ok, I've read a lot of the old 300+ page Enhance thread and I think I have a pretty good understanding of what is good and bad for an enhance shammie but I have a dumb question about Yo's Sim

When I ran a sim a month or so ago with gear setup A and found weights etc etc and plugged those into lootrank.com I was given a list of best pieces to get.

Now when I run a sim with my current gear some of those items that were deemed awesome fall way down the list and some of the items that I replaced are deemed better.

This confuses me. Can anyone help me understand this? or just flat out tell me if I'm being retarded. Thanks.

It all depends on what you put into the sim in terms of stats and buffs. As you gain in one area, say crit, it will become overall less valuable as you approach diminishing returns so the values of other stats will increase.

Remember, you should get a new set of EP weights after every gear upgrade since that will have an impact on the next piece of loot you get.

Also, if you get good pieces of gear that aren't quite upgrades, sometime down the road they may be so hold onto them. I am using [Boots of Utter Darkness] but my [Edgewalker Longboots] appear to out EP them using higher values as I move into epic gems and higher gear levels.

1) 35% average damage reduction for glancing blows, comes from Glancing blow - WoWWiki - Your guide to the World of Warcraft
2) 3% crit reduction for Boss mobs, comes from Critical strike - WoWWiki - Your guide to the World of Warcraft
If you find this information outdated please point to some info on it.
Viewed few wws reports, they suggest close to 25% average damage reduction for glancing blows indeed.

If we consider:
Melee Combat Mechanics

which simply is more recent and way more consistent than the wowwiki posts.

Glancing blow is 25% (I think the 24% in the link above is a typo)
CritReduction is 0.04% per Skill difference point, e.g. 0.2% per Level = 0.6% less Crit.

So basically I think the current values are:
-0.6% crit
-28% hit (9+18%)
+1.5% dodge

So we should use 28% hit penalty and 0.6% crit penalty and use 6.5% dodge and 25% glancing probability and 25% glancing reduction.

But I think we should really look into the crit reduction a bit further.

I tried looking for the awnser to my question but i was unable to find it.

Currently i have T6 bracers/belt, wil get the boots next week probably if they drop. Now i was wondering if its worth taking a 4th T6 part for the 4piece set bonus.

If yes, which piece would you suggest i take?

My guild allows me to loot leather gear (rogue's do not get preference), so i basically have acces to the best possible gear for the T6 slots.

Thank you for your input

The ideal setup does not include 4 pieces of T6. However, 4 piece T6 might be an option while you're acquiring the rest of your ideal setup. It depends on which gear you've already picked up. The EP difference between the 4th piece of T6 and the item you're replacing to use it has to be less than 70.

Answered in his FAQ:

Yep, figured as much... it's just 5am here and I'm having a stupid moment

Hmm looks like I'm gonna write my own mini simulation version for the enh shaman. I need to find out why all the closed form models are +hit heavy while the simulations are not and the reality is somewhere inbetween.

Especially my reworked model is really one the +hit side, while being highly detailed and accurate, if i hardcode the flurry vvalues to 85% (result of Yo's sim with the testgear) and without haste and expertise, the result equation of melee dmg looks like this (for one specific gear only)
finaldps := 886.6602909+9.344605262*hit+9.143708172*crit+.9729473683e-1*crit*hit

obiously in this gear +hit is more worth that +crit.

Might take awhile though, but currently I have no idea where the difference is, since with Yo's answer I can account for every difference in results between my model and his sim, the only difference is that mine is +hit heavy.

Tornhoof --
In my experience, Yo's simulator is very close to correct. Taken over time, every upgrade it has suggested so far has increased my DPS by roughly the amount predicted, and many others have the same experience.

Thus, I find it likely your closed form equations are missing something important. Don't waste your time creating a mini simulation, chances are it'll be based on the same logic. Instead, post them somewhere where they can be critiqued.

Hmm, I will rewrite my closed form model in an easily digestable way here.
My previous experience just is, that nobody takes their time to discuss these models.

Heya, just wanted to add a small tidbit of info in hope that Yo´s! sees it and adds it to the simulator.

When doing the stat weights for Thottbot the line given is missing the value for expertise, that is because thottbot internally calls it Skill Rating, it is the number that follows the value of hit rating.

For example the sim gives me this line:

http://thottbot.com/score/B2.2,1.95,0,0,0rP0,0,0,1.83,2.0,1.8,0,1,0,0,0,0,0.34sdO22,22,22,70A,,,0,0W,0,0,,,0,,0,0,0,0,0

And in the main page i see that expertise rating actually has a 3.47 value, the line with expertise rating accounted for thottbot should look like this then:

http://thottbot.com/score/B2.2,1.95,0,0,0rP0,0,0,1.83,2.0,1.8,3.47,1,0,0,0,0,0.34sdO22,22,22,70A,,,0,0W,0,0,,,0,,0,0,0,0,0


It is just a small change, but with it thottbot turns into actually a very reliable loot comparison tool.


EDIT TO ADD:

damn i forgot to turn off the auto parse links, so fixed that and bolded the 0 of the sim and the correct value example.

Ok let's see about this model:
float asmh = ws1 / ((1 + flurryvalue) * (1 + haste / 100)); // fully flurried attack speed
float asoh = ws2 / ((1 + flurryvalue) * (1 + haste / 100)); // fully flurried attack speed
float attackspeed = (asmh+asoh)/2;
float aatime = attackspeed /2 ; // how much time for each auto attack (3*attackspeed/(2*3)
float flurryprob = min<float>(1 - pow(noncrit, 3 + 2 * (1 - dodge / 100.0f) * (1 - pow(1-wfprocprob,aatime) + aatime * (3/10.0f))),1.0f);

The flurry pro bability is fine so far, it was discussed and analyzed before and should do the trick.

Now the interesting part:
float wsmh = ws1 / ((1 + haste/100) * (1 + flurryprob*flurryvalue));
float wsoh = ws2 / ((1 + haste/100) * (1 + flurryprob*flurryvalue));

flurryvalue is 0.3 usually.

The idea for the new model is not modelling the pure dps, but modelling the swings we do and calculating their dmg.

So Runtime is 1000 for now (1000s)
float runtime = 1000;

1. we need to know how many normal white mh and oh hits we do:

float mhhitcount = (hitwhite - dodge) * 1/100.0f * (runtime/wsmh);
float ohhitcount = (hitwhite - dodge) * 1/100.0f * (runtime/wsoh);

We have these now, nothing big there, we divide the runtime by weaponspeed and multiply by the attack table

2. we need to know how many stormstrike hits we do:

float ssmhhitcount = (hityellow - dodge) * 1/100.0f * (runtime/10);
float ssohhitcount = (hityellow - dodge) * 1/100.0f * (hityellow - dodge) * 1/100.0f * (runtime/10);

We do stormstrike every 10s, and the oh ss only gets done if we hit with the mainhand. nothing unusual here though.

3. we need to know how many windfury procs we do from white and from ss, we assume 36% proc rate.

float wfprocmh = ws2 / (ws1 + ws2) * wfproc; // mh wf proc prob
float wfprocoh = ws1 / (ws1 + ws2) * wfproc; // oh wf proc prob
// hit count calculation
float wfmaxcount = runtime/minwftime;
float wfprocmhcount = (hityellow - dodge) * 1/100.0f * wfprocmh * mhhitcount;
float wfprocohcount = (hityellow - dodge) * 1/100.0f * wfprocoh * ohhitcount;

The wf proc probability is slightly skewed towards the faster weapon, otherwise we just do basic math.
every possible white hit might proc windfury up to a certain count. We also need the wf procs from stormstrike

float ssmaxwfprocs = min<float>(wfmaxcount - wfprocmhcount - wfprocohcount,runtime/10.0f);
float ssmhwfproccount = min<float>(ssmaxwfprocs, ssmhhitcount * wfprocmh);
float ssohwfproccount = min<float>(ssmaxwfprocs, (ssohhitcount - ssmhwfproccount) * wfprocoh);
wfprocmhcount += ssmhwfproccount;
wfprocohcount += ssohwfproccount;

We calculate the max count of wf procs at all (basically every ss can proc wf until we reach the wf limit imposed by the 3.0s cooldown). Then we calculate the amount of wf procs from ss mh, then the amount of procs from ss oh (the ss oh can only proc wf if the mh does not proc it).

We sum the procs up etc.
The remaining stuff is straight forward again:

float mhwhitedmg = 0.01f * 1.1f * (100 + crit - glancing) * (wdps1 + ap / 14) * ws1;
float ohwhitedmg = 0.5f * 0.01f * 1.1f * (100 + crit - glancing) * (wdps2 + ap / 14) * ws2;
float mhwfdmg = 0.01f * 1.1f * (100 + crit) * 2 * 1.4f * (wdps1 + (wfap+ap) / 14) * ws1;
float ohwfdmg = 0.01f * 1.1f * 0.5f * (100 + crit) * 2 * 1.4f * (wdps2 + (2*wfap+ap) / 14) * ws2;
float ssmhdmg = 0.01f * 1.1f * (100 + crit ) * (wdps1 + ap / 14) * ws1;
float ssohdmg = 0.01f * 1.1f * 0.5f * (100 + crit) * (wdps2 + ap / 14) * ws2;
float mhdps = (mhwhitedmg * mhhitcount + mhwfdmg * wfprocmhcount + ssmhdmg * ssmhhitcount)/runtime;
float ohdps = (ohwhitedmg * ohhitcount + ohwfdmg * wfprocohcount + ssohdmg * ssohhitcount)/runtime;
float armorreduction = max<float>((MobArmor - m_Equip[50])/(MobArmor - m_Equip[50] + 10557.5f),0);
float dps = (1.0f-armorreduction) * (mhdps + ohdps);

Every white hit can either glance, crit or hit normally for Xyz dmg, same for all the other swings.

That's it.
It should pretty much everything from my point of view. It is still heavy on the hit side.

few changes:
1) The bonus AP from Windfury does not have an offhand penalty, thus:
float ohwfdmg = 0.01f * 1.1f * (100 + crit) * 2 * 1.4f * (wdps2 + (2*(wfap+ (0.5f * ap))) / 14) * ws2;

2) Your white damage crit calculations will be naturally low because you are evaluating crit after you've already removed misses from the table.

You're also removing glances from crits as if the reduction from glances had equal weight to the increase from crits.

ex:
35% of the time, we do 100 damage.
25% of the time, we do 70 damage.
30% of the time, we do 200 damage.
10% of the time, we do none.
With 100 tries, this is 35 * 100 + 25* 70 + 30 * 200 = 11250 damage.
However, your equation is more like this: .9 * (100 + 30 - 25) * 100 = 9450 damage.

Given this, hit will have more value than crit, because in your model you must hit before you can crit. And crit will have less value overall, because you have glances removing bonus damage at an equal rate.

You can fix this thusly: if we hit 90% of the time, we really aren't dealing with 100% any more -- we're only dealing with damage outcomes. So your white damage caluclations, rewritten to reflect this:

float notmisswhite = hitwhite - dodge;
float damageadjustments = (crit * 2 + glancing * 0.7f + (notmisswhite - crit - glancing) * 1) / notmisswhite;
float mhwhitedmg = 1.1f * damageadjustments * (wdps1 + ap / 14) * ws1;
float ohwhitedmg = 0.5f * 1.1f * damageadjustments * (wdps2 + ap / 14) * ws2;
...
mhdps = (mhwhitedmg * mhhitcount + mhwfdmg * wfprocmhcount + ssmhdmg * ssmhhitcount)/runtime;
ohdps = (ohwhitedmg * ohhitcount + ohwfdmg * wfprocohcount + ssohdmg * ssohhitcount)/runtime;

3) I now understand why your model predicted the badge MH over Dragonstrike. Long and short is, you're assuming all WF procs happen as soon as possible. In reality, when misses and dodges are considered, most WF procs occur 2 or more weapon cycles outside of the cooldown. This is resulting in more WF procs in your model than can be expected in game or in simulation. avgwflag > minwftime. Haste will shrink the value of either variable by some amount, but it'll shrink the value of the average lag by more, resulting in a higher value for haste and haste procs than in your model.

I also believe your equations undervalue the imbalance of slow weapons vs fast ones. You do model that a fair number of WF procs stem from SS, and that this gives a slight benefit to the main hand, but you don't model that this removes from consideration a portion of your autoattack swings. Instead, you model SS procs as fill-ins due to MH and OH misses, when in reality 20% or more of all WF procs come from SS. This constant "launch mechanism" for WF procs insulates us to a large degree from increased WF lag due to slow weapons. Again: you've essentially got more WF procs that is possible, and this is clouding the fact that in game and in simulation, slower weapons have a major benefit.

In general, your WF model leaves me very uneasy. Aside from the average WF lag issue I mentioned, there's still a question of MH vs OH bias -- I do NOT buy that this is simply an issue of ratios and weapon DPS, as this does not predict proc rates I've seen in my own logs, which selects for the MH at (if memory serves) around 60% of the time. And because questions of WF are really the most important part of playing a shaman, I'll stick with the simulator for now.

Still -- thanks for sharing your equations man, it's obvious you've put a fair amount of time into this.

Someone asked if RED could be worse than Swift Skyfire Diamond, because of socket requirements.

RED requires 2blue and 2 yellow. Additionally, I'll stipulate it requires cat's swiftness instead of dexterity.
Swift Skyfire Diamond only requires 2 yellows.

As a result SSD provides: 24ap, 6agil, 8str vs 12agil and 3% crit.

This is 32ep vs 6agil and 3%crit. Is 3% crit damage better than 20ep?

<First time posting here>

This question has also got me wondering, i read through the thread and didn't see anything about DW DS testing but i'm also thinking it would be crazy dps.

I have 1 Dragonstrike and have the other one ready to be made come 2.4.3 and was wondering if someone had tested it and could either tell me to not waste my time or tell me it's awesome

1.)
I'm already doing this, I think I wrote (I should have after all) (2*WFAP+ap)/14, the 2 only being a factor for the WFAP
2.) Ah yes, you're right, I apparently oversaw this in this model (well it's the first attempt for a model on swing level, my other models don't have that problem fortunately
3.) I will see into that, thanks, I think you have a point there.


Edit:
I think your point 2 covers, why this model is extremly more hit dependant than my previous models. So it's time to get the wf stuff done right

Yes, most values Ive heard evaluate 3% as between 50 and 70 EP. Just to confirm I ran the Simulator with and without the RED meta (as the only difference) and it came out to about a 25dps loss, roughly 80EP.

It is not that simple -- you are not considering socket bonuses in this case i think. If there are any beneficial socket bonuses with blue sockets then it will be even less than 20 ep against 3% crit. Further making the improvement clearer.

Can anyone confirm this? If it's true I might have to buy that badge OH after all.

Here's the post with the screenshot of his combat log showing a refresh: http://elitistjerks.com/737042-post3567.html

So well there was a discussion going on if I really am a supporter because I have not skilled WF-totem. Well I found no information concerning the benefit of skilling it. I hacked some math together and got a rather surprising result (as long as my math isn't wrong that is).

So lets start. I won't describe every single calculation step, because I used MATLAB to do that for me
I had following model in mind:
Warrior, auto-attacking, WF-totem on mainhand.
So evaluating gives this:
DPSDiff=3/7000000*(100+Haste)*(100+Crit)*ExtraAP

So it seems the difference is only dependend on Haste and Crit of the warrior and the extra AP of the WF Totem.
Well I was wondering how much it really matters, so I have plotted the whole thing and this was the result:

(The different colors are for certain DPS diefferences as the colorbar on the right shows)
The minimal DPS difference is 1.9455 and the maximal DPS difference is 7.6286.
This DPS would be lost if you haven't skilled the WF totem and well it's rather tiny but my model is quite easy but there should be no misses or parries only dodges, but before I continue to make this thing more complex I would like to know if this small model seems to be right or if I messed up everything.

PS: As the old Theorycraft thread is closed and there is no new one, I thought it would fit in here, the thinktank shouldn't be used for discussion like this, at least that's what I understood.