I am trying very hard to understand this.

So does ABSpam04MBAM mean

Spam AB, and cast AB again followed by a 4 stack MBAM?

It means spam AB, and cast an MBAM if it is available and your Arcane Blast stack is either 0 or 4.

Thank you so much!

I know in previous posts it was discussed that 4T8 is going to be even worth less dps for an arcane mage in 3.2.2 than before...but with regards to this cycle, is it as bad as everyone thought?

In other words, if you are at a 0 stack AB, and you cast MBAM if 4T8 causes you to keep your MBAM proc, you would cast MBAM again correct?

That is as correct now as it has always been, yes.

Unless I'm mistaken, it looks as though the delayed MBAM procs have been hotfixed to the way it was before (i.e. proccing immediately after cast).

I think I am either understanding your wrong, or I think you might be off on the 92% thing. Yes, the flat chance of getting a proc after casting AB 5 times is 1 - 0.6^5 = 92.22%, but you cannot treat it as being independent if we know some of those 5 AB casts have already failed to proc it.

What I mean by that is, if you cast AB 10 times, and it never procced, and then cast AB another 5 times, in those 5 times, your chance to proc at least once is 92.22%, just as expected (the fact that the previous 10 did not proc it makes no difference at all). However, if you cast AB 4 times, it did not proc, and then cast AB once more, you cannot say the chance of getting a proc in those 5 casts is still 92.22%. Basically, it would be incorrect to say that in 92% of the cases, if you failed to proc AB by the 4 stack, it'll proc by the 5th one.

We can look at this in terms of 4 AB casts because those 4 AB's are always cast either way, completely independent of procs. After that, we only keep casting if we don't proc, which means you have to look at each cast independently. You can say the chance to proc in 2 casts is 64%, but that number has no meaning once we're already at 4 stacks, because of the possible outcomes: NN, PN, NP, PP (P = Proc, N = No proc), we only care about these two: NN and NP. In the case of PN or PP, we wouldn't cast the second AB because it has already procced.

Have one question about trinkets:
Its clearly showing on Rawr ( last update ) that Embrace of the Spider is actually on 8-9th spot for "best trinkets" for Arcane specc. But when i watching ABSpam04MBAM on Rawr ( as your main cycle ) it showing that Embrace of the Spider is far more better then any trinket except Reign of the Unliving ( both 245/258 iLvl )
So question is , is this Actually true or is just showing my current gear/stats ( by watching ABSpam04MBAM )?

/cheers

So, the most mana conservative rotation would be:
000200102220: 6725.712 dps, 17.99562 mps = AB4ABar1234MBAM

Which is telling me to spam AB to 4 stacks cast ABar if no MB procs and cast AM if MB procs on AB stacks 1, 2, 3 or 4?

But the highest DPS rotation with reasonable MPS is:
000000020020: 7199.522 dps, 153.0721 mps = ABSpam04MBAM

Ok, so just so I get my head around this, without T8 I should cast AB to 4+ until MB then cast AM.


But really this one seems to be the best as far as DPS versus MPS:
000200002220: 6975.286 dps, 65.87646 mps = ABSpam1234MBAM


Which is basically telling me to cast AM as soon as MB procs as long as I have an AB stack?

Am I getting that right?


And really what I am getting out of this is that it is going to be up to me and my mana as to when I cast what, but that I should never be casting AM without the MB proc. Is that also the case?

Thanks.

Well with the fix to MB we're back to where we used to be. Here's the new old cycles in case you forgot (with the same setup as for previous numbers). Note that these are optimal regardless of whether you have ABar.

Cycle Code Legend: 0 = AB, 1 = ABar, 2 = AM

0: AB0,ABar0,MB0-
1: AB0,ABar1.832449,MB0-
2: AB0,ABar1.832449,MB13.83245+
3: AB1,ABar0,MB0.2843535+
4: AB2,ABar0,MB0-
5: AB3,ABar0,MB0-
6: AB4,ABar0,MB0-
7: AB4,ABar0,MB13.12741+
8: AB2,ABar0,MB0.04183626+
9: AB3,ABar0,MB13.12741+
10: AB1,ABar0,MB0-

00000000000: 7262.698 dps, 677.8624 mps = ABSpam
00000002000: 7154.938 dps, 179.5613 mps = ABSpam4MBAM
00000002200: 7064.082 dps, 143.3241 mps = ABSpam24MBAM
00000002220: 7011.305 dps, 125.1452 mps = ABSpam234MBAM
00000022220: 6759.125 dps, 59.91507 mps = AB4AM234MBAM
00000201220: 6488.157 dps, 25.91806 mps = AB3AM23MBAM
00002001200: 6068.341 dps, -10.45493 mps = AB2AM

As discussed in the Rawr.mage thread, the new armory parsing system has broken some trinkets. Reign of the Dead/Unliving is fixed in build 36978 of the SVN but not yet released. Unless you can compile and build Rawr yourself you'll have to wait for 2.2.20

Not quite. Since people wearing 4T8 will still be able to MBAM twice in a row, I believe you forgot that in your arcane cycles calculation. Since you've removed ABSpam04MBAM cycle from the latest build of Rawr I can't prove this with numbers, but looking at the values for MBAM, it says for me 7597dps (-179mps) while AMSpam4MBAM at 7452 dps (204mps). So if I were wearing 4T8 still I'm assuming I'd still want to use the extra MBAM proc immediately.

So don't even use 2.2.19?

I think this is both right (in its rejection of bad probabilistic thinking) and wrong (in its focus on averages as inputs to the decision tree). First to paraphrase Morthis more informally/intuitively and with less statistical jargon: when you start chaining AB, there are a certain number of permutations (think of it as a coin flip where heads has a 40% chance), some of which result in a proc in the first 4 applications, some of which result in mana dumps of 5, 6, 7, 8 or more. When you finish casting that fourth and MB has not yet procced, you can informally think about it this way: you now know that the chances are much higher that you're in one of the really bad permutations (7, 8 or more), because you know you're not in any of the good ones. The chance of it proccing in the next four ABs is exactly the same as the chance was in the first four: 1-0.6^n.

If you allow yourself a new decision point after that fourth cast--and I don't see why you wouldn't--what do your options look like then? There are three basic paths you can take: ABarr or AM now, 1) take the dps hit and the relatively small mana hit and start over; 2) cast a few more times then abort if it hasn't procced; 3) commit yourself and spam. Your risk-reward on 1 is obvious. If 2 works out for you it may have been worth it (as Casstor says your MPS average has gone up), if 2 doesn't work you've dumped a few AB3s and now you don't get anything AND you have to pay to get out of it. Option 3 will, over the long haul, result in the DPM figures Kavan has posted. If it's good, 3 will look like the good 2. If it's bad, you will be out of mana and sitting there with your thumb in an unpleasant place, waiting for your evocation cooldown.

I think that the error that's being made by people who are pushing AB spam (and who are undoubtedly statistically literate) is essentially that they're not accounting for variance or not weighing variance with cost. Specifically, how tolerable, exactly, is running out of mana with no way to get it back? Personally, I would say catastrophic. I just don't want it to happen, and, not to be condescending, 90%+ doesn't mean "round up to 100%". Running many, many statistical trials is what DPS is all about. That means it WILL happen. An arcane mage who plays this way, to me, is somewhat analogous to the rogue who puts out really good numbers then goes and stands in the fire. What is the cost of being unreliable to your raid?

Basically, averages be damned if you can't tolerate the variance. When you're at 40% mana left and you have to decide whether to keep spamming, you can't, I would argue, make the decision as Casstor says based on your preferred average MPS. You have to account for the very real possibility that when you finish that rotation you will be OOM. If everything goes to plan, you pay for a 4 stack of AB. If not, you have no mana. That is the full range of possibilities, and the catastrophic case--even in a single trial--is unlikely but in no way statistically insignificant. That means the decision isn't based on averages--it's based on worst case scenarios.

(Not statistically insignificant, you say? 8 stacks without proc = 1.68%. Take 25 rotations as an example fight, chance of it happening once is 42%. 15 rotations, 25%. Stack multiple fights per raid, and you get the idea...)

Sorry for being long-winded, but this is a cool discussion as we try to figure out what the best practices are. Kavan, I'm really interested in AB4ABarr4MB, if you'll humor those of us who aren't sold on spamming.

Well remember when calculating probabilities there's many different ways of looking at the same event. What I calculated was a binomial distribution with a 40% "success" rate, specifically the probability of getting at least 1 "success" in 5 casts.

So what I was saying was that if you're at 4 stacks and want to know whether or not to cast again, you can think that 92% of the time you will see at least 1 proc in 5 casts. That doesn't mean there's a 92% chance to proc on the 5th cast, or that the previous 4 casts have anything to do with your next cast's chance to proc. It just means that the chances of seeing at least one proc in 5 casts is 92%. How a person interprets that and applies it is up to them

I searched the whole thread and no results came up so I'm wondering does the proc from the legendary mace count towards activating IA? Because my guild runs with 3 full-time raiders who have it so I'm almost always guaranteed to have at least 1 of the shields on me which would definitely make IA worth taking regardless of the fight.

Yes.

Yes, you want to use that extra MBAM proc immediately or else you run an incredibly high risk of losing the proc. I theorycrafted this pages ago in this thread, and with the changes to AB and MBAM it has only become an even greater loss.

It isn't catastrophic. A famous comparison from BC theorycraft threads was the comparing of scorching to save mana to running out of mana and wanding -- turned out running out of mana and wanding until you could cast the fireball was better.

The key is looking at the dps/mps ratio of your tradeoff.

Using kavan's numbers, it is better to drop to the AB4AM234MBAM rotation than to spend any time in evocation.

Presuming the ABspam4MBAM rotation as a baseline that cannot be maintained with poor MB luck, we start with 7155 dps expected.

Falling back to AB4AM234MBAM loses 395 dps but gains 120 mps for a ratio of 3.3.
Evocating loses 7155 dps but gains around 2050 mps (kavan's numbers appear to be assuming around 250 mps between spirit, gems and raid buffs, and I assumed 24kmana when raid buffed) for a ratio of 3.49.

So, in fact, for the spam/nospam decision, whether your evocation is up is completely irrrelevant. If you have to use evocation at all, it's because you are in a very long fight where you can't even sustain a 60 mps rotation, or you missed switching to a lower mps rotation at the proper time. This is not a huge error, so it pays to be somewhat aggressive, resulting in sometimes using evocation. But it also means that it should be highly unusual to be in a fight without painful mana mechanics or a mispull where evocation on cooldown and you are anywhere near out of mana. You want to ride the line of using evocation *at all* on a fight less than 6 minutes, not using it on cooldown. Even if you do run out, wanding to regenerate mana is not a catastrophe, it's just bad mana management. You only have to do it long enough to be able to cast the 75% dps mana gaining rotation of AB2AM, which you then maintain until your evocation is back up. Comparing running out of mana to dying in a fire is way over the top. At worst, running out should mean some time at 3/4 dps, and rarely more than a few seconds of actual wanding.

Using Kavan's numbers, each step down the dps tree works like this in terms of dps/mps

Pure ABspam v. spam4MBAM: .214 (about as low as you can get, i.e. use pure spam only when mana is effectively infinite)

spam4ABAM v. spam24MBAM: 2.52

spam24MBAM v. spam234MBAM: 2.94

spam234MBAM v. Evocation: 3.29

spam234MBAM v. AB4AM234MBAM: 4.2 (this is worse than Evo, so evo *should* be taken when necessary in cases where mb is proccing and you are still low on mana. If MB is not proccing at all, then the 2 middle steps don't exist as options, either you evocate or drop directly to AB4AM234MBAM -- tradeoff v. evocate for that is as done above).

AB4AM234MBAM v. AB3AM23MBAM: 7.97 (Bad, but much better than wanding)

AB3AM23MBAM v. AB2AM: 11.667 Worse yet, but still a lot better than wanding.

AB2AM v. Wanding: 16.9

Gearing for mp5 vs. spellpower: 17.2

Didn't feel like doing the calc for spirit/int regen, gearing for spirit or int is probably better than wanding (due to crit), but since AB2AM actually regenerates mana in normal raid situations, there's no reason to do either except on mana drain fights or with other gimmicky problems.

I'm much more interested in AB4AM, not as a replacement for the spam based rotations, but as a fallback when MB is killing us. Should be higher dps and slightly more mps than AB4AM234MBAM the question is whether it's a better dps/mps tradeoff vs. the various spam cycles.

You seem to have forgotten haste here. Evocation should only take around 6 seconds to complete in typical arcane gear.

To me the answer to this is rather obvious. Once my mana gets below around 35%, I am guaranteed to get maximal efficient use out of evocation (unless I get frost warding procs or mana tide or other things that I can account for depending on fight and group makeup). This accounts for the 60% I get back from evocation, plus a margin of error and accounting for spirit regen during the cast. Until that point, I fish. Once I get below that point, especially if there is around a minute or more left until the evocation cooldown, it makes sense to 1) give up the fishing strategy and 2) even sometimes start using procs earlier, at 2 or 3 ABs if I have it by then, depending on how low my mana is dropping.

Effectively, at that point it means that whatever mana I'm not spending now, I'm guaranteed to have available to spend later. It means that even if I reach that 35% point and am forced to adjust to a strategy that lowers my dps, that loss is mitigated not only by the decrease in risk of going oom, but also by the higher dps I'll be able to output later with the extra mana I have available due to the strategy switch.

There's a concept called the "knee of the curve" in cost/benefit analysis. Basically, that's the point at which you see a marked drop-off in returns for additional investment. To me, that knee is clearly at 35% mana, adjusted for whatever other recovery options are open to you, and so that is when you should shift to a more efficient rotation if evocation is still a ways off.

Here's the optimal cycles for those still using 4T8 (numbers not comparable to the ones before due to different setup):

0: AB0,ABar0,MB0-
1: AB0,ABar1.789023,MB0-
2: AB0,ABar1.789023,MB13.78902+
3: AB0,ABar0,MB11.65406+
4: AB1,ABar0,MB1.169254+
5: AB2,ABar0,MB0-
6: AB3,ABar0,MB0-
7: AB4,ABar0,MB0-
8: AB2,ABar0,MB0.4352674+
9: AB3,ABar0,MB13.05504+
10: AB4,ABar0,MB1.005268+
11: AB1,ABar0,MB0-

000000000000: 6963.628 dps, 651.9622 mps = ABSpam
000200000020: 6900.03 dps, 150.955 mps = ABSpam04MBAM
000200002020: 6824.473 dps, 115.2184 mps = ABSpam024MBAM
000200002220: 6779.815 dps, 97.92279 mps = ABSpam0234MBAM
000200022220: 6552.298 dps, 40.43221 mps = AB4AM0234MBAM
000200202210: 6307.53 dps, 10.48285 mps = AB3AM023MBAM
000020202210: 6267.814 dps, 6.185516 mps = AB3AM123MBAM
000021002010: 5981.563 dps, -18.89301 mps = AB2ABar12MBAM
000022002010: 5908.036 dps, -23.6222 mps = AB2AM12MBAM

I think what's being missed in this discussion is a consideration of maximum mana usage. With the fine-grained control we have of our overall consumption, we need not be constrained to a fixed rotation--that is, I don't think these rotations that advocate Arcane Barrage after N Arcane Blasts without a Missile Barrage make sense. Rather, I think there is a level of mana below which you can no longer afford to fish to make it to the next Evocation. In this, I think runciter is on the right track, but the idea can, I imagine, be improved with more rigor.

What I conceptualize, though, is that for 2 minutes of chaincasting, there is some maximum number of 4-stack Arcane Blasts you can afford to use in a fishing cycle, and once you exceed that point, you must resort to a more conservative cycle. Which more conservative cycle to use is a question of optimization and could be considered a function of how long one chooses to fish, but even then, with Missile Barrage random, one has to further determine how that selection should work--do we choose the conservative cycle based on worst-case consumption? Do we resort to this cycle early to have more leeway later? These are open questions and not ones I'm entirely prepared to answer, but I think this is where the direction of rotation theory needs to go to help Arcane mages determine what to do when their "luck" is worse than expectation.

Well we already know given a starting state which rotations are optimal given starting mana and available mana tools down the road. Which cycle pair is optimal is ultimately dependent on total mana available (ignoring effects of AP on optimal cycles and similar fluctuations). This means that as you're moving along in the fight you can potentially redo this whole calculation all the time taking into account your actual mana situation. You basically compute optimum cycle pair given your current situation. It's not hard to see that what will happen is that if things go close to average you won't have to adjust at all and otherwise you might have to change your cycle pair up or down the list depending on whether you're above or below average on mana.