# 5.1 Warrior Simulations – Part 3 – Stat Priorities

In case you missed the previous entries, in part one we detailed the model and analyzed a few basic queues that used either Shield Block or Shield Barrier exclusively.  In part two, we analyzed a number of other queues that tried to combine the two spells.  From that analysis, we narrowed down our choices to three different queues: Shield Block spam (“SB”), Shield Barrier spam (“SBr*”), and a first-come first-serve strategy with a high-rage bleed valve (“F-110″).  For more information on how those queues work, please consult part two of this series.

As the title suggests, in this installment we’re going to take a look at how different stat configurations affect the results of the simulation.  You may recall that I posted some simple TDR stat weights for warriors last September.  I received a lot of feedback on that post, varying from surprise to insistence that those weights must be completely wrong – in fact, as I mentioned in part one, that was part of the impetus for returning to these simulations.  Today we’ll have enough detail to see whether the criticism was warranted.

However, unlike the last round, we have much more sophisticated data and metrics to work with.  In addition to looking at how the stats affect TDR, we can analyze smoothness metrics.  One of the arguments that many people have put forth, myself included, was that hit and expertise were likely to have a very strong effect on damage smoothing much like they do for paladins.  So while they were distant last-place entries for TDR, they may still end up being a very high priority for survivability because they make your intake more predictable.

There’s always the additional argument that hit and expertise are great for DPS, but it’s incredibly difficult to draw a numerical comparison there.  Exactly how much survivability is it worth giving up to boost DPS?  There’s no simple answer.  That’s going to vary based on the needs of your raid team and your level of comfort with your own tanking.  So it’s not something we can analyze with a simulation like this.  However, if the stats give us a significant defensive benefit, stacking them becomes a win/win situation in both departments.

Files:

I’m using two new files in this post, but both are just scripts that call the workhorse function warr_mc.m, which is unchanged from the previous installment.  The scripts are

warr_mc_stats.m – performs the stat analysis for our queues
warr_mc_stats2.m – detailed hit/expertise level analysis

Gear Sets:

First, let’s look at the gear configurations we’ll be using for this simulation.  I’ve chosen to use gear sets that are very similar to the ones I used in the paladin simulations, which approximate an average item level of 496.  The sets each have 11k Strength, 63500 armor, and 17150 rating worth of secondary stats distributed amongst parry, dodge, mastery, hit, and expertise.  This is enough rating to cap hit and expertise and still have 9500 left over for avoidance and mastery.  Just as with the paladin setups, I’ve enforced a minimum of 2000 dodge and parry and 1500 mastery to account for the fact that we rarely get complete control over our stat allocation.  That still leaves 4000 points for us to shift around at our leisure. The stats are summarized on the table below.

|    Set: |  C/Ma |  C/Av | C/Bal | C/Bal-NC | Avoid | Av/Mas | Mas/Av |
|     Str | 11000 | 11000 | 11000 |    11000 | 11000 |  11000 |  11000 |
|   Armor | 63500 | 63500 | 63500 |    63500 | 63500 |  63500 |  63500 |
|   Parry |  2000 |  4000 |  3167 |     3167 |  7325 |   6000 |   4000 |
|   Dodge |  2000 |  4000 |  3167 |     3167 |  7325 |   6000 |   4000 |
| Mastery |  5500 |  1500 |  3166 |     3366 |  1500 |   4150 |   8150 |
|     Hit |  2550 |  2550 |  2550 |     2450 |   500 |    500 |    500 |
|     Exp |  5100 |  5100 |  5100 |     5000 |   500 |    500 |    500 |

We start with four “control” gear sets that prioritize hit and expertise cap before anything else.  Control/mastery (“C/Ma”) prefers to stack mastery with the excess 4000 points, while control/avoidance (“C/Av”) splits that extra rating between dodge and parry.  I also have a control/balance (“C/Bal”) set that tries to keep equal amounts of dodge, parry, and mastery.  And finally, there’s a second version of control/balance (“C/Bal-NC”) that shifts 100 hit and 100 expertise into mastery to see whether being about half a percent below hit and expertise caps has a significant effect on survivability.

At the other extreme, we have the avoidance-based sets.  These drop all but 500 hit and 500 expertise, freeing up additional stat allocation.  Thus, we have 10650 rating to allocate with these sets.  The pure avoidance set (“Avoid”) takes all of that and splits it between dodge and parry.  The avoidance/mastery (“Av/Mas”) set sheds a little of that avoidance to bring mastery up to 4150, while the mastery/avoidance (“Mas/Av”) set shifts the majority of that itemization into mastery, pushing it all the way to 8150.

Metrics:

We’ll use all of the same metrics as before to evaluate these gear sets.  Since I’ve covered these in the last installment, I’ll just quote myself:

S% is our Shield Block uptime in decimal form
mean is our mean damage intake in percentage of maximum DTPS
std is the standard deviation of damage intake for a 5-attack moving average
SBr(k) is the total number of Shield Barrier casts, in thousands
SBr<60(k) is the number of Shield Barrier casts at less than 60 rage, in thousands

RPS is our rage generation rate (i.e. X rage per second)
xsR(k) is the amount of excess rage, in thousands

80% is the percentage of spikes that are above 80% maximum throughput
90% is the percentage of spikes that are above 90% maximum throughput

If you see me use shorthand like “90% events for 4-attack strings,” it means we’re looking at what percentage of all 4-attack strings fell above 90% of maximum throughput – in other words, how many times did we take 4 attacks in a row that summed to 90% of 4 full hits.

Also note that the 90% metric is a subset of the 80% metric.  In other words, the 80% category tells us about all of the events above 80% throughput, which necessarily includes all of the 90% events too.  If we want to know how many events fall between 80% and 90%, we have to subtract the number in the “90%” row from the number in the “80%” row.

Shield Block Queue (SB):

The first of our three queues is the simplest: “spam Shield Block.”  We have a few expectations for how this queue will perform already.  Since it ends up wasting a lot of rage due to the Shield Block charge mechanism, it shouldn’t gain a large benefit from hit or expertise above what’s necessary to maintain Shield Block.  On paper, that’s 6.667 rage per second, but due to discretization and RNG we expect to need an average that’s a little higher than that to improve smoothing.  And Shield Block’s synergy with mastery should make mastery builds stronger than avoidance builds.  Let’s see how many of those expectations show up in the data:

|      Set: |     C/Ma |     C/Av |    C/Bal | C/Bal-NC |    Avoid |   Av/Mas |   Mas/Av |
|        S% |   0.6667 |   0.6667 |   0.6667 |   0.6667 |   0.6637 |   0.6657 |   0.6665 |
|      mean |   0.5841 |   0.5849 |   0.5857 |   0.5837 |   0.5372 |   0.5361 |   0.5350 |
|       std |   0.1340 |   0.1440 |   0.1393 |   0.1394 |   0.1611 |   0.1538 |   0.1431 |
|       RPS |   8.2311 |   7.9791 |   8.0795 |   8.0686 |   6.9497 |   7.1282 |   7.3737 |
|    xsR(k) | 938.5090 | 787.2990 | 847.5390 | 841.0140 | 187.7190 | 282.4960 | 425.3690 |
|    ------ |    --- 2 |   Attack |   Moving |  Average |   ------ |   ------ |   ------ |
|       80% |  22.4295 |  24.0775 |  23.4485 |  23.1582 |  20.2805 |  19.3815 |  17.7728 |
|       90% |   8.1348 |   7.8957 |   7.9970 |   7.9205 |   7.4148 |   7.1505 |   7.0875 |
|    ------ |    --- 3 |   Attack |   Moving |  Average |   ------ |   ------ |   ------ |
|       80% |   8.4790 |   9.7037 |   9.2218 |   9.0358 |   8.0490 |   7.2245 |   5.9550 |
|       90% |   0.0000 |   0.0003 |   0.0000 |   0.0000 |   1.0903 |   0.5763 |   0.1998 |
|    ------ |    --- 4 |   Attack |   Moving |  Average |   ------ |   ------ |   ------ |
|       80% |   6.6083 |   8.9138 |   7.9930 |   7.6997 |   6.9833 |   5.6407 |   3.8593 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0000 |   1.2390 |   0.6270 |   0.2055 |
|    ------ |    --- 5 |   Attack |   Moving |  Average |   ------ |   ------ |   ------ |
|       80% |   4.5538 |   7.2175 |   6.1227 |   5.8265 |   5.4863 |   4.0145 |   2.2655 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0000 |   0.4768 |   0.2352 |   0.0798 |
|    ------ |    --- 6 |   Attack |   Moving |  Average |   ------ |   ------ |   ------ |
|       80% |   0.0000 |   0.0000 |   0.0000 |   0.0000 |   1.1093 |   0.5568 |   0.1750 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0000 |   0.1623 |   0.0907 |   0.0290 |
|    ------ |    --- 7 |   Attack |   Moving |  Average |   ------ |   ------ |   ------ |
|       80% |   0.8303 |   1.4797 |   1.1968 |   1.1203 |   1.6803 |   0.9902 |   0.4093 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0000 |   0.1500 |   0.0823 |   0.0275 |

First, let’s look at the TDR metric (“mean”).  There’s not a lot of variation amongst the different control strategies, at least not enough to draw statistically significant conclusions about mastery and avoidance.  But there’s a very clear drop in damage taken when shifting to the avoidance gear strategies.  Shifting 6650 rating from hit and expertise into dodge and parry reduces damage taken by a little over 8%.  The difference between pure avoidance and avoidance/mastery isn’t that large either, but there seems to be a slight decrease in damage taken as we shift more of that avoidance into mastery.

So this simulation suggests that for TDR mastery is a little ahead of avoidance and both are far ahead of hit and expertise.  In other words:

mastery>avoidance>>hit/expertise.

How does that compare with our results from September?  Well, to refresh your memory…

N=50, tau=10000, stat=1500

|          |  dodge |  parry |    hit |    exp | mastery |
|     mean | 0.8957 | 0.9008 | 0.0556 | 0.0438 |  1.0000 |
|      std | 0.0626 | 0.0675 | 0.0677 | 0.0544 |  0.0580 |
| std_mean | 0.0088 | 0.0095 | 0.0096 | 0.0077 |  0.0082 |

…. we see again here the weakness of hit and expertise, and the dominance of mastery/dodge/parry for TDR

Yup, pretty much exactly the same.  Mastery ahead of dodge and parry, hit and expertise trailing far behind.  So we can feel pretty confident that those results were correct.  And they aren’t hard to rationalize either – the excess hit and expertise isn’t doing much for TDR because the change in Shield Block uptime isn’t very large.  We’re well over the 6.667-rage minimum required for Shield Block maintenance in the Avoidance gear set despite having minute amounts of hit and expertise, so we’re losing less than 0.1% of our uptime. That just doesn’t translate into a significant TDR cost.  We just generate less of the excess rage that we’d be wasting anyway.

What it does cost us is smoothness.  The control strategies do a much better job of eliminating 90% spikes. The avoidance gear sets do well for 2-attack strings, but we start to see lingering 90% spikes for 3+ attacks, and as the number of attacks goes up the pure avoidance set falls behind in 80% spikes as well.  And all of it is due to inconsistent rage generation, leading to periods of extended Shield Block downtime.

Within the different control sets, control/mastery  is clearly ahead of control/avoidance and control/balance.  Apart from 2-attack strings, which we generally disregard anyway, control/mastery consistently provides the lowest spike presence.  Curiously, the “uncapped” control/balance set is an improvement over the regular control/balance set.  This suggests that being exactly at or over cap isn’t as critical for warriors as it is for paladins, who saw a noticeable loss in this same comparison.  We’ll touch on that in more detail very shortly.

It’s worth noting that the mastery/avoidance set performs surprisingly well.  While it doesn’t completely eliminate 90% spikes, it has a significant advantage in overall spike presence.  In many cases, it experiences half as many spikes as the control/mastery set.  In a situation where an 80% spike is enough to kill you, the mastery/avoidance gear set is arguably the best choice, especially for the 4- to 5-attack strings we generally focus on.

You may have noticed that the avoidance gear sets are in the vicinity of 7 rage per second, which is above the theoretical 6.667 Shield Block threshold.  And despite that, we’re still seeing this gapping effect that leads to 90% spikes.  That and our interesting results for the uncapped control/balance set lead us to ask, “Exactly how much rage generation is optimal?”  We need to be a bit more precise than that if we want numbers though.

In particular, what we’re really interested in is how much hit and expertise we really need to maintain Shield Block.  We know that 6.667 is the theoretical limit, but we also know that it isn’t sufficient in practice because rage generation isn’t a nice, smooth, continuous thing.  So let’s try and determine exactly how much hit and expertise it takes to push 90% spikes below an acceptable threshold – let’s say 0.001%.  To do that, we’ll take the control/balance gear set and just hack away rating until we get to zero.  Technically we’d be able to allocate that itemization somewhere else (say, mastery), but for this calculation we’ll pretend that we can’t, and that we’re simply losing itemization.  Maybe it’s being shifted to Stamina, for example.  The reason is that it will give us a better “worst-case” estimate on our hit and expertise needs.  Since the 4-attack moving average seems to have the worst time with 90% spikes in the avoidance gear set, we’ll use that.

If we perform that calculation, it looks something like this:

Spike damage presence vs. combined hit and expertise percentage. The inset shows an expanded view of the results above 15% combined hit and expertise.  The dotted line on the inset is 0.001%.

With no hit or expertise, our spike presence (the percentage of events exceeding 90% throughput) is a little over 3.5%.  It drops fairly dramatically with hit and expertise up until about 10% on the x-axis, where it’s starting to slow down in effectiveness.  The surprising part here is just how much hit and expertise it takes to eliminate those spikes.  Even at 15% combined hit and expertise (5% hit, 10% expertise), about 0.03% of all events exceed the 90% spike threshold.  While that’s certainly not a lot of spikes, it’s still well above our self-imposed 0.001% limit.  It isn’t until around 21% combined hit and expertise (7% hit, 14% exp) that we consistently drop below 0.001% spike presence, and it’s only above 22.3% hit and expertise that we finally hit zero.  The behavior in that final section of the x-axis is a little easier to see on this semilog plot, for those that are interested.

So while it’s not critical to be exactly at hit/exp cap, this plot tells us that we start opening ourselves up to 90% spikes if we’re even 1% shy of the cap.  It also suggests that warrior rage generation is “bursty” enough that it pays to be well above the theoretical minimum rage threshold for Shield Block, because a missed Shield Slam at the wrong time is all it takes to create a dangerous spike event.  As such, it’s probably not a bad argument to maintain caps if you’re going with a control strategy.

In summary, this data seems to suggest two viable gearing paths for the “Shield Block spam” finisher queue.  We either stack hit and expertise to cap and then focus heavily on mastery (the standard Hit/Exp > Mastery >> Avoidance control strategy), or we try and push mastery sky-high with a side of avoidance (the Mastery >> Avoidance > Hit/Exp strategy).  Either of those two gives a pretty good result, with one being better at eliminating the high 90% spikes and the other being better at minimizing spikes above 80%.  I also wouldn’t be surprised if a Mastery >> Hit/Exp > Avoidance strategy would perform as well as the mastery/avoidance strategy – the high mastery value seems to be the key component of that queue’s performance.

Shield Barrier Queue (SBr*):

We expect the Shield Barrier queue to tell us a very different story.  Since Shield Barrier isn’t limited by a charge system the way Shield Block is, we should be able to take advantage of rage generation much more effectively.  That should make hit and expertise much more powerful stats than they are for the Shield Block queue.  Mastery will likely suffer a bit given the lack of guaranteed blocks, while avoidance should get stronger because it doesn’t compromise the Shield Barrier absorb mechanic.  Let’s see if the data confirms those predictions:

|      Set: |     C/Ma |     C/Av |    C/Bal | C/Bal-NC |   Avoid |  Av/Mas |  Mas/Av |
|        S% |   0.0000 |   0.0000 |   0.0000 |   0.0000 |  0.0000 |  0.0000 |  0.0000 |
|      mean |   0.4254 |   0.4044 |   0.4137 |   0.4155 |  0.3935 |  0.4087 |  0.4257 |
|       std |   0.2002 |   0.2044 |   0.2030 |   0.2054 |  0.2180 |  0.2164 |  0.2123 |
|    SBr(k) | 100.0000 | 100.0000 | 100.0000 |  99.9620 | 98.1860 | 98.1530 | 98.2380 |
| SBr<60(k) |  66.2540 |  65.9270 |  66.2830 |  66.3450 | 73.8420 | 74.0180 | 73.6990 |
|       RPS |   7.3601 |   7.3745 |   7.3637 |   7.3252 |  6.3873 |  6.3768 |  6.4139 |
|    xsR(k) |   0.0000 |   0.0000 |   0.0000 |   0.0000 |  0.0000 |  0.0000 |  0.0000 |
|    ------ |    --- 2 |   Attack |   Moving |  Average |  ------ |  ------ |  ------ |
|       80% |  17.6300 |  16.7297 |  17.2497 |  17.2058 | 15.4113 | 16.2243 | 16.7793 |
|       90% |   8.9105 |   8.1948 |   8.4980 |   8.5545 |  8.5440 |  9.0595 |  9.5497 |
|    ------ |    --- 3 |   Attack |   Moving |  Average |  ------ |  ------ |  ------ |
|       80% |   5.7390 |   5.3152 |   5.5655 |   5.5515 |  5.0933 |  5.4630 |  5.6943 |
|       90% |   4.1412 |   3.8148 |   3.9960 |   3.9415 |  3.6448 |  3.8937 |  4.1227 |
|    ------ |    --- 4 |   Attack |   Moving |  Average |  ------ |  ------ |  ------ |
|       80% |   2.3103 |   2.1128 |   2.1852 |   2.4530 |  4.0907 |  4.3530 |  4.5382 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0090 |  0.4575 |  0.4857 |  0.4927 |
|    ------ |    --- 5 |   Attack |   Moving |  Average |  ------ |  ------ |  ------ |
|       80% |   2.1305 |   1.8900 |   2.0082 |   2.3093 |  3.3967 |  3.6865 |  3.8115 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0025 |  0.1555 |  0.1677 |  0.1752 |
|    ------ |    --- 6 |   Attack |   Moving |  Average |  ------ |  ------ |  ------ |
|       80% |   1.1298 |   0.9695 |   1.0538 |   1.2115 |  1.7618 |  1.9700 |  2.0035 |
|       90% |   0.0000 |   0.0000 |   0.0000 |   0.0005 |  0.0513 |  0.0598 |  0.0610 |
|    ------ |    --- 7 |   Attack |   Moving |  Average |  ------ |  ------ |  ------ |
|       80% |   0.4732 |   0.3762 |   0.4383 |   0.5012 |  1.1652 |  1.3668 |  1.4605 |
|       90% |   0.0053 |   0.0040 |   0.0032 |   0.0080 |  0.1808 |  0.2100 |  0.2143 |

The TDR stats are a little surprising.  The control strategies certainly compete much better here, but they’re still slightly behind what one accomplishes with the Avoidance gear set.  So hit and expertise have improved considerably, but they still fall behind avoidance in terms of TDR.  Mastery suffers as predicted, with avoidance/mastery and mastery/avoidance falling behind the pure avoidance set.  This disparity is mirrored in the difference between control/mastery and control/avoidance.  In fact, mastery has fallen behind hit and expertise in this data; going from the control/avoidance set to the mastery/avoidance set, which is equivalent to shifting 6650 hit/exp into mastery, gives us a net increase in damage taken.  So our TDR priorities seem to be fairly simple in the Shield Barrier queue: Avoidance > Hit/Expertise > Mastery.

The smoothness metrics are a little different though.  There isn’t a whole lot of difference between the gearing results for 2- and 3-attack strings, though avoidance has a slight edge there as usual.  But as we start looking at longer strings, the control strategies rapidly pull ahead in both 80% and 90% categories.  Control/avoidance performs better than control/mastery, which isn’t too surprising since an avoid extends the life of a Shield Barrier bubble.  But any of the control schemes are a better choice than the avoidance ones for smoothness.  Thus, our smoothness stat priorities would be Hit/Exp >> Avoidance > Mastery.

It’s clear from this that mastery is the odd man out for Shield Barrier spamming.  While it’s not a bad stat, it just doesn’t stand out when you eliminate one of it’s main synergies (Shield Block leading to more critical blocks).  The TDR differences are relatively small compared to the differences we see in the smoothness metrics, so if we want to come up with a general set of stat priorities for this finisher queue I think hit and expertise “win.”  Overall, I’d go with Hit/Exp > Avoidance > Mastery if I were spamming Shield Barrier.

First-come first-serve queue (F-110):

This one’s a little harder to make predictions for.  Since it uses both Shield Barrier and Shield Block, it’s got a weird mix of mechanics.  Shield Block makes mastery strong and weakens avoidance, while Shield Barrier does exactly the opposite.  The only thing that I think I can successfully predict here is that hit and expertise will be strong for smoothness metrics since they seem to always be strong for smoothness. My guess is that they’ll probably fall behind significantly in the TDR department, though, because they’re only a little better for TDR with SBr* and a lot worse for TDR with SB.  Let’s see what the data says:

|      Set: |    C/Ma |    C/Av |   C/Bal | C/Bal-NC |   Avoid |  Av/Mas |  Mas/Av |
|        S% |  0.4255 |  0.4024 |  0.4123 |   0.4126 |  0.2749 |  0.2846 |  0.3067 |
|      mean |  0.4776 |  0.4746 |  0.4763 |   0.4759 |  0.4453 |  0.4534 |  0.4600 |
|       std |  0.1562 |  0.1642 |  0.1619 |   0.1617 |  0.1861 |  0.1835 |  0.1770 |
|    SBr(k) | 57.4520 | 59.7650 | 58.7690 |  58.7180 | 71.0220 | 70.1040 | 68.0820 |
| SBr<60(k) | 55.8980 | 58.8670 | 57.6260 |  57.5240 | 70.8290 | 69.7730 | 67.5520 |
|       RPS |  7.9169 |  7.7466 |  7.8128 |   7.7934 |  6.6260 |  6.6983 |  6.8472 |
|    xsR(k) |  0.0060 |  0.0030 |  0.0150 |   0.0010 |  0.0000 |  0.0010 |  0.0000 |
|    ------ |   --- 2 |  Attack |  Moving |  Average |  ------ |  ------ |  ------ |
|       80% | 14.1575 | 14.5208 | 14.5025 |  14.2623 | 14.5012 | 14.8190 | 14.6805 |
|       90% |  5.2595 |  5.3152 |  5.3080 |   5.3175 |  6.9227 |  7.1638 |  7.1575 |
|    ------ |   --- 3 |  Attack |  Moving |  Average |  ------ |  ------ |  ------ |
|       80% |  5.6348 |  5.8228 |  5.8165 |   5.6595 |  5.2422 |  5.4765 |  5.3887 |
|       90% |  3.3908 |  3.4208 |  3.3955 |   3.3407 |  3.3657 |  3.5343 |  3.5292 |
|    ------ |   --- 4 |  Attack |  Moving |  Average |  ------ |  ------ |  ------ |
|       80% |  1.6260 |  2.0728 |  1.9735 |   2.0240 |  3.7503 |  3.8285 |  3.6620 |
|       90% |  0.0915 |  0.1422 |  0.1265 |   0.1265 |  0.4427 |  0.4622 |  0.4025 |
|    ------ |   --- 5 |  Attack |  Moving |  Average |  ------ |  ------ |  ------ |
|       80% |  1.3987 |  1.8645 |  1.7230 |   1.8560 |  3.1955 |  3.2955 |  3.0950 |
|       90% |  0.0000 |  0.0000 |  0.0000 |   0.0025 |  0.1138 |  0.1293 |  0.1172 |
|    ------ |   --- 6 |  Attack |  Moving |  Average |  ------ |  ------ |  ------ |
|       80% |  0.6168 |  0.8048 |  0.7702 |   0.8377 |  1.5735 |  1.6660 |  1.5755 |
|       90% |  0.0000 |  0.0000 |  0.0000 |   0.0013 |  0.0398 |  0.0473 |  0.0415 |
|    ------ |   --- 7 |  Attack |  Moving |  Average |  ------ |  ------ |  ------ |
|       80% |  0.3443 |  0.4570 |  0.4295 |   0.4820 |  1.1385 |  1.2465 |  1.1983 |
|       90% |  0.0003 |  0.0005 |  0.0017 |   0.0065 |  0.1555 |  0.1620 |  0.1470 |

It looks like I was partly right about the TDR metrics.  Control falls behind avoidance, suggesting that in this mixed queue avoidance is significantly ahead of hit and expertise.  But the disparity is not as great as I expected, either.  I attribute part of that to the sheer efficiency of Shield Barrier – in terms of raw TDR it just eliminates more damage per rage than Shield Block can, and that makes up for a lot of ground.  We see that the mastery/avoidance gear set gives up the bulk of the TDR advantage that the avoidance gear configuration demonstrates, suggesting that mastery is strictly worse than avoidance for TDR.  Mastery/avoidance is still ahead of control/avoidance, though, suggesting that mastery’s still ahead of hit and expertise by a good bit.  So our TDR stat priorities end up being Avoidance > Mastery > Hit/Expertise.

Once again, the smoothness metrics seem to favor the control strategies.  We see a significant reduction in both 80% and 90% spikes for 4+ attacks by shifting from an avoidance-based configuration to a control-based configuration.  I think a lot of this disparity is due to the differences in Shield Block uptime.  The control sets boast over 40% uptime, while the avoidance sets barely manage to maintain 30%.  There’s also a significant disparity in rage generation, with avoidance sets nearly a full 1 rage per second behind the control sets.

There’s also a bit of a chicken-and-egg scenario going on here.  The control sets are generating more rage and thus having to push Shield Block casts back less often, resulting in higher Shield Block uptime.  But that increased Shield Block uptime leads to more critical blocks, which increases rage generation through Enrage.  There’s diminishing returns on that feedback loop, of course, but the interaction is definitely having an effect.

Mastery seems to be a better choice than avoidance in the smoothness category as well.  Control/mastery consistently outperforms control/avoidance by a rather healthy margin.  And the mastery/avoidance set also outperforms the pure avoidance set. though the differences are much smaller.  Again, the lower Shield Block uptime is hurting mastery in the avoidance sets, suggesting that its value is heavily dependent on how often you’re casting Shield Block.  All told, this indicates that the smoothness stat priorities for the F-110 queue should be Hit/Expertise >> Mastery > Avoidance.

It’s interesting that the TDR and smoothness stat priorities differ so completely.  But I think that the TDR variations are so much weaker than the smoothness effects that we can more or less ignore them.  So the overall recommendation for this queue would be Hit/Expertise > Mastery > Avoidance.  Though I think you could make a pretty good argument that hit and expertise  should be “>>” (much greater than) the other two based on the smoothness metrics.

Conclusions:

There are a few obvious trends that stick out when you consider all three data sets.  First, hit and expertise are consistently good smoothing stats.  No matter what queue you happen to be using, hit and expertise tend to do a good job of smoothing out rage generation and subsequently finisher use, giving smoother damage intake profiles.  The SB queue is the only situation where anything else came close in the smoothness department, primarily because of the sheer amount of excess rage being wasted due to the lack of some sort of low-cost rage dump.

On the other hand, hit and expertise were generally poor for TDR.  It was only in the SBr* queue that they managed to crawl out of last place, and only because mastery has no interaction with Shield Barrier.  In the other two cases, they were distant runners-up.

This is pretty similar to how the paladin simulations came out.  Hit and expertise aren’t that great for raw TDR, but dominate the control category.  From a design standpoint, that’s how it ought to be, because it builds a trade-off into the system.  Want smoother damage intake?  Then be prepared to take a little more damage overall.  It ensures that there’s no “king stat” that excels in every category, to the point that you ignore everything else.

It’s worth noting that, just as in part 2, the F-110 queue gives SB a serious run for its money.  Both seem to excel with a control/mastery gearing scheme, though SB seems to perform well with a mastery/avoidance build as well.  But while F-110 does permit more 90% spikes, it maintains lower overall spike presence in control/mastery than SB does in either of the two gear sets.  I think it’s safe to say that F-110 is a legitimate competitor when it comes to finisher queues.

But no matter the queue, it’s clear that for damage smoothing you’ll want to prioritize hit and expertise rather highly.  For any queue that includes Shield Block, it seems that  Hit/Exp >  Mastery > Avoidance will be your go-to gearing priority.  If you’re ignoring Shield Block entirely, then mastery drops to last place for smoothing.

This entry was posted in Tanking, Theck's Pounding Headaches, Theorycrafting and tagged , , , , , , , , , , , , , , , , . Bookmark the permalink.

### 16 Responses to 5.1 Warrior Simulations – Part 3 – Stat Priorities

1. Omega says:

I stick with hit/exp >mastery for now until i get out of a tier that has lei shi. FUCK that boss. no revenge procs, no melee damage what so ever so avoidance isn’t worth a shit, neither is mastery for that matter. it’s a fight where hit/exp are EVERything and everything else is stamina. that boss still kills me more than any other boss by far. can’t wait to get to a tier again that has none of that. heh

• Dreadski says:

Gems and reforges are cheap. I have a reforge yak now and also happen to be a JC so none of it is really an issue.

• Dreadski says:

But the data from Theck’s report still states that Hit/Exp capping is better for smoothing damage. Smooth damage is what we want, so F TDR until it becomes useful in some imaginary distant expansion. This is the old avoidance vs stacking stam and armor argument, except avoidance became mastery and avoidance, and effective health became keeping up shield block.

2. bryjoered says:

Fights like Lei Shei are one of the main reason why gemming for stamina seems to be the best all around solution if you don’t like to change up your gear per encounter. Mastery is very valuable if there is blockable damage, but it’s nothing compared to what it was in Cataclysm. Until my healers are complaining about running out of mana healing me I’m sticking with Stamina as my number one stat, now followed by hit/exp>Mastery.

3. tigerlol says:

Great stuff as always. I think there is a typo at this sentence: “At the other extreme, we have the avoidance-based sets. These drop all but 500 hit and 50 expertise”

• tigerlol says:

Or am I understanding the table incorrectly?

• Theck says:

Nope, definitely a typo. Should be 500 hit and 500 expertise. Thanks for catching that, I’ve corrected the text.

4. bryjoered says:

Theck, how would you weight hit/exp compared to Mastery for shield block only? Is it way higher up till cap?

• Theck says:

I’m pretty sure I answered that question at the end of the Shield Block queue section.

5. Whaler says:

AMR does the stat wieghts. Looks like cap hit + exp to cap, then everything to mastery.

• Theck says:

AMR doesn’t do any simulation to determine the stat weights. They get them from theorycrafting, like, for example, this post.

Where do you think their paladin stat weights came from?

6. bryjoered says:

Well, technically capping hit is very slightly more important than capping expertise because Thunderclap cannot be dodged or parried. Hit, in my opinion, is only more valuable because it your aoe tanking talents like shockwave, dragon roar, and TC all can’t be dodged or parried. There is aoe tanking in encounters in every tier.

• bryjoered says:

After tanking terrace of the endless spring, I really can’t begin to tell you how valuable stamina is in that raid. Every encounter it seems has huge magic damage coming in at all times, which not only makes stamina much more valuable, but barrier and therefore hit/exp as well. I seriously feel at this point, that any warrior tank that isn’t getting as much stamina as they possibly can is only making it harder on themselves and their healers. I mean, you should generally go for socket bonuses, but every gem should have stamina on it period in my opinion.

• Tim says:

Theck has stated in several of his other posts that he doesn’t include stamina into his simulations because of the complexity it brings. I can’t find the exact post he states it in, but he mentions effective health (EH) is the single largest priority for tanks. These simulations he has been doing for both paladins and warriors is trying to help determine the best survivability stats after maximizing EH.

7. Pingback: Revenge Combinatorics | Sacred Duty

8. Krennick says:

I have no painfully long reply to this. It is good work. I just wish a viable queue that strives for S% at 0.6667 was included. (Or three)