So, from the moment I first saw Dragonspine Trophy start dropping for people, I've been sort of curious about it - it's clearly good, but the mathematician in me has always been curious about just *how* good. Well, the other day I saw someone post a combat log for half an hour in the blasted lands, so decided to run some analysis on it. If you want to see the source material, it's linked from the wowhead comments on the item - url is http://putstuff.putfile.com/57121/1853616

Now, there were a couple of interesting revelations here.

First off, lets look at the duration of the procs. There were 51 periods of haste all told:

9.86
9.89
9.906
9.922
9.937
9.953
9.953
9.968
9.968
9.969
9.969
9.984
10.0
10.015
10.016
10.016
10.016
10.031
10.031
10.031
10.046
10.047
10.047
10.062
10.062
10.062
10.063
10.063
10.063
10.141
10.156
10.156
10.157
10.157
10.157
10.172
10.172
10.828
12.453
14.266
14.39
14.453
14.968
15.594
16.188
17.5
17.562
19.203
20.812
20.859
23.344

Now, 37 of those are 10.2 seconds or under - presumably the 10 seconds that it's supposed to run. But 14 of them are longer than the usual duration - 3 of them by more than a factor of 2. Hence, these must reflect it proccing and refreshing itself while already up.

Observation 1: Dragonspine Trophy has no hidden cooldown like Hourglass of the Unraveler does.

Counting in the (at least) 17 refreshing procs, there were actually 68 procs all told. Hence the proc rate, massaging some numbers around, is about 1.7 PPM in this test. Unfortunately, there's not enough information here to determine if it's a fixed percent chance or a proc per minute chance.

So, that's all pretty basic stuff, and I'm sure someone has already figured it out. The interesting (and in fact, concerning) part of it is what happens when the actual effect of the haste is measured.

So, the methodology of this is as follows: I wrote a python script that went through and stripped out the time between each two offhand swings (I chose offhand swings rather than MH so as to remove the effects of sword specialization). I then compiled two lists of swing spacings, one for swings that occurred when haste was not active, and one it was. I through out any swing that spanned the start or end of haste. That is: haste starts (and stops) between two swings. Since that attack is "partially hasted" I chose not to count it towards either list. The only swing timers included are those that lie entirely inside or outside a hasted interval.

Now, there were a few high outliers in this data. Some of these were due to gouge/bandages with intervals of order 20 sec. However, there were also a handful of shorter ones, still well above the usual swing interval. My interpretation of these were that they were indicative of times when an attack missed or was dodged, since there are no misses or dodges listed in the log. Hence, I removed these from the data as well.

These corrections made, we're left with 961 unhasted swings (ranging from 1.125 to 1.953 seconds) and 452 hasted swings (ranging from .859 to 1.469 seconds). Computing the average swing duration in each case, we find that the regular swings averaged 1.422 seconds between attacks, while hasted swings averagesd 1.169 seconds between attacks. This means, that in a given interval of time, with the haste effect active, one will land roughly 21.64% more swings than one would unhasted.

But wait! 325 haste rating was, at level 60, 32.5% haste rating. This was reduced by the standard factor of 1.577 to 20.61%, and then buffed by 50% in patch 2.0.7, so it *should* be giving 30.91% more attacks - why, then, is it only giving 21.6% more attacks? I see only once possible explanation:

Observation 2: Dragonspine Trophy is not benefitting from the 50% buff to haste that theoretically occurred in patch 2.0.7. Note that is *says* it gives the right amount - see http://hem.bredband.net/perwin84/AFGM/dragonspine.gif (linked from Thottbot entry for the item) for instance - but the actual attack rate is not consistant with this display

It is undetermined as of now whether this is true of all items that give haste rating, or just of dragonspine trophy. It would be interesting to test with another item that gives haste rating (Abacus of Violent Odds, for instance) and see if it exhibits the same behavior.

Ald

A factor you're not considering that throws off such data from solo farming mobs is the reduced swing timer after you parry a mob's attack. I suspect this is a large portion of the wrinkle in your data.

I'm not actually familiar with such an effect... how is it supposed to work?

Regardless, throwing out the attacks that span a parry, we still have 433 hasted attacks and 912 regular attacks, the means are 1.422 and 1.168, and the speed increase is 21.77%. On the whole: this does not appear to be a significant effect.

Ald

According to the combat mechanics testers parrying an attack resets your melee swing timer to 0 (or some very low figure) ie if you are half way between two swings, and parry an attack you now get to attack right away, instead of having to wait for the remaining half of your timer.

Ald

IIRC the posted data on these threads, a parry does not reset your swingtimer but reduces it significantly by a given percentage.
But my common sense says this would only apply to your mainhand swings. So parsing OH swings only should be not affected by "parry haste".

Just unequip any haste items, and test your parser. Should give an accurate measurement of your OH weapon delay.

yeah and get some flowers for the offhand or a fish

Well, I don't actually have the trinket myself, so can't take additional data - this was based on data taken by someone on wowhead. However, if someone who *does* have the trinket would like to post their combat log (preferably with no OH weapon, and with misses/dodged/etc. displayed) I can certainly run it through the same analysis. In a totally ideal world it would also be done with a weapon of a different speed so as to be able to infer something about the proc rate.

It doesn't necessarily have to be giving the actual 325 rating, it could be a discrepancy between tooltip and function.

I don't think you should be discarding all the swings that were "partially hasted". Last we saw in the hunter math threads, there's no such thing as partially hasted swings. The swing timer works by using your current haste to determine time until your next swing. If you have a haste effect up when swing1 is fired, that haste will be used to calculate when swing2 lands, even if the haste expires before that time. However if the haste does expire before swing2 lands, then the time between swing2 and swing3 will not be reduced by haste.

Would it be helpful to post a combat log from a hunter? We can guarantee no misses/dodges/parries, no sword spec or off-hand/mainhand filtering, and probably no bandaging/gouging either - I assume your scripts could be run on other combat logs without much trouble?

I could only provide data from abacus and kots though, no dragonspine trophy for me (yet...). Anyone else want to chip in with the data?

edit: this is probably better placed in the Combat Mechanics forum btw

Interesting - I'll take a look at that and see if I can confirm/deny it. I wouldn't expect it to make a large difference though.


Sure, I can run numbers on other data sets. However, see below for an interesting update on the nature of this bug.

I made some additional modifications today to run on main hand attacks. I threw out all the stuff that I threw out before, but I additionally dropped any attack where the swing time was interrupted by a parry or sword spec. This resulted in 193 hasted MH attacks and 407 unhasted OH attacks, with the following result:

The regular swings averaged 2.76 seconds. The unhasted swings averaged 2.06 seconds - an increase of 34%, much closer to the predicted 31%.

Observation 2, revised: It appears that 2.0.7 haste buff was applied only to MH attacks, and not OH attacks.

I didn't see a combat mechanics forum - that forum is called "class mechanics", which this technically isn't. But if the powers that be feel it deserves to be elsewhere, they're free to move it. I don't see a way of doing so myself.

Maybe he missed seeing it go off? 30mins with 37 procs seems to be rather solid data that you shouldn't go tooo long without seeing a proc. He may just have had an unlucky streak however.