The step range is 1s, as it would be a bit more difficult to account for gcd in this way of doing it. (I would imagine).

Maybe the steps could be changed to e.g. 0.1 and a new column with wether a gcd is active or not can be added?

Just a guess into the blue, because i am on my way into the weekend no hard math, but could the culprit be that you generate more combo points than you can spend with your 10energy/sec?

I respecc'd to HaT last night. Ran a couple of 5 mans with more or less same dps as I had achieved in Combat Sword form, then ran the PvP raid as only rogue in a Pug. Achieved 3200dps according to recount. (Sorry when Santa brings my new box I will run all the stat gatherers, I promise)
Just wanted to check thoughts on ideal weapon/poison set up for this build.

Actually, the lack of change from 2.5 to 4.9 makes sense. Those combo points are wasted (since all abilities require 5 cp to use). If you generate 2.5 CP per second, and you're waiting for 5CP to fire off an ability, then unless you have exactly 2.5 CP when you generate the next 2.5CP, something is going to be wasted. You could fire off a Hemo to increase dps during that dead GCD, which the simulation does when it doesn't drop your energy too low to use an Evisc on the next GCD, SnD is active, Rupture is active, and neither needs to be refreshed in the next GCD or two.

There is something to be said for a rotation that does not require 5s/5r/5e. I will be looking into that next. Also, this model doesn't allow for 'player ingenuity', if you will. Ability usage is relatively fixed (the rules governing when the simulation is allowed to use an ability). A player may be able to sqeak out a little bit more dps with creative ability usage, but for the most part, the results are a solid baseline (imo).

Changing the granularity won't make a difference due to the GCD. I may be mistaken, but all the used abilities trigger a GCD of 1 second.

You are dead on correct. You are Combo Point limited until 2.5, which is why you see scaling from 0.0 to 2.5. From 2.6 to 4.9, you are energy limited, even with full Relentless Strikes. At 5.0, your abilities effectively cost 10 energy, which you regain every GCD. In a sense, all CP and Energy limitations are removed once you hit 5CP/s.

-----

I understand this isn't a perfect model, and I have plans to update it further for more accuracy and flexibility. This model was just a quick and dirty I threw together yesterday. However, it does point out two important things: what is limiting our HaT based DPS in each range of CP generation, and why the HaT bug causes the problems it does.

Thanks for your feedback. Please continue to make suggestions for improvements.

The modelling is not optimal between 2.5 CP/s and 5, because of one of these restrictions :
- use finisher only if 5 combo points (2-4 CP evi someone ?)
- prioritize Hemo over not using GCD (waiting a fraction off GCD for 5 combo points may be better than using Hemo)

Correct on both accounts. The requirements for 5CP ability usage is an severe limitation, but necessary for this first pass. This will be looked at in my revised model. The Hemo CP Thresholds and Hemo Energy Thresholds actually allow for some control over the 2nd point, but there isn't true forward looking code to optimize it correctly. I will work this into my next model as well.

Thanks. :-)

You are correct. The finer granularity would make a difference ain the 2.5-4.9 range. I will attempt to bring it down to 0.1, since that is the energy regeneration interval.

I didn't actually assume Expertise and Hit Caps for white, only yellow damage. I actually pulled my white and poison damage estimates (and miss rates) from wws reports (Ippon in Naxx).

Yeah, the spreadsheet needs more customizability. Anything specific you'd want?

http://wowwebstats.com/3yfewofagm6aq

WWS log from our 25-man Naxx this week. Evizerage is the HAT rogue, Corixie the Combat rogue. Gear is close to approximate, but we have both gotten 4-5 upgrades since that raid. I (Corixie) was using Crimson Steel MH, Paper Cutter OH. Eviz was using Calamity MH, Cutter OH. Biggest difference in gear is my Darkmoon card vs his FotFF. On most of the fights, I was anywhere from 100-400dps behind him, with him taking huge leads on crit-oriented fights like Loatheb.

We did everything we could to make sure we had the same buffs. Same group, ToT eachother every CD with macros, no Hysteria on us or anything. It looks like, even without the multiple HAT rogue functionality, HAT is still a viable raid spec.

Another note, on Sarth tonight we realized that holy priest with CoH will generate 1 CP per crit. I'd post WWS but they got borked by the phasing :P

Onyxia is deep breathing more, too.

It can be definitely concluded that this is a myth.

Your own post proves that there wasn't even a correlation, so let's move on and leave the discussion to the topic of the spec, it's current state of bugginess, etc.

Cheshyr:

I only meant yellow attacks in my post, so we're on the same page at this one.

I would be interesting to see at what number of CP's one should use Eviscerate especially to make best use of your GCD's. You have already done something similar with the Hemo tresholds, and if the GCD is kept track of, this would be a nice feature in my opinion. Depending of your cp generation, it might be better to wait for a 5 point evisc or throw in two 3 pointers, if you have enough energy, even though this would require an AP input by the user. Or something

A non expertise capped method of doing this could be by utilizing the =randbetween function of excel to generate random numbers which could correspond to a finisher missing or hitting. But then you would probably need to change the rows used for calcs from 120 to a hundred times more.

Alot of other features would be helpful (like beeing able to plug in average AP/crit values, enable/disable Rupture, talent points etc.) but then it will begin to differ from the illistrutive version as it is now. And since CP generation is as sporadic as it is, it might be a waste of time to model many of these things

The two main weaknesses in that simulation are how you handled CP generation, and the strategies you used.

CP generation is not constant. Let us assume that each party member has a certain probability of contributing a CP each second, and that this probability does not change from second to second (critting in second 1 is independent of critting in second 2, etc. Not entirely accurate but close enough). Using four of these proportions (one for each group member besides yourself) one can easily construct a distribution to model the number of CP gained per second, and each second use RNG to get a number using the distribution. If you want to go simpler, just assume every party member has the same chance of contributing a CP each second and use a geometric distribution.

Now, on to the case of strategies. You need to define a strategy using parameters (since the strategy you defined is clearly not ideal in some cases). The parameters determine when evis is used, when hemo is used, etc. Then your program can, given any distribution, try out all possible combinations of strategy parameters and determined which one is the "best." The best strategy is likely to change depending on the CP distribution, but if you allow for enough possibilities with your range of parameters, it should be fairly adaptable and accurate.

If all of this is done, the largest source of inaccuracy in the simulation will be that our distribution of combo points does not exactly fit what we encounter in the real world. And then there's the inconvenient problem that we have to consider a different distribution with every group, because it will always be different.

Exactly my thought, since it means a difference of throwing an Evisc once every 2 seconds to once every second. I think it's a good rough estimate, but some part is missing in this segment of the modelling.
- 2-4 pt Eviscerates may still yield 0-2 new combopoints (depending on build)
- CP generation isn't constant, 2.5 CP/s can mean as much as 2 CP/s half the time and 3 CP/s the rest of the time, as it could mean 5 CP/s during 25% of the time and 1.7 CP/s during 75% of the time. Such spiky behaviour will be demanding of a rogue to use optimally, but will allow some adjustment of rotations still. I tend to use these dry-spell moments myself to refresh SnD's for example, to fully use the CP generation during better times for Eviscerates (Eviscerate scales up per CP used, while SnD scales down per CP used). But the point is, there is a difference in style required between 2 - 3 CP/s and 1.7*3 - 5 CP/s.
- At 4.5 CP/s one can Eviscerate almost every cooldown, it basically means Eviscerating once every 1.1 second instead of every second. Similarly, at 4 CP one can Eviscerate every 1.2 seconds. So in other words, Eviscerate damage will be lower by 10 and 20% respectively, not by ~40%. There is definitely some cross-overpoint where it starts becoming more interesting to sit out waiting for CP's than it is to use less CP's on finishers, or generate your own CP's.
- Because of this, it's probably safe to assume that there is a very linear increase in the last part of the scale untill it caps out at 5 CP/s, probably starting between 3 CP/s and 3.5 CP/s. Somewhere there, this line of "Sitting out and waiting for max CP's" will start to outperform rotations that (even occasionally) involve own CP generation.
-Sitting just below the 5 CP/s while exclusively using Eviscerates also means you're constantly capping on energy. Effectively, you'll often have enough energy to weave in your own CP generation anyway should the crit rate in your party drop for a couple of seconds.

1. It is 1 CP per second per party member, Ghostcrawler explained as much on the US forums. Higher CP generation per partymember per second has been seen, but this is probably due to lag and some bugs.
2. In the case of such a slow cycle, one would use one Snd, two eviscerates and one rupture. While you do want to maximize Rupture uptime, you should never clip ruptures.
3. Even 10 men Naxx can be crit heavy. Pally and priest healers crit, hunters crit, mages crit. Getting the ideal group and ideal set of buffs/debuffs just becomes more attainable in a 25 men raid, but it's still possible in 10-men
4. Preferably 5 pointers only, especially on the Eviscerates
5. If energy regen permits, yes. You don't ever want to use more energy than what you have coming in, since energy regen is rather low. Finishers cost only 0-10 energy, but CP builders all cost at least 30 energy (talented), so use them very sparingly.
6. No, preferrably even less. The answer to your first question should explain why and how.
7. Use 2 fast daggers under good circumstances, for poison procs, 1.3 speed in both hands would be very ideal. You shouldn't need CP builders much anyway. If CP generation is very low, using a slow mainhander becomes more useful, for your own CP builders, like Hemo.
8. WP/WP
9. In Mainhand too

Since this is indeed the case, I for one would find it helpful to see which rule set would be optimal to apply for 1 CP/s, 2, etc.. Because if we know the optimal way to handle a specific CP generation, then we would be able to adapt to the varying number of CP's. So I do believe that the spreadsheet made could yield some new information on how to play the HaT spec more optimal. I we want EP weighings, then of course a tool like Roguecalc or something would be optimal.

Ok, I've got a lot of information to process thanks to all the feedback. I forgot the level of simulation accuracy demanded by EJ members. ;-)

The initial purpose of this spreadsheet was to see why the HaT bug was causing such huge dps gains, and provide a little bit of insight into HaT scaling. I had never intended it to be used for individuals plugging in their own values and evaluating rotations based on party make-up within a raid. However, that appears to be what is in demand, so I'll attempt to turn this informational hack into a full working simulation. It may take a couple days, however.

I'll put together the list of features I intend to implement and which ones I'm going to 'wish list' for now, and post them in a bit. Feel free to leave more ideas. I don't intend to compete with RogueCalc or any of the other dps spreadsheets. This is going to be specific to HaT builds; please keep that in mind. :-)

Thanks again.

Edit:
I have concerns about adding 'randomness' to the simulation. If the random number generator is truly random,then average values should suffice. If we want to model how the simulation reacts to variance, then perhaps an adjustable deviation from the average (gaussian, as was mentioned earlier) would suffice for that specific simulation, but I think it's be more useful as a post-sim test. Simulate to get your baseline, then monte-carlo the baseline. Is that what you were suggesting, and I just misread it?

I had heard WP/DP. Did this change recently?

I see what you mean. However, does that information do us any good? We don't have any control over that distribution in a meaningful way. The most we could do is to coordinate high-crit chance abilities across a party to create predictable bursts of CP for the rogue, but that seems extremely difficult to me. I'm trying to generate information we can act upon, and I'm too short-sighted to see how distribution could be used.

As far as the only "randomness" that would be taken into account in the first place was finisher miss/dodge chance, and Ruthlessness procs, then I guess an average dps could be found within a reasonable variation with using random numbers as I stated, however I have no idea if this i actually the case, with excel's/oo's limits. However if this is not the case a gaussian distribution could probably be used in some way.

However I still think that the most important thing would be to add AP, crit and debuffed boss AC, to determine a basic rule set to follow while playing HaT. At his moment at least.

The general consensus seems to be that keeping SnD up is always worth it, however Wow Web Stats shows us 3 rogues in top2, only one of them (the 2nd one) uses SnD and even he doesn't have full uptime on it (4 snd:s in 2:01 fight), yet they're doing mindboggling dps.

The only thing they seem to do is one garrote / ambush to start off, followed by normal swings and evis, with a couple
longer cooldowns popped in the mix.

the first one seems to be better geared since average hit damage is noticably higher, 2nd one has higher crit and uses SnD, 3rd one even lower crits and no snd but still missing only 500ish dps.

Also in Wow Web Stats
rogues are not using snd constantly, as well as some more skills and different poisons seen.

Anything to go forward in those?

Hi all -

I tried HaT in Naxx this week, as many of you have it seems, but there was one thing that I thought would significantly boost my DPS.......a macro.

I found myself sometimes sitting at 3 combo points but still not energy capped, and waiting for that next combo point to come in. Then all of a sudden 5 combo points would come in at once, effectively wasting 3 of them.

Is there any macro modifer that will do the following word explanation? "If I have 4 or more combo points, eviscerate, otherwise do nothing".

I know this would severely dumb-down any sort of "skill" involved in the spec, but it would also increase DPS substantially.

Thanks.

-Jared

Not possible

The goal is to predict DPS, and to do that we need to consider combo point generation. Even if we cannot control combo point generation, we still need to predict it in order to obtain an accurate model for DPS. Just as our actual DPS varied in a non-linear fashion across varying levels of CP generation, chances are that different CP generations will yield different EP weighings... not to mention a different "optimal" strategy. If your CP generation varies only slightly, you can get extra hemos in with little risk of wasting combo points. If your CP generation varies wildly (even if the mean amount is the same) you are more likely to end up with waste when you trigger GCD which reduces the relative utility of hemo, meaning it would get used more rarely. As you can see, even with the same mean CP generation, our playing strategy may depend on how much the CP generation varies, so it is very important that we look at the distribution as a whole.