(Hi, from Ask Mr. Robot again). We are actually working on an in-game reforger to our addon We're trying to get it to import all of the changes you need from our site. We hope to have that out in a few weeks, so keep your eye out. If you want to be notified as soon as it launches, send me a PM and I'll keep you posted.
Ask Mr. Robot Human Minion
By the time it is released (if it ever does get released), 6.0 (the new expansion pre patch) will be around the corner, which will completely change combat mechanics again, rendering his new spreadsheet useless.
As much as I would have loved to have had the use of his completed spreadsheet for Tier 14 and Tier 15 progress, he let us warriors down again. His compendium is completely abandoned also, I would not recommend anyone buying it now, as all you'll find are some very outdated macros from 5.0
Darkfriend: Are there any specific things you feel that SimulationCraft needs to fix to better model Warrior DPS? You mentioned something about it not handling the action priorities well. Can you give me some examples and reasoning?
Hmmm...The action list can be changed to take into account certain "big" gear set changes (well, even small ones but becomes pretty complex to make a "default" priority list to cover everyone). I haven't checked out the Warrior action list but the Shadow Priest one for example for Tier15H changes the default action list if the Lei Shen trinket is present. If there is a trinket for Warriors that has such a big impact then definitely possible to change the action list to detect its presence or lack thereof and just the rotation.
The fight length can be adjusted easily. max_time=<length in seconds> if using the command line version or if you pass it as a text option in the GUI. The GUI also lets you pick some other lengths via a drop-down.
That fight length by the way is an average length. What it does in practise is try to centre the thousands of iterations on that given length and then spreads out the iterations so that half of them are less than that amount and half are more, in a nice smooth coverage. So if say you specify a fight length of 500 seconds, you will have iterations going from 450seconds to 550 seconds, and if you had say 10000 iterations then it would adjust the fight length by 1/10th of a second every iteration. And at the end it will average the DPS from all of those iterations. As such it will pick up on slight differences in fight length pretty well.
Trinket luck is again taken care of by the number of iterations. SimulationCraft is a monte-carlo type simulator. That means in each iteration is runs through the fight as if it were running with the in-game mechanics, and then at the end of those thousands of runs through it gives the averages (and standard deviations and plots etc). So you might have lucky trinket procs on some iterations which bump up the numbers, but the law of averages says that over thousands of runs they'll tend to be balanced by those runs where you didn't have lucky procs. We still keep a track of the minimum and maximum DPS iteration, and for that matter show a curve of all of the iterations so you can see the DPS spread. The final number is just the average of them all.
Would you like some more detail on any of these things or have any more specific things you'd like to discuss. If there are some direct changes to the action list that would increase DPS (or be a little more generic for those with different gear) please let me know!
TO address the simple issue of fight length, I didn't mean in Sim-C itself, I meant in an actual fight itself which will change (generally getting shorter) each week. Depending on the fight where certain damage buff portions occur can drastically raise or lower any classes DPS.
Overall, my main point is that one shouldn't run 2 separate classes through Sim-C and say "look monks sim x number of DPS higher than us warriors are shit FML." Although Sim-C can be used as a tool to gauge roughly a classes DPS, it isn't accurate enough to account for everything for a realistic DPS number. It's strengths are in its stat weights and for comparison between items.
Other things that sim-c can't really model are movement (hurts some classes/specs far more than others) unless the movement is very very close to how that individual fight operates. Although you can use the difference between a patchwerk fight and a low movement fight to compare how much movement impacts any spec, again it's the strength is in the difference between the two.
Player skill is very hard to model as well, and again impacts different classes in different ways.
Now on to more specific things. I haven't messed around with custom priority lists enough to know exactly how to fix these things, but I'll just list the things that seem to be slightly off below.
1.) Mogu potion use is shows as being (mogu_power_potion,if=(target.health.pct<20&buff.recklessness.up)|target.time_to_die<=25). It should be used before reck, since you'll be using CDs->DR into execute phase, and this will keep DR from benefiting from either skull banner or potion, or force you to clip reck with DR in execute phase which you generally want to avoid (although there are exceptions).
2.) Cool downs aren't being saved for execute phase. This is a big one.
3.) Berserker rage is being misused *I think*. Looking at the sample sequence I can tell that it for sure is being used too early right on the pull, for example.
4.) Minor things with rage values. Anytime you limit to only using HS/WS when certain rage values are met can lead to problems, because it's going to vary in any fight. This is a fairly minor one though, and I don't really know of any good way to fix it.
5.) There kinda should be some specific sequence for feather procs during the pull and during execute phase. A proc at the right time will change around the rotation quite a bit to maximize AP based abilities during high feather stacks.
6.) Since Shout is literally the last thing on the prio list, it really doesn't need to have the under 70 rage thing. Under 90 is probably better, since the next BT will bring you to 110 assuming a crit.
The biggest things are feather and execute hase.
Thanks for the feedback!
The movement stuff while by default not modelling any particular fight still does try to model periods of actually having to move although I admit it's by no means perfect. Improving the "raid events" as we call them is one of the things I'm considering tackling myself. I no longer actually play WoW personally (plenty of the other SimC devs still do of course) so I'm more likely to tackle more "meta" things like that.
The same can be said for the player skill thing. In fact that is definitely one of the areas that I feel we don't handle too well. We do have some sort of player skill mechanism though by default it's turned off and it plays as well as it could...after a fashion. It still can't react to procs that it doesn't "know" about yet for example. So there is still both some sort of a lag + reaction time delay with some procs, but that's how it should be after all.
Player skill modelling is probably the second main area I'd like to tackle at some point. (The third is an audit of our latency modelling. I think our current approach is a fine approximation but I'm a stickler for "correctness" and I'd like to more clearly separate the sim internally into having the concepts of a "client" and a "server" when it comes to latency modelling. Mostly just to simplify the number of latency variables we have.)
I would probably have to disagree with you a little on your claim that SimC isn't accurate enough to model a class's DPS. But of course with the caveat that a) In any one run it's only modelling it in a certain circumstance (i.e. a Patchwerk fight if that's what it chosen) of a certain length. b) We're not perfect and it's certainly possibly that some class modules might have the occasional bug, especially in the more complex abilities, and c) The default action priority lists and BiS gear choices etc. are only as good as our own efforts combined with the feedback we get. While we do have some "class experts" helping directly on some modules, action lists and gear choices, SimC development is mostly about getting the underlying code correct so that people can play around various things like action lists and gear.
Thanks for your action list check. If I get some time I'll try comparing it with the default warrior action lists and seeing if I can improve it. If it's not looking for feather procs then yes, that would definitely be a likely good place to improve it. All of those points raised by the way look like they're things that can be adjusted purely in the textual action list without requiring any change to the simulator itself (other than of course changing the default action list it generates once we have an improved version ready to go).