As a Mut rogue, I have a problem figuring out Envenom's damage. When I total my raid-buffed AP (flask, imp. BoM, GotW, imp. Kings, Agi food, and Unleashed Rage, as well as the possiblity of mongoose and SoC procs) and raid buffs like Earth and Moon, my maximum envenom crits should only be hitting for about 5060 (5500 with double goose and disdain) but I am constantly seeing crits around 6100. I have accounted for 3/3 Vile Poisons. I really cannot think of what is buffing me to get these kind of numbers when the formula for envenom's damage does not support it.

Also I've been noticing a higher DPS when I leave out Rupture and just use Envenom. I don't have any WWSs, but in our last two M'uru kills, I had exactly the same buffs and the kill time was only a few seconds off, but my DPS on the latest kill where I only used Envenom was about 150 DPS higher.

You might think so, but no. For determining large differences, such as whether Vindicator's Brand or MH Glaive is a better MH weapon, yes, you can determine it in a relatively small number of trials. But many of the differences we're looking at are relatively small - 10 DPS or less. And in order to resolve differences on that scale, even assuming ideal conditions, takes hundreds if not thousands of hours of combat. And when you factor in the variability of player skill - whether player A is better than player B, whether player C was having a good day or a bad day, and so on - one quickly is forced to acknowledge that no reasonable number of WWS will be able to resolve the differences in, for instance, socketing. There's just too much statistical noise and variance for a small number of samples to do much good.

after finding out this astonishing news (to me at least), does anyone think --this-- is accurate? if so, looks like my best shot is dual trackers is best for what im clearing..

edit: silly typo

Agreed - I propose "KSpree" but anything OTHER than "KS" that isn't equally confusing would be appropriate.

That ranking appears to be at least in the right ballpark. There are no obvious major problems, though the exact ordering is a little questionable in places. You couldn't certainly do worse than following it - though, frankly, you could also just use the spreadsheet for slightly more detailed answers specific to your situation.

Given that Kidney Shot has no place in PvE cycles, I don't see that there's a lot of problem with using "KS" to mean Killing Spree only for purposes of these discussions. That said, if people prefer an alternate term, I would suggest either KSp or Spree - I don't think spelling out the full KSpree is necessary, given that both KSp and Spree are unambiguous.

Hunger for Blood and Find Weakness?

You see that trend because better gear generally has more hit, and your assertion doesn't assume equal levels of gear. The two players would have to have the same exact gear, but one gemmed for hit and the other gemmed for agi, but even then there are way too many other factors that affect dps that WWS wouldn't be helpful for such a minute difference.

Also, hit isn't the only stat that affects how many hits you have. Haste will give you more attacks, including more crits, which scale better now due to prey on the weak.


As for the sample size, as aldriana mentioned, there basically is no sufficient sample size, since the amount of improvement the theory suggests is within the margin of error for execution on any given fight.

Completely regemming wouldn't account for a consistently noticeable difference the same way a new tier of upgrades would, etc.

I'm going to ignore the first part, because plain and simple making comparisons based on the performance of different players isn't going to show anything, there are too many variables. You need to have something constant to compare with. You can compare an individuals performance over time if it doesn't change, but there are still many other variable factors there.

The target dummy for one player is much better, but a 15 minute sample is still fairly small considering things such as variation in cycles, how random effects modify your cycles, cool down usage, etc. You'll get close, and that would probably be sufficient to show if there is a 50 dps difference, but 5 or 10 would need more.


The main problem I have with your analysis is you are drawing conclusions from your observations that completely contradict theory without any controlled basis for the correlations. It seems you are entering into the analysis with pre-formed conclusions, then looking for any observations that may support them, but the support isn't there.

Allow me to run some numbers, if I may.

First, it should be noted that on trash and the like, the relative attentiveness of the players and their decision-making in picking targets is going to dwarf any difference in gear. So unless you have 2 players with identical playstyles paying identical attention to the raid, those differences will overwhelm all but the largest gear differences.

Hence, to get true comparisons between gearing, you need to look at boss fights - and not all boss fights are suitable. For instance, comparisons cannot be made at all on M'uru if the rogues in question are on different doors. And even if they're on the same door, differences in positioning affecting the value of blade flurry, differences in getting gravity balled or not, and so on will affect results. On Felmyst if one gets encapsulated and the other doesn't, that will affect things. And so on.

Hence, only on a truly sustained fight like Brutallus can you consistantly see differences more due to gear than other factors.

So, the question becomes, in the best-case scenario, how long does it take you to find a difference in the damage output of two rogues on Brutallus? Well, the first thing we need to know is how large the difference in question is. Due to socket color restrictions and the like, I would say that switching from full Agi to full Hit is going to give a swing of, say, 60 stat points. So, what's the EP difference between agility and hit? Well, it's probably under .5 EP per point, so we're looking at a difference of at most 30 EP. 1 EP at 70 gives around .5 DPS, so we're looking for a damage difference on the order of 15 DPS.

Now, I would say that the standard deviation on damage in a Brutallus attempt is probably close to 50 DPS. For the sake of argument lets say it's only 25, but it's certainly not less than that. Hence, we know the *difference* in DPS between the two rogues with a standard deviation of 25 * 1.4 = 35 DPS.

Now, in order to distinguish with 95% certainty a difference of 15 DPS, we need to get the standard deviation of the difference under 8 DPS. Thus, we need to do (35/8)^2 ~= 19 Brutallus attempts to determine the difference.

So, to determine the difference between socketed *full* agility, and *full* hit, with two rogues with otherwise *identical* gear, of *identical* skill, using the exact same strategy on Brutallus, would require 19 Brutallus parses to tell the difference, even under our optimistic assumptions. But in practice, these are unrealistic assumptions - the data just isn't that good. So we need even more data than that to tell the difference.

It should be becoming apparent by this point that the logistical hurdles of in-game testing are significant and nontrivial to overcome, which is why as a community we generally put more faith in simulations and calculations than we do anecdotal evidence in-game.

I'd just like to point out that there's a very large chance that the numbers derived from such a survey are off by quite a bit, someone with a bit more math skill than me could probably tell you exactly how large the variance would be. But asking 1000 in order to figure out the opinion of 300 000 000 is obviously going to be off by a rather large margin. (Even if you account for the fact that these surveys are usually conducted by asking people of sorts and sizes.)

At any rate, I hope you understand that a WWS or two is not enough to draw conclusive evidence from.
Even if they were done on dummies and with people of equal gear (apart from the gems in question), you would still need hours upon hours of flawless play to discern such minor differences.

If a rogue does 5k dps, and one gem over another gives a theoretical dps increase of say; 1dps.
Then that's a 0.2% difference, that's very minor. Very hard to test.

(Sorry if I'm just re-posting here)
(Edit: Aldriana beat me to it with some hard facts and numbers)

Even observing same player, on same fight with same gear and same group setup I would estimate the difference in DPS is several percent, perhaps even in the area of 5-10% depending on the fight. How then could we evaluate gear differences that are below 1% level? There are just too many random factors affecting DPS numbers and you would need huge samples to reduce randomness contribution well below of what you are actually measuring.

They key point is that, statistically speaking, the size of the pool of voters (which, as a note, is more like 150 million than 300 million) is not really that relevant towards determining the accuracy of a poll. What matters is the sample size. So lets consider, for instance, the standard deviation of a poll with 1000 responses. Lets assume for the moment that the responses are split 50/50; then the variance in the result is 1000 * .5 * .5 = 250, and the standard deviation is the square root of that - or a bit under 16. Thus, with the 95% confidence, the true value lies within 1.96 standard deviations of the mean, or about 30 responses - which is 3%. Note that such polls usually site a margin of error 3 or 4% - which is exactly what we just came up with.

Now, add to the fact that on any individual attack, we have a lot more sources of variance than a simple yes/no poll, so we need significantly more than 1000 attacks even to get within 3%; and 3% (or 75 DPS for a level 70 rogue in Sunwell gear) isn't really enough resolution to tell us much anyway. To get to the half-percent and below type accuracy we needs requires hundreds of thousands of attacks without the parameters of the experiment changing - and that's just hard to do.

I do concede the point about comparing two people -- it is not theory crafting and is not particularly relevant here. I just posted that WWS to show that the biggest difference and what was holding me back was hit. But enough about that.

What you are proposing is faith in math rather than anecdotal evidence. No arguments here.

I don't trust simulations of something with as complicated math as the wow Combat system -- to many ways to mess it up.

I want to stress the target dummy again. With the target dummy we don't need to simulate. You will get an accurate picture and proof of the theory on a dummy, not just anecdotal evidence. If you turn on auto attack and walk away with a data collector going you will get almost everything you need to prove the Theory. In most cases White damage is all you need to compare gear. Whether it is an adjustment Haste, Hit, AP, Agi, Str, or Crit, the change will show in White Damage. Sure bonus crit damage is improved for special attacks, but it still shows on the white damage and you know that the actual DPS difference is larger. This method will solve a lot of side grade comparisons. [Shoulderpads of the Silvermoon Retainer], [Swiftstrike Shoulders], and [Shoulderpads of Dancing Blades] are perfect examples of gear that can be compared in this way.

Set bonus and cool down abilities make it more difficult -- well not more difficult, just not something you can walk away from. These dummies allow you to poison them as well as unleash endless rotations, so it still works. In order to keep your sanity you may have to do this for shorter periods more often, but it can be done.

I do trust the math in the theorycrafting, but with some of these I really want to see it in practice. I brought up the shoulderpads above because this is the best example of my personal theorycrafting, and I went with the Swiftstrike because it gave me something like 10dps (I am not at the computer with the DPS spreadsheet). Now I want to compare these to see if the Theory is correct.

Auto attacking wouldn't be sufficient at all. One of the main reasons that there are different stat weightings now are because of talents that affect our energy use reacting to an improvement in certain stats and not others.

Focused Attacks procs more with higher crit, and gaining more energy doesn't affect our auto attacking whatsoever, it affects how we use our abilities.


There is no point at all to just test auto attack dps, since over time, it should equal your paper doll dps, so you can just switch gear out and see how it affects that number instantly if you're curious.


There are also other stat levels that determine how incremental gains affect dps. Expertise for example has good benefit to prevent dodges on finishers that affect our cycles, not just on auto attacks. The same thing goes with hit... it has vastly reduced benefit once you pass the spell hit cap, as it is only improving white dmg from that point on, while there are other stats that improve white dmg as well as specials and talent procs.

Autoattacking really does not provide a viable assessment of the DPS output of a given item, particularly if one is lacking raid buffs. The interdependencies between white damage and yellow damage in terms of procs and the like are too sophisticated to by simply ignored when comparing gear. Is basing a comparison on an autoattack parse better than nothing? Sure. But it's not really very good, either. Hence, the goal of this community is to perform tests of that sort to determine the underlying mechanics governing the behavior, and then simulate or model those mechanics on a scale that cannot be achieved in-game.

Lets be clear: our mathematical models aren't perfect. Our simulations aren't perfect. There are probably aspects we're missing in them, and even with perfect knowledge and implementation they'd still be estimates. But, despite any problems they may have, they are still a better assessment of real game mechanics than any reasonable amount of in-game testing. There are simply too many variables for in-game testing to be a viable technique, so, flawed though it may be, the mathematical models and simulations are the best tools we have access to at this time.

When figuring this out we don't count individual attacks. My 1.4 weapon has twice the number of attacks as your 2.8 weapon. The goal is DPS. This means an even longer sample size. The answer is we need to get a sufficient sample size measured in seconds not attacks. The number of seconds is quite likely as much as 10,000 or more. 3 hours is 10,800 seconds, assuming only half a raid is spend in combat, you can reach a sample of 20,000 seconds in one week of raiding.

I am really intrigued, and I want my reference materials back. I'd like to see at what point you get a statistically significant sample -- and if it is feasible -- after all 1,000,000 seconds is over 11 days.

It also seems that there is a major contradiction here: If we can't determine if we are actually doing more damage then why are we figuring out how do more damage in the first place?

All of us know if one piece is clearly better than the other. What we want we know what separates items that are at the same level.

Exactly what question would this in-game test of yours answer?
What do you hope to achieve with it?
I'm curious.
Because testing items on a case by case basis seems like, well, a full-time job.

Good Question. For me it is seeing if the Hypothesis (Theory) is valid in Real Life.

Your target dummy example also ignores a more practical fact that many (most? all?) of us are constrained by: we don't all get to have every piece of loot that might be useful to us, or that we'd want to test. I may wonder which 2 daggers out of a choice of 5 are best for me, but I likely don't have the DKP to buy them all, and even if I did, who wants to get 5 mongoose enchants when only 2 of them will end up being used?

For that reason, we turn to calculations and simulations to guide us so that we can make gearing decisions ahead of time, without having to obtain the item first. Testing every piece of gear on the dummy is just not practical, as Mækk said.

Wow Web Stats
Other rogue in our group missed less than I did, and I pulled quite a bit more dmg than him.

Making assumptions on 2 equations that arent equal is a bit folly?

I changed things a bit last night, trying dual wound, instead of the deadly/wound combo I tried last time. I tried to work in killing spree and had an aggro issue on the pull, and had to pause dmg. In the end, I had close to 4100 dps, which I think can be due to the poison change and slowing threat for a short period, but it was not that far off the 4200 I had last week. Our fight took 30 seconds longer, that alone lowers the value of heroism, which would drop dps.
Wow Web Stats

I'm 99% certain that Setup is currently bugged. The "chance to gain a Combo Point upon resisting a spell" component appears to be disabled; I'd say due to the fact that Spells now Miss, rather than Resist. Can anyone confirm my 1% suspicion? I've only tested it against one mob.

I've been browsing some of the instance drops that mmochampion has access too, and I've noticed a frightening trend.

Most gear is becoming incredibly STA heavy, at the cost of any AGI at all.

I was wondering if there are any rogues in beta that can confirm or deny this in a broader sense? I don't see how STA helps a raiding rogue. Or half the DPS classes in general.