About Attack Duration buffs

[Gabli]

I’ve recently taken the time to make a bunch of measurements concerning the effect of “- x% Attack Duration” bonuses. The basics, which are not that basic, are from this thread from moebius92.

As a reminder, here is the way things seem to work when you fire a “standard” skill (neither marked Fast nor Immediate in the tooltip):

• If the skill has an induction, it happens first.
• After the induction time is ended, the skill starts activating. This lasts for a duration noted “ACT” and is visually spotted by the skill’s icon borders being green.
• After the skill has finished activating, its cooldown phase starts. For some skills, any other standard skill that is queued will not start until a post-skill delay (“PSD”). The sum of ACT and PSD, called “global cooldown” and noted GCD, is noteworthy as Attack Duration buffs are directly affecting this time laps.

Variations occur when using a “Fast” or “Immediate” skill while a previous skill is activating:

• Fast skills allow to negate the previous skill’s PSD: the latter starts activating immediately after the previous one finished activating.
• Immediate skills start activating the moment they are clicked/keyed. Such a skill negates not only the previous skill’s PSD, but potentially its ACT too (bar anti-macro bugging). As such they also allow to start the previous skill’s cooldown sooner. According to moebius92 a third skill can’t start activating until the last of both skills’ GCD have ended: this is not something I tried to confirm but there is no reason to assume it works any other way now.

You’ll find in the following posts in this thread:

• The main results of my tests.
• The general idea of how ACT, PSD and GCD are measured.
• An explanation of how measurements’ accuracy is estimated.
• Some details on tests conducted on my (dwarf) Champion.
• Some details on tests conducted on my (female elf) Hunter.
• Details on the determination of the cap value of Attack Duration bonuses.

[Gabli]

Main results of my tests:

• Attack Duration bonuses are additive (unchanged).
• Attack Duration affects GCD and ACT differently (likely unchanged):
  
  • GCD is fully reduced: a skill with 1s GCD without any buff would have a 0.7s GCD if your total Attack Duration bonus was 30%, as long as the PSD is not reduced to 0 (see below).
  • ACT is reduced by a fraction of your Attack Duration buff. This fraction depends not on the skill type (melee/ranged, focus/induction, …) but on the weapon type (1H sword, 2H axe, bow, …). ACT times seem to usually be reduced by something close to AD%/x where x is somewhere between 4.5 and 5: 4.54 for 2 handed axes, 4.74 for two handed swords (Champions), 4.83 for bows (Hunters).
  • Preliminary tests, that I have not really pushed, seem to indicate that ACT times do not depend on character’s race but do depend on weapon type, same as the fraction of Attack Duration that reduces ACT durations.
  
  
• A skill’s GCD cannot go lower than its ACT. That means that if a skill has no PSD, additional Attack Duration buffs have lower value.
• There is a hard cap to the Attack Duration bonus, probably at either -39 or -40%. Beyond this point, Attack Duration has no influence whatsoever on skills and rotations execution times.

If some people are interesting in the xlsx spreadsheet that contains all the data gathered and used, I don’t mind sharing but I simply do not know where to host it.

[Gabli]

General idea of how to measure a skill’s ACT or PSD:

Thanks to this and moebius92’s plugin, it is possible to measure a skill’s ACT, PSD and GCD in various conditions as the plugin basically tells you the time at which any used skill goes on cooldown. Depending on what you want to measure, you either spam a given skill or follow it with a standard skill. The method is summed up in the schema below:



In the case of skills that don’t have cooldown like hunters’ Quick Shot, it is still possible to indirectly measure ACT and GCD by sandwiching the skill with two others.
For example, if we use the sequence [Barbed Arrow – Quick Shot – Low Cut], the plugin tells us when Barbed Arrow and Low Cut respectively finished activating. The difference between the two is composed of Barbed Arrow’s PSD, Quick Shot’s induction time and GCD, and Low Cut’s ACT. If Barbed Arrow’s PSD, Quick Shot’s induction time and Low Cut’s ACT are known we can deduce Quick Shot’s GCD. Uncertainty propagation (see next post, only for the brave) however means that the results are less precise than with straight spamming a skill.
Replacing Low Cut with Penetrating Shot allows to measure Quick Shot’s ACT rather than PSD.

[Gabli]

Measurements’ reliability: (be warned that this is a rather math heavy section)

Because I wanted to get an idea of the accuracy of the results, I tried repeating them. For example, if you spam Wild Attack (champion skill) the measured duration between two times it goes on cooldown will vary. Most values are fairly packed around the average, but you get variations of different orders of magnitude:

• Most are a few ms (0.001 s) away from the average. I suspect these variations are due to slight fluctuations in latency, but I can’t be certain as I don’t know exactly how the plugin works.
• You also sometimes get results a few 0.01s or even 0.1s away from the previously mentioned balled up results. I suspect these are caused by either “internet traffic” such as short but large latency fluctuations and packet losses, or anything that might or might not happen on the server’s side. Bottom line is these measurements are assumed to be way off and discarded.

Given this, it’s obvious for example that to measure a skill’s ACT it is more interesting to measure it x times by using it twice on cooldown rather than just using the time it takes to spam it x times on cooldown and dividing by the resulting time by (x-1).

The first method allows both to single out and discard obviously off measurements and prevent them from “polluting” the resulting average, and to measure the uncertainty u(ACT) that gives an idea of how far away from the average you can expect the “actual” value to be.

• The uncertainty of a single measurement would be the standard deviation (with 1 deduced degree of freedom) you’d get by repeating it an infinite number of times. The closest estimation would be your set’s standard deviation. This is not what is interesting to us.
• What we want to know is how close the actual value might be to the average of the measured data points. The best estimation of the actual value is the average you get, and the uncertainty on the measurement is the standard deviation divided by the square root of the number of data points.

Assuming a normal distribution (the famous gaussian or bell-shaped curves) of measurements around the average, one interpretation is that the actual value has a 68% chance not to be more than u(ACT) away from the measured ACT and 95% chance of not being more than 2*u(ACT) away. That is, obviously, assuming the average’s limit when repeating the measurement is the actual activation time.
Now, some measurements result from linear regression from data or direct calculations from same data.

In the first case, uncertainty on the slope and intercept can be evaluated with a Monte-Carlo simulation:

• For every data point, draw a random value around the measured one according to its uncertainty (there is a sqrt(3) factor to account for uniform distribution of the script’s RNG).
• With simulated points, perform a linear regression and store the slope and intercept each in an array.
• Repeat the previous steps N times.
• Evaluate uncertainties on the slope and intercept from simulated values, treating them as data points.

In the second case, the formula for propagation of uncertainty is used (the one with partial derivatives). For example, suppose we have measured a skill’s ACT and PSD and their respective uncertainties u₁ and u₂ and want to deduce GCD and its uncertainty u(GCD). Because we have GCD=ACD+PSD, we also have u(GCD)=sqrt((u₁)²+(u₂)²)
I also sometimes used the z-score to check whether it is reasonable to assume that two measured quantities are actually the same. The formula is Z=|x-y|/u(x-y): it’s a way to tell whether the difference between x and y can be assumed to be 0. If it is the case, the measured difference should like differ from 0 by less than 2 times its uncertainty. As such, Z<2 is usually a good indication that x=y is a reasonable hypothesis.

[Gabli]

Measurements done with a (dwarf) Champion:

With my dwarf Champion I measured the ACT duration of the following skills: Wild Attack, Swift Blade, Blade Wall, Brutal Strikes, Ferocious Strikes. These measurements were done both with a two-handed axe and a two-handed sword, for different “-x% Attack Duration” buffs. The AD buff was changed by using different combinations of Flurry proc (-20%), Hiddenhoard armour set (-10%), Heroic Alacrity words of craft (-3%) and Battle Reflexes trait (-[0-10]% in 2% increments). When I did not want Furry to proc, I used blue specialization to avoid the Precise Strikes buff to Attack Duration.

For all these but Swift Blade, you get a very nice graph when plotting ACT duration versus AD buff, with data points lining up almost perfectly in the [0;33] range of Attack Duration buff:



For reasons that I can’t explain, ACT measurements of Swift Blade are surprisingly off.

It can be noted however that going from -40% to -43% AD has no impact on skill activation time. There are two possible explanations:

• AD stops reducing ACT after a point somewhere around the 38-40% mark.
• Heroic Alacrity is bugged.

Since Heroic Alacrity does reduce ACT properly by bumping AD from 30 to 33%, the logical conclusion is that there is a cap to Attack Duration effect on skills’ activation time. Most measurement with my dwarf champion, whether with a two-handed axe or a two-handed sword, tend to indicate a cap at -39% Attack Duration (see last section).

With Brutal/Ferocious strikes for example, you can see that you shave off about 3.4 ms of activation for every 1% Attack Duration when Battle Reflexes is tiered from 0% to -10%. Given that for every data point the uncertainty is about 0.3 ms, a difference of 1% in Attack Duration is fairly easily measurable. This is a great opportunity to verify that AD buffs are indeed additive by comparing ACT duration with Flurry (-20% AD) and Battle Reflexes with Hiddenhoard armour buff (10%+10% AD). Were the buffs multiplicative, the latter total AD would be 19%. However, when comparing ACT of all skills in the two conditions we get extremely close results (numbers below are obtained with a two-handed sword):



It is also worth noting that z-score (see previous post) indicate that ACT measured with Flurry and BR+Raid bonuses are consistent with both ACT being equal and that AD bonuses are indeed additive.

Next thing I did was comparing ACT durations when using either a two-handed sword or axe, still for varying Attack Duration buffs. Axes being slower, I plotted the relative ACT increase (in %) when going from sword to axe versus Attack Duration buff:



Z-scores (see previous post) indicate it is reasonable to assume that relative increase in duration is the same for all skills except Swift Blade (for which, again, I have no idea why numbers are off). For that reason, I computed the average relative increase in ACT duration for the other skills and while the alignment is nowhere perfect, it is consistent with uncertainties on the measurements.
As we can see the relative difference gets smaller with larger Attack Duration buff, which is an indication that Attack Duration is more effective to reduce ACT with an axe than it is with a sword. Yet the sword is always notably faster than the axe.

Whether with an axe or a sword, it can be inferred that the effectiveness of Attack Duration on ACT is the same for all skills by using data points before the Attack Duration cap:



Because the lines don’t cross the abscissa at AD=100%, it means that if there was no cap you would need more than -100% Attack Duration to have “instant” activation times. From there, we can get a decent estimation of AD’s effectiveness by plotting s versus ACT₀ where ACT₀ is the ACT of a skill without any AD buff and s is the slope of the same skills ACT(AD) plot:



The fact that this plot intercepts at (0;0) means that effectiveness is indeed the same for all skills with a given weapon. Then respective slopes allow to measure AD’s effectiveness (eff=x means that -1% Attack Duration reduce ACT by x%)

I have not tried every useful dps skill in the Champion’s toolkit, but it seems that most do not have a post-skill delay. At least it is the case for Brutal/Ferocious, Blade Wall and Wild Attack. This is a fairly big change with previous tests by various players, that is undoubtedly connected to the update they did a while back to “fluidify” skill usage.

As such, Attack Duration buffs really seem to not be worth a lot, ceiling at a whopping -8.2% or 8.4% ACT reduction (depending on whether AD caps at 39% or 40%).

I have not taken the time to make the same in-depth measurement with another race, but the few I did with a Man and a High-Elf with two-handed swords seemed to show the same ACT for Wild Attack without any AD buff. If anyone wants to double-check, be my guest!

[Gabli]

Measurements done with a (female elf) Hunter:

I’ve done pretty similar tests with my Hunter, although I tested more skills but only with one set of weapon types (bow for ranged attack, sword mainhand and dagger offhand for Low Cut).Since Quick Shot does not go on cooldown, its ACT can only be measured by sandwiching it between too skills that do (see previous post). Because you need to measure both a skill’s PSD and another’s ACT to get to Quick Shot’s ACT or GCD, the measurements are less precise and the data points’ alignment is understandably slightly worse. It is however consistent with a linear model: the points are fairly close to the regression when accounting for the uncertainties, and they seem to be fairly randomly distributed around that regression. For other skills, ACT data lines up pretty nicely before a cap at close to 40% (more on that in the next post):



As for champion skills, we can plot projections of the linear part of the ACT(AD) plots to assume all bow skills are evenly affected by Attack Duration. Only Quick Shot is marginally apart (we have already explained that QS’ ACT measurement were less precise), and very slightly apart for Heart Seeker.



Then we can evaluate AD’s efficiency same as before:



Concerning Merciful Shot, it is obviously a b*tch to test for so I only did a few tests with 0% and -25% Attack Duration buffs. It seems it has the same ACT as other focus spenders (PS, Blood Arrow, Exsanguinate, Split Shot, Upshot) and a small PSD somewhere around 0.013 s with 0% Attack Duration buffs, but no PDS with -25% AD.While Exsanguinate also seems to have about the same PSD than Merciful Shot at 0% AD but either none or only a few ms with -3% AD.It seems that among DPS skills, only Barbed Arrow and Heart Seeker have relevant and similar PSD, from 0 to cap Attack Duration. We can plot their GCD versus AD, and approximately verify that every -1% to Attack Duration does seem to reduce the GCD by 1% of its initial value, until the AD cap is reached. On the other hand, Quick Shot also has a PSD but that is fully negated with close to -10 or -11% Attack Duration, after this its GCD seem to closely follow its ACT:



The implication this has on Hunter gameplay is that to optimize rotation time, “Fast” skills (Upshot, Penetrating Shot, Blood Arrow, Exsanguinate, Rain of Arrows, Merciful Shot) should as often as possible be used after Barbed Arrow and Heart Seeker. Even at Attack Duration cap, it can save you about 0.18s every time so about every 4 times you use a fast skill rather than a standard one after HS or Barbed Arrow, you have gained enough time for an additional Quick Shot.

[Gabli]

Measurement of Attack Duration hard cap: (be warned, some skippable math in the beginning)

With each measurement of either ACT or PSD as a function of Attack Duration, we can measure the value at which Attack Duration caps as long as we have at least one or two data points after the cap, two obviously being preferable to make sure ACT or PSD seem to be constant rather than simply having one point off of the regression. Data initially forms a descending line before going constant: AD cap is the value at which both lines intersect.

Given that we have already obtain the slope and intercept of the descending line, if we note x the value at which AD caps, s and i the slope and intercept of the descending line, and a the ACT or PSD value after cap, we get:



Given that the slope and intercept with their uncertainties can be computed through Monte-Carlo simulations (see earlier) and that a is simply the average of data points after the cap occurs, we can get the percentage value at which Attack Duration seems to cap along with its uncertainty, for any skill tested with any weapon.

The first thing to test with a given weapon, is that AD cap measurement for different skills are consistent with each other, through z-score evaluation. Then, we just compute the average value for hard-cap. This was done with a dwarf Champion both with a two-handed axe and a two-handed sword, and with a female elf Hunter:





As can be seen, measurements indicate a roughly -39% Attack Duration cap for the dwarf Champion and a -40% Attack duration cap for the elf Hunter. If we try to compute the average and uncertainties for both and evaluate the z-score we get z=7, which means that both measurements are incompatible.
Possible explanations:

• My measurements are f*ed up.
• There is -1% Attack Duration buff that I missed somewhere on my Champion, which would mean I reach the same -40% cap than with my Hunter by applying an additional -39% bonus.
• The hard cap depends on the class and/or race.

I truly hope the last option can be ruled out. Given that with the exception of avoidances hard cap for stats are usually round numbers (25% critical chance, 200% Mastery, 40/50/60% Mitigations, …), I suppose the right value for Attack Duration cap is -40%. Which means that either my measurements with my Champion are botched, or that I missed a -1% buff, or that there is hidden -1% Attack Duration buff somewhere for Champions, or Dwarfs, or two-handed weapons.

[JustALittlePeril]

First of all, well done. I did a similar analysis on the champ and my results were similar. I did not catch either of the following though:

1. 2H swords are faster than 2H axes. Ug. now I have to decide if I really want my dwarven champ to wield a sword instead of an axe. Just seems wrong.
2. The 40% attack duration cap. I only measured at 10, 20 and 30% as HH was not yet out, so I missed that. It would be really interesting to see if that cap applies when external AD buffs are applied, like Anthem of Prowess, or if they are in a different category and still applied. If the 40% cap does still apply, it might make sense to ditch some of the points in Battle Reflexes, if you have the HH set bonus, to avoid being over capped when buffed.

One additional thing I did find, is that certain champ skills do not seem to benefit from attack duration buffs:
Fury of Blades
Horn of Gondor
Horn of Champions

(actually the horns saw a very small benefit going from 0% to 10%, but nothing after that.)

[Cordovan]

Wow, that is an impressive amount of work! Thank you for the analysis.

[Gabli]

First of all, well done. I did a similar analysis on the champ and my results were similar. I did not catch either of the following though:

1. 2H swords are faster than 2H axes. Ug. now I have to decide if I really want my dwarven champ to wield a sword instead of an axe. Just seems wrong.
2. The 40% attack duration cap. I only measured at 10, 20 and 30% as HH was not yet out, so I missed that. It would be really interesting to see if that cap applies when external AD buffs are applied, like Anthem of Prowess, or if they are in a different category and still applied. If the 40% cap does still apply, it might make sense to ditch some of the points in Battle Reflexes, if you have the HH set bonus, to avoid being over capped when buffed.

One additional thing I did find, is that certain champ skills do not seem to benefit from attack duration buffs:
Fury of Blades
Horn of Gondor
Horn of Champions

(actually the horns saw a very small benefit going from 0% to 10%, but nothing after that.)

Dwarfs with swords just feel wrong I'm with you .

I would be surprised if Attack Duration buffs from external sources, such as Routing Cry or Anthem of Prowess, behaved a different way. I was not able to test this as I don't have a second account to buff my character during the tests, and I would never ask someone to just stand there and buff me once every minute or so while I'm doing my tests.
In a discord server, someone made a list of buffs that showed you could stack AD up to something like -78% on a Champion if you have a Captain and a Minstrel to help.
If anyone wants to do this test the outcome would be interesting : we could first figure whether or not these external buffs stack additively with the rest or not, and if they do whether the supposed hard cap still holds for large AD buffs.

I had not tested the skills you listed, so I double checked Horn of Gondor and it looks like you are correct. To be more precise it looks like Attack Duration stops reduction HoG's ACT after 2% at most, which is the smallest increment I could test. That's the data I gathered :

Attack Duration (%)	ACT(s)	u(ACT) (s)
0	1.5104	0.0003
2	1.5028	0.0003
6	1.5021	0.0004
10	1.5010	0.0005
16	1.5024	0.0004
20	1.5027	0.0003
30	1.5024	0.0005
33	1.5024	0.0004
40	1.5022	0.0003
43	1.5025	0.0006

That's the corresponding plot :


I don't have anything more than speculations to explain this. One possibility would be that at some point, in order to have more "fluid" gameplay, devs could have tweaked stuff in a weird way like giving certain skills a specific AD buff which would mean you would reach the hard cap that much "sooner" in terms of additional buffs ? Had there been more than one data point before the constant portion, I could have used the slope to checked if its consistent with the ACT with 0% AD for a given weapon type... I'm out of ideas here.

[Gabli]

Wow, that is an impressive amount of work! Thank you for the analysis.

That's not really "work" to me though .

By the way, any chance to have some input from someone on the team on whether these items are working as intended or not ?

• Some weapon types are slower than others, for example axes are slower than swords
• Skill's execution time being affected slightly differently by Attack Duration buffs depending on the type of weapon being used.
• The existence of "hard cap" on Attack Duration somewhere around -40%.
• The fact that this cap seems to be different for different skills, and possibly also for different classes or races.

[JustALittlePeril]

I tend to agree that it is probably all one category of buffs that cap at 40%. If that turns our to be correct, heroic alacrity and battle reflexes start to seem like a waste in a raid environment. All you need is flurry, anthem of prowess and penetrating cry to hit 40%.

My brother and I have been known to run some tests like this. I'll post results here if we get around to it.

[JustALittlePeril]

It almost looks as if the 2H sword has a base 1% AD buff. Did you do 2H Axe tests at the cap? If the 2H Axe cap turns out to be 40%, it looks like a 1% AD buff for 2H swords might explain both the speed difference between 2H axes and 2H swords, as well as the odd 39% apparent cap on the 2H sword. The 2H sword and 2H axe should then be the same speed at the cap. My dwarf champ could sleep well at night knowing that his axe is just as fast as those silly swords when it matters most!

EDIT: Nevermind, I see now where you already verified that the axe caps at 39%, too.

[Vastin]

By a rather surprising coincidence, I was going over much the same issue a couple weeks ago, doing some game archaeology into the skill execution timing system - though in my case I was approaching it from the actual code and data side, in addition to a fair bit of testing. It looks like you have better testing tools than I do.

TL;DR your conclusions appear to be largely accurate.

Skills are defined by an Animation Duration, and a Skill Duration, and they generally use the longer of the two.
- Skill Duration can be modified by attack speed buffs by +/- 40%.
- Animation Duration can be modified by only +/-10% however.
- So if the duration are similar - or animation duration is longer, then the skill will respond very poorly to attack speed buffs.
- Fast attacks always use Animation Duration, so they never respond well to attack speed buffs, ironically.

The system was originally designed with the idea that Skill Duration would always be substantially longer than Animation Duration, but over the years that's simply often not the case any more. With many skills they are the same, or skill duration is a bit shorter than animation duration - so these skills can't adjust much.

There are also some flat out problems with how the math is being done in one or two places, though with the tight clamp ranges currently in place that particular bug is hardly noticeable - but it would be if I loosened the clamps without us fixing that first.

Weapon Type *shouldn't* have any appreciable effect on attack duration these days unless the weapon is giving you a specific buff to Attack Speed, but I'll double check that and make sure.

We do have plans to go in and dust the cobwebs off of all of this and get it modernized and working nicely again, though I don't have a timetable on that yet.

In any case, nice sleuthing!

[Gabli]

Thanks a lot for the answear.

I suppose it indeed fits, "animation duration" is what we called "ACT" and "skill duration" is what we called "GCD".

That would mean that the only things not WAI with the current systems are skills almost unaffected by Attack Duration at all (Horn of Gondor for example), weapon types having slightly different speeds, and the hard cap that seems to be different for different classes or skills.

[Vastin]

Thanks a lot for the answear.

I suppose it indeed fits, "animation duration" is what we called "ACT" and "skill duration" is what we called "GCD".

That would mean that the only things not WAI with the current systems are skills almost unaffected by Attack Duration at all (Horn of Gondor for example), weapon types having slightly different speeds, and the hard cap that seems to be different for different classes or skills.

Yep. Unfortunately the upshot is that from the player POV, the way skills respond to attack speed mods is quite arbitrary, which isn't great.

My hope is to do a pass that actually allows all skills to respond similarly to attack speed mods, ideally within a range of +/-50% (so 50-150% execution time).

Shouldn't be that difficult in basic implementation - the issue of course is that the balance ramifications are pretty dramatic, so it'd have to happen alongside a lot of testing, skill tweaks, and a re-examination of effects that actually grant skill speed mods, as it would become a much more significant factor in combat calculations of all sorts.

As an aside, inductions aren't affected by attack speed, and aren't supposed to be - though the windup/winddown animations on either side of the main induction timer can be, so inductions can see a slight improvement due to that.

-Vastin

[JustALittlePeril]

Yep. Unfortunately the upshot is that from the player POV, the way skills respond to attack speed mods is quite arbitrary, which isn't great.

My hope is to do a pass that actually allows all skills to respond similarly to attack speed mods, ideally within a range of +/-50% (so 50-150% execution time).

Shouldn't be that difficult in basic implementation - the issue of course is that the balance ramifications are pretty dramatic, so it'd have to happen alongside a lot of testing, skill tweaks, and a re-examination of effects that actually grant skill speed mods, as it would become a much more significant factor in combat calculations of all sorts.

As an aside, inductions aren't affected by attack speed, and aren't supposed to be - though the windup/winddown animations on either side of the main induction timer can be, so inductions can see a slight improvement due to that.

-Vastin

It would also be nice if the response to attack speed mods were as advertised, instead of ~22% of the advertised rate, as the testing above shows. If that means smaller advertised buffs, so be it.

[Gabli]

It would also be nice if the response to attack speed mods were as advertised, instead of ~22% of the advertised rate, as the testing above shows. If that means smaller advertised buffs, so be it.

I suppose it would be so after the pass. I don't think it's possible right now, as the "advertised" mods does apply for some skills but not all. The current Attack Duration buff/debuff system just does not fit the way skills work basically.

[LagunaD2]

What about attack duration *debuffs* of enemies?

Are these generally effective (skill duration >> animation duration) or not?

[Gabli]

What about attack duration *debuffs* of enemies?Are these generally effective (skill duration >> animation duration) or not?

I guess the first step would be to confirm that skills for monsters function the same way as freeps (animation duration vs skill duration). Only I think you could reliably test for this is making use of PvMP :
Step 1 : check whether skills have PSD.
Step 2 : check whether ACT & GCD are buffed/debuffed in a similar fashion.

I don't have any hard data for this. Usually in PvE instances, you see mobs/bosses going idle in between hits so I think that if these terms apply, they have larger skill duration than animation duration ? And as such Attack Duration debuffs are probably full value until -40% ?Still does not change the fact that as long as damage is so unevenly distributed among boss' skills, AD debuffs won't have a big impact on TPS for tanks as it does not change the tankbusters' cooldowns.

Edit : I just realise I've probably misread your post and you are talking about debuffs applied to freeps.
In that case, the most reasonable asumption is that the overall mod applied to Attack Duration is the sum of all buffs and debuffs, being capped between -40% and +40%, fully applied to "skill duration" but only about 20% of it applied to "animation duration".One implication that was discussed somewhere else, is that a freep with mostly "fast" skils sur as a Beorning will be pretty be unaffect by Attack Duration debuffs which was confirmed with the slugs puddles in the Sunken Labyrinth.

[Gabli]

As an aside, inductions aren't affected by attack speed, and aren't supposed to be - though the windup/winddown animations on either side of the main induction timer can be, so inductions can see a slight improvement due to that.

I had missed this bit. I was not aware the induction did not start until after a windup animation. Though if I look at the tests I've done for induction skills with respect to Attack Duration, both wind up and down animations seem to be affected by Attack Duration the same way than "animation duration", and both would have been concatenated in the "ACT" in the model I used.

According to your statement, those windup/down animations then should not currently be able to be dilated/contracted by more than 10% ? And total "skill time" would be the sum of induction (obviously unaffected by attack duration buffs) and both wind up/down animations ?

[Vastin]

I had missed this bit. I was not aware the induction did not start until after a windup animation. Though if I look at the tests I've done for induction skills with respect to Attack Duration, both wind up and down animations seem to be affected by Attack Duration the same way than "animation duration", and both would have been concatenated in the "ACT" in the model I used.

According to your statement, those windup/down animations then should not currently be able to be dilated/contracted by more than 10% ? And total "skill time" would be the sum of induction (obviously unaffected by attack duration buffs) and both wind up/down animations ?

I think that's likely correct. Not positive as I didn't spend much time testing inductions specifically - but I *believe* Skill Duration is generally irrelevant for them unless it somehow comes out longer than the entire induction cycle which is probably never the case, though I'd have to comb through to be sure about that.

- Vastin

[Vastin]

What about attack duration *debuffs* of enemies?

Are these generally effective (skill duration >> animation duration) or not?

Attack Speed Debuffs are calculated the same as buffs (it's just a +% rather than a -% and they can fight each other) but should be more effective on average because the system ends up taking the LONGER of skill or animation duration, and skill duration is usually at least close to if not longer than animation duration, so chances are good you'd end up seeing up to a full 40% slowdown due to attack speed debuffs.

Fast skills will be effectively capped at +10% debuff however, as they only care about animation duration, and changes to that are much more tightly capped currently.

- Vastin

[Gabli]

I think that's likely correct. Not positive as I didn't spend much time testing inductions specifically - but I *believe* Skill Duration is generally irrelevant for them unless it somehow comes out longer than the entire induction cycle which is probably never the case, though I'd have to comb through to be sure about that.

- Vastin

According to the tests I've done, skill duration does come out on top with a reasonable margin for both Heart Seeker and Barbed Arrow, even with -40% Attack Duration. For Quick Shot, "skill duration" becomes irrelevant after -10 or -11% Attack Duration, and it never is relevant for Swift Bow.
Those are the only induction skills I tested.


Edit : the skill duration being fully affected by Attack Duration is only consistent if "skill duration" does not include the induction window time.

[LagunaD2]

Attack Speed Debuffs are calculated the same as buffs (it's just a +% rather than a -% and they can fight each other) but should be more effective on average because the system ends up taking the LONGER of skill or animation duration, and skill duration is usually at least close to if not longer than animation duration, so chances are good you'd end up seeing up to a full 40% slowdown due to attack speed debuffs.

Fast skills will be effectively capped at +10% debuff however, as they only care about animation duration, and changes to that are much more tightly capped currently.

- Vastin

Thanks.

I'll keep using Sign of Power: Command then...