), only half of Felguard's style damage
Felstorm does the same damage as Wrathstorm (100% weapon damage per second), 33% spellpower coefficient in the tooltip has no effect

realCool · 07/19/12, 7:52 PM

As the new build should have no influence on the singel target priority list of demonology (except that you don't have to wait for an Molten Core Proc anymore to trigger decimation, and as such had no special influence on the priority list) I tried to change it so that the benefit of Dark Soul would be used better. At the moment it is used at the start of the fight where meta is only up quite shortly. So I tried adding the following:

actions+=/blood_fury,if=demonic_fury>=750|target.time_to_die<=25
actions+=/dark_soul,if=demonic_fury>=750|target.time_to_die<=25
actions+=/service_felguard,if=talent.grimoire_of_service.enabled&buff.dark_soul.up

This was also done to assure that all 3 cooldowns are stacked as they have the same cooldown. This was a dps decrease, because the average number of used Dark Souls in a patchwerk fight with 450sec and 20% variation decreased from 4.3 (19% uptime 28% benefit) to 3.9 (16% uptime 25% benefit)
When I started decreasing the demonic fury value in the if condition the dps started increasing again, reaching a maximum at 250 fury, which now gives an ~700 dps increase to the standard action list. What I don't understand is that with a value of 500 I already reached the same number of average Dark Souls used. So at first I thought the further increase in DPS comes from the earlier application of doom on the target (as meta is now used later on then normally). So I included the following conditional:

actions+=/metamorphosis,if=buff.dark_soul.up|dot.corruption.remains<5|demonic_fury>=900|demonic_fury>=target.time_to_die*30|dot.doom.remains<5

This did not change anything. As such I am unsure where the DPS increase from further decreasing the demonic fury value comes from.
All simulations done with the simc 503.1 release version and the Warlock Demonology T13H profile

ultrajustin · 07/19/12, 9:06 PM

Last I tested the corruption mechanic with demonology was thus

1) apply corruption
2) turn demon form and when all your buffs are synchronized you recast void ray 1 time, this updates the corruption to your new stats/demon form bonus
3) afterwards touch of chaos infinitely extends the extra powerful corruption

I am under the impression the sim may use this technique in some manner, thus maybe your simulation is waiting too long for fury and is losing its corruption.

• Zakalwe · 07/19/12, 11:08 PM

The sim does not model the lack of buff recalculation when touch of chaos extends corruption. It's obviously not something that is intended or desired.

• Zakalwe · 07/19/12, 11:58 PM

After going over the results of some testing done by others (thanks realCool and FlyingCow), I'm quite confident that the spell power coefficients we use are correct. Yes, there are some discrepancies between the tooltips and the actual in-game behavior, but I appear to have caught the last of those when I discovered the correct shadow bolt coefficient the other day.

I think a more likely reason for the discrepancies between our results and GC's internal numbers is simply the complexity of our action lists and conditionals, plus other extraneous things like the specific gear/trinkets/glyphs/talents used by our profiles, when we pop bloodlust, etc. To explore what sort of effect these things might have, I made extremely simplified versions of the three pre-raid profiles, with no glyphs, no trinket usage, no potion usage, no bloodlust, no doomguard, etc. I also simplified the action lists and removed any tricks to squeeze out extra DPS or synchronize cooldowns with dots or anything like that, and I made each profile use Grimoire of Supremacy for a nice static pet contribution.

Basically, these profiles run through a perfectly naive version of each spec's basic rotation, and any external influences that might provide synergies that GC's internal tools might not take into account have been removed. The end result is very interesting:

These results perfectly match the relative differences between the three specs that GC described in his post. And this is my theory for why their internal numbers don't match ours - I think their tools simplify the situation too much. Only a theory, of course - I have no idea what sort of tools they have at their disposal or how they use them - but it's a possibility worth keeping in mind.

Fallensaint · 07/20/12, 2:38 AM

Found another hidden mechanic, this time for Destro. If you spec GoSac and have the buff active then FoX doesn't cost any embers. As far as I can tell, this particular effect only happens with Destro. There is no comparable effect with Affliction, though Demonology does have a brief window where they can use Demonic Rebirth to recover their pet.

Also, GoSac boosts Shadowburn in the beta, though it's not really hidden I'm not sure when it was slipped onto the list.

With a free, instant summon and a bit of additional damage, I'm curious to see how the talent shapes up.

krno · 07/20/12, 9:02 AM

Since the latest build Chaos Bolt has its cast time reduced by 30% when backdraft has more then 3 stacks, but it does not comsume then. Guess this is a bug.
And it seems like Chaos Bolt proccs from DTR comsume Burning Embers.

Sartré · 07/20/12, 11:42 AM

Nice find Zak, please do report this in beta class balance thread and we hope GC shares how they handle finding their own numbers he mentioned.

Edit: Oh it seems you have already done this. Nevermind.

realCool · 07/21/12, 8:10 AM

I think the reason for the dps increase with decreasing the fury value is the improved corruption. As Dmg modifier and int procs are still snapshotted for dots it is an dps increase to get the best possible corr version (meta + trinket/weaponenchant procs + Dark Soul + Blood Fury) on the target as early as possible. That means you should probably start quite soon so that corr is rolling with that in a single target fight. It should then be a dps increase to get a higher amount of fury for subsequent dark souls, but looking at the demonic fury resource graph this seems to be the case nevertheless.

Nardale · 07/21/12, 1:17 PM

I see the resource per second gain and loss of health in the sims, but they do not include any tier one regeneration talents. Are they modeled at all? I am away from my desktop so I can't test in SimC for a few days. I just want to see the net health over a fight so maybe we can get some hard numbers in the beta class balance thread.

• Zakalwe · 07/21/12, 1:22 PM

Dark Regeneration is not currently supported by the sim, but Soul Leech and Harvest Life both are.

zinnin · 07/27/12, 4:13 AM

Is Kil Jaeden's Cunning modeled in simcraft so we can test if it is actually a dps increase while moving?