Hmm. Maybe if you parry a swing when there's less than 20% on the swing timer, it doesn't just drop the parry, but hastes the *next* attack instead?

That would then appear in the logs as a weapon swing of 2.16 seconds without any parry occurring during the swing. If I understand what you're doing correctly, you wouldn't see these, as you're only looking at swing intervals which do contain a parry.

However, if you got a parry-hasted swing immediately following one of these "saved" parry charges, you might get a double parry effect. That is, the swing would be hasted twice even though only one parry occurs during the swing itself. The total amount of haste on such a double-parry should be 2.304 seconds, which is very close to your observed values.

To confirm this, could you check:

a) Whether there are any "hasted" swings (i.e. swings at an interval of 2.16 seconds) which do not contain a parry.

b) If there are, do they always follow a "wasted" parry (i.e. a parry that occurs within the 20% cap zone)

c) Whether the "super-hasted" swings always follow a "wasted" parry.


Edit: Hang on, just noticed the other sheets in the workbook - I'll check myself

Edit2: And the answer is that I lack the brain power to work out what's going on. I tried to have a look at the swing times for hits immediately subsequent to "wasted" parries, but couldn't make out a pattern. There were hits at 3.6 seconds, 2.6 and even 4.6 seconds. There also seemed no obvious pattern of "wasted" parries immediately preceding the super-hasted swings. So. A mystery.

Since I appear to be blessed by God with an insanely stable (ping wise) internet connection, is there anything else I could do to help you guys out? I don't have access to a shaman of appropriate level, but maybe I could use the PTR or "borrow" someone else's.

To be objective, maybe yours is the norm, and it's just that mine is super sucky ;-) In any case, if a pally or rogue would be able to do some testing, I'll let you know, but you've already been incredibly helpful, so hopefully no more multi-hour test sessions will be necessary.

I finally got around to parsing some of my test runs. The lag's pretty bad -- this was off the 2.2 PTR after all -- but I've got a lot of data. It looks like main hand attacks get the parry haste, and off hand attacks do not.

These are using two 2.7 speed swords.
MH: [Protector's Sword]
OH: [Protector's Sword]

Awesome work - thanks man!

As far as the cost is concerned, it takes 22.4 parry rating to get a full 1%, whereas it takes 18.9 dodge rating to get 1%, so the cost is not huge at all. Against hard hitting bosses that allow liberal use of Heroic Strike, the haste from parry is naturally going to allow for more Heroic Strikes (assuming you have the rage for it). Realistically you're not going to often be choosing between dodge or parry unless you're gemming, but personally I think the significant haste makes the small extra cost worth choosing parry over dodge.

You guys should be commended for this work. Good job on the data gathering and analysis. I think its simply a verification in a way, although the OH parry mechanic is interesting.

Did anybody test with a mob parrying?
If so, it could prove useful for the weapon skill -> parry mechanic discussion.

That is, if you still know your weapon skill and the mob level.

Thanks for this research. I would love to have your ping.

For practical purposes, I'd like to update the guide on the WoW-US forums. What are the caveats, if any, we are seeing to the original Parry entry on the WoWWiki?

The original Parry entry was:


Anyway, great work again, it was difficult to follow this before because the theory had not been verified nor did we know the researcher. As brought up here, the people testing have to have pretty darn good ping, or be willing to test for a very long time, to be accurate.

I'll leave the formal mathematical proof to someone else, but from the graphs in this threat and the theoretical numbers the average haste increase from parry can be fairly easily deducted:

40% of your swings get the maximum of 40% speed reduction (to 0.6 * weapon speed).
20% of your swings get no speed reduction at all (1.0 * weapon speed).
The remaining 40% get reduced by on average the middle of these (0.8 * weapon speed).

So your average swing speed as a result of a parry is 0.4*0.6 + 0.2*1 + 0.4*0.8 = 0.76. In other words, 1% parry is worth 0.24% haste.

This is independent of your weapon speed, as everything here are percentages.

This also makes it possible to calculate the value of the parry in terms of dodge and haste rating:

1% parry is approximately equal to 1% dodge + 0.24% haste.
1% parry is 23.65 parry rating.
1% dodge + 0.24% haste is 18.923 dodge rating + 15.76*0.24 = 3.782 haste rating, or a total of 22.705 rating.

In other words, even with the cheaper parry rating from patch 2.1 and the more expensive haste rating from patch 2.1.2 (?) parry rating is still too expensive, even not taking into account the fact that haste rating is generally a poor tanking stat.

I would expect a lower haste value than calculated if mob attacks faster than the player (due to double parries not being as effective). However the mob was attacking slower than you when you were using the dagger and you still got lower than theorycrafted results.

What I really wonder if your simplification really causes a higher number than reality in all these examples.
For a mob attacking slower than you it should be easy:

On each attack your parry chance would be parry*1.8/2.0

This is always lower than yours (no matter what parry chance), with 23% parry you were supposed to get ~12.6% more hasted attacks than you calculated. Then the second order calculation would say that the extra haste made you have an average weapon speed lower than 1.8 thus requiring re-adjustment of the original formula, and same can be done for 3rd order and so on until sufficient accuracy is reached.
With almost 6% haste your speed changes to almost 1.7 resulting in 5.8% less procs, which is still more than what you calculated (as we multiplied by 1.126 and divided by 1.058).

Calculating a slower wepaon is more complicated due to double parries, but unless we can excuse your 1.8s weapon results to variance I wouldn't say ALL your tests were lower than expected because of the formula's inaccuracy, as you can see for the 1.8s weapon I fixed the formula to be accurate enough to show that the values don't match. This is unless your data pool is too small, which you could in fact calculate to see the estimated error in your measurement as error in number of parries = number of parries ^ 0.5, and thus error in % of attacks getting parried (at least using the first order, linear formula) would be estimated as (number of parries^0.5)/(number of attacks).

In 1 hour you would get 2000 attacks and 414 of them should have been parry hasted, so if you would get an amount of attacks hasted close to the theorycrafted value the error should be 20, or ~5%. Since 5% of 5% is very small, even though the error estimation is very very rough, I doubt it's much higher than this and thus the statistical error in your measurements is probably not the explanation either.

There must be something horribly wrong with the theorycraft that wasn't related to in my post that would exaplain the difference between the calculated and measured results at least for the 1.8 speed weapon as the formula for chance for an attack to be parry hasted (as well as how to figure out its 2nd order, 3rd order and so on until sufficient accuracy is met) is pretty obvious.


On a side note you say your formula is a first order apporximation but a first order (unless mixing math and english is failing for me again as english isn't my first languege) is by definition linear, and yours is not.

EDIT: Just noticed that the slower the weapon the closer your formula gets to the measured number, so assuming the statistical error is small on the other weapon speeds too (which it honestly should be with so many measurements, as shown above), this fact can help finding what is wrong with the theory. But the bottom line is that it's so off even after using my more accurate formula for the 1.8 speed weapon that you can't excuse it for being "chance calculation inaccuracy".