Fast Cast and other haste stuff

[Kirschy]

Past while I've been trying to finish up our model of haste. The following is what I found. At the bottom I'll post some of the experimenting and methods I used to obtain these values and findings.

To start, here are the values of Fast Cast as they affect recast timers. (I make zero inferences as to how it affects casting speed.)

Fast Cast I: 50/1024 or 51/1024
Fast Cast II: 72/1024
Fast Cast III: 102/1024
RDM AF1 Hat: 51/1024
RDM AF1+1 Hat: 51/1024
RDM AF2 Body: 51/1024
Loq Earring: 10/1024
Homam Pants: 20/1024
R. Pumps: 9/1024
Marduk Body: 20/1024

When trying to determine the recast when combining fast cast and haste, it's important to know the order in which they are applied. Haste is applied first, and then truncated to the 10ths place. Fast Cast is then applied to the truncated number. The number is then floored(decimal truncated) for the final recast value. The equation would look something like this:

New Recast = floor( [1-Fast Cast] * ( [1-Haste] * Recast ) )
(Round to the 10ths place after each step.)

There are two caps to be aware of when determining a recast. The lower cap is 50% of the original Recast. The upper cap is 255 seconds.

I also tested Hasso/Seigan as they have a dominant effect on recast values. Hasso/Seigan add a 50% multiplier to the original Recast, and is applied before Haste or Fast Cast. With Hasso/Seigan active, the equation would look something like this:

New Recast = floor( [1-Fast Cast] * ( [1-Haste] * ( 1.5 * Recast ) ) )
(Round to the 10ths place after each step.)

The last piece of information I was interested in obtaining was where Slow occurs in the equation. The easiest way to test was by using Slow +% equipment. Unfortunately the math didn't pan out, and I was quick to realize that Slow Equipment appears to work differently than a Slow Effect from Magic/Ability. (Not 100% confirmed). Slow from equipment is subtracted from your total Haste value. It acts as a negative value of Haste. I was also suprised to learn that the values of of Hecatomb were slightly higher than one would assume. I tested three pieces:

Hecatomb Cap: 93/1024
Hecatomb Mittens: 52/1024
Hecatomb Leggings: 41/1024

When I originalli conducted haste testing, I found that you could predict recasts /w slow values from hecatomb by considering them negative haste values. Later when trying to determine the Slow from Demons in Dynamis-Xarc, I found that Haste and Demon Slow appeared to be seperate terms, and concluded a fault in my original tests. I now believe that Magic/Ability slow is a seperate term than slow from equipment. To test this, I ran out to Boyahda tree and started to kraken spiders in order to get slowed. Unfortunately the Spider Slow wasn't lasting long, and the slow inflicted varied. First Slow was between 330/1024-335/1024, and the second Slow was 183/1024-187/1024. If I'm able to find a way to get myself slowed by a consistant source, I'll try to finish the work I started. Otherwise I will wait til I find the time to ballista, and do my testing in one of those reserved pvp things. (Why do they require ballista points to reserve :/)


~~My methods and experiments~~
Honestly, I have about 8 notebook pages worth of stuffs, I'm gonna try to post just the important material, so that others interested in the testing can try this themselves, or continue the testing. I apologize in advance that some of this is hard to read or doesn't make sense. For example, I sometimes refer to 113/1024 and 911/1024 as the same value of haste. (This is because 113/1024 is how we read it, but 1-(113/1024) = 911/1024 and is how I use it when calculating.)

Majorty of the testing was done by using a BRD and varying the amount of the singing skill to raise the haste value in 1/1024 increments. Then casting Reraise or Dread Spikes and recording the recast values.

75 DRK/37 RDM: (Dread Spikes) (Fast Cast II)
/w no haste eq - 2:47 [This puts the value at 69-73/1024]
113/1024 Haste - 2:28
118/1024 Haste - 2:28
119/1024 Haste - 2:27

75 RDM/37 WHM: (Reraise) (Fast Cast III)
/w no haste eq - 54 seconds [86-102/1024]
75 RDM/37 NIN: Utsu: Ni (Fast Cast III)
/w no haste eq- 40 seconds [This narrows the range to 92->102/1024]
75 RDM/37 WHM: (Reraise) (Fast Cast III)
119/1024 Haste - 47 seconds
114/1024 Haste - 47 seconds
113/1024 Haste - 48 seconds

Using the above, the only way the math all works out is by having Haste applied first and rounding the 10ths place. I wish I had a better way to explain it. I proceeded to check and make sure this was correct by determining points at where the recast would change and verifying it. An example of such experiment:

75DRK/37RDM
Speed Belt + Victory March(535 Skill)
61/1024 + 118/1024 + 72/1024(FCII) = 1:18
Speed Belt + Victory March(542 Skill)
61/1024 + 119/1024 + 72/1024(FCII) = 1:17

Having 100% success after performing plenty of these experiments, I went to determine the fast cast on Hat/Body. The following can be used:

75RDM/WHM using Reraise
/w Fastcast Piece - 51 seconds [871-887]
113/1024 Haste - 45 seconds [Narrows Range down to 871-865]
119/1024 Haste - 45 seconds [Narrows Range down to 871-870]
240/1024 Haste* - 39 seconds [Narrows down to 871]
1024-871 = 153/1024 total fast cast... 102 from FC III means 51/1024 from Fastcast Piece.
(*240/1024 Haste = 119/1024 Victory March + 61/1024 Speed Belt + Dusk Feet+1 + Dusk Hands.)

I apologize all of this is messy =x. That experiment works for AF1+1 Head/AF1 Head, and AF2 Body.

Homam Legs can be deduced by the follow. (I had to assume 30/1024 on the haste portion.)

75DRK/--- (Dread Spikes)
/w just pants - :51 [Range of 21-16/1024]
113/1024 - :31 [Narrows to 19-21]
118/1024 - :30 [Narrows to 20-21]
117/1024 - :31 [Narrows to 20/1024]

Most of these are expected values. I don't think it's anything new. I think the last important piece to post is proof that Hasso/Seigan is applied before Haste. I had to find values of haste which would cause a difference if one was applied first as opposed to the other way around. Here are two cases:

@880/1024 Haste
1024-880 = 144/1024 - 114 March + 30/1024(3% Eq Piece)
880/1024 * 180 = 154.6 * 1.5 = 231 seconds
1.5 * 180 = 270 * 880/1024 = 232 seconds

The result came out to be 232 seconds, implying Hasso/Seigan was applied first.

@861/1024 Haste
1024-861 = 163/1024 - 113 March + 30/1024(3% Eq) + 20/1024(2% Eq)
861/1024 * 180 = 151.3 * 1.5 = 226 seconds
1.5 * 180 = 270 * 861/1024 = 227 seconds

The result came out to be 227, giving further proof that Hasso/Seigan was applied first.



To put all of this work together, I mixed and matched equipment and attempted to predict the exact recast value.

113 Victory March + Advancing March + Ace's Helm + Homam Legs + Heca Hands + Heca Feet + Hasso + Loq Earring.

Hasso = Yes~ 1.5
Haste = (113/1024 + 96/1024 + 40/1024 + 30/1024 - 52/1024 - 41/1024) = 186/1024
Fast Cast = (20/1024 + 10/1024) = 30/1024

Recast = floor( 1 - [30/1024] * ( 1-[186/1024] * (1.5 * 180) ) )
(Rounding to the 10ths after each step)
Recast = 3:34

The recast does indeed turn out to be 3:34.

EDIT: Added Fast Cast for R. Pumps and Marduk Body~

[_chuchu]

Great info, Kirschy!

[cdgreg]

nice job

[Tsuki]

Excellent work, can we get an r.pumps/warlock mantle value? ^^

[Phalanx]

nice job

[Tubbers]

Can we add a Kirschy smiley already?

[Eraphiel]

Excellent work, can we get an r.pumps/warlock mantle value? ^^

I brought this up as she was testing with things that have fast cast on them...the only problem is that the mantles are annoyingly rare on Fenrir, so testing it may prove difficult. .__.;

[Asmoranomar]

I'm a 75 BRD with r.pumps/warlock mantle (37RDM sub). I'm familiar with fraps and virtual dub if any recording is needed. While I'm personally not too concerned with haste values, I have done extensive testing on Fast Cast and -Song Spellcasting Time (in regards to casting times). (speaking of - anyone have links to casting time testing?)

If I can be any help, let me know and I'll see what I can do.

[Kirschy]

There was some testing /w Cure times and Light Arts/Fast Cast/Cure Merits/etc... But I believe it was found that were tiers. This is similar to what I found when I tested attack delay from mobs under varying amounts of slow.

What it comes down to is that either the game doesn't graphicalli represent intricate intervals when displaying animations. Or the game actualli does round/truncate when figuring out casting speed/attack delay values. Mostly for this reason, I've been trying to steer away from this kind of testing. I'm more interested in calculating exact and precise values.

Like Era said, finding a wizard's mantle is tough on Fenrir. I was lucky enough to have Era let me borrow his RDM AF2 body for testing =x. Finding the mantle will be a bit more difficult (not to mention designing the tests.). I'll get around to doing Rostrum Pumps soon~

[Kirschy]

Thanks to Rama and his testing, we've been able to ascertian the fast cast on R. Pumps and Marduk Body~

R. Pumps:
BRD/SAM using Mazurka(60s recast)
110/1024(March) + R. Pumps = 53 [0-9/1024]
111/1024(March) + R. Pumps = 52 [8-26/1024]
50/1024(Turban) + 112/1024(March) + Hasso + R. Pumps = 75s [9 or Higher]
50/1024(Turban) + 113/1024(March) + Hasso + R. Pumps = 74s [9 or Lower]

The only value of fast cast that works for all four of these recasts is 9/1024.

Marduk Body:
BRD/SAM using Mazurka(60s recast)
61/1024(V. Belt) + 109/1024(March) + M. Body = 49s [1-20/1024]
61/1024(V. Belt) + 110/1024(March) + M. Body = 48s [19-38/1024]
50/1024(Turban) + 114/1024(March) + M. Body + Hasso = 74s [20 or Higher]
50/1024(Turban) + 115/1024(March) + M. Body + Hasso = 73s [20 or Lower]

The only value of fast cast that works for all four of these recasts is 20/1024. This makes the Fast Cast value 2%.

Thanks again to Rama for performing these tests. It's nice to finalli nail down some values on these items.

[Valyana]

I find it rather surprising that Pumps and Loq. Earring would have different values.

[Suiram]

Regarding spider slow, is it possible that the potency varies with the amount of TP they have? Maybe try stunning WSs to 300% with a spider of constant level and see if it always slows the same amount?

Also another mob that slows in Diremites, and I know that Filimented Hold as a BLU spell is supposedly a fixed potency, so if you can't get consistent results from spiders it might be worth checking those to see if the slow from the WS is fixed, too.

[Ophannus]

Euhvi's slow seems extraordinarily potent. I was soloing one on my DRG and it hit me with it(forgot the name of the TP move) but I didnt swing for like 20 seconds.

[Spider-Dan]

Euhvi's slow seems extraordinarily potent. I was soloing one on my DRG and it hit me with it(forgot the name of the TP move) but I didnt swing for like 20 seconds.

I thought max attack slow was +100%? (i.e. it takes twice as long as normal to swing)

[Wizerd]

I thought max attack slow was +100%? (i.e. it takes twice as long as normal to swing)

To be fair, 16.4 seconds (double normal polearm delay?) probably seems like 20 seconds when it takes that long to swing. <_<

[Asmoranomar]

Kirschy -

May I have a re-cap of what it is you specifically do in performing these tests? How are you acquiring the recast times (Via interface? Recoring frames? Memory? etc?). I'm not testing haste values, but I'd like to have something to baseline my work against to ensure its accuracy. Unfortunately, I'm not comming up with anything remotely similar to your tests.

Edit: To clarify what I'm asking, I understand the purpose of using Hasso/Seigan, Dread Spikes and the stepping of Marches. What I'm really interested in is how your taking your measurements and what your doing to account for inconsistancies.

[Eraphiel]

What she had me watch for when testing the Duelist's Tabard and Warlock's Chapeau vs Warlock's +1 was the recast from the spell menu.
Stepping up the haste from march using the varying skill was the most controlled way to make minute changes to narrow down the possible values. Further combinations of haste gear were then added to test an expected value she derived from the formulas, and see if they matched up with the expectations.

As far as attack animations, to my understanding (as I've asked her this and volunteered to test things like Spirit Surge haste insofar as weapon delay/etc.), Fraps is used to record a set period for attacking, then watching frame-by-frame. Since the game runs at a frame rate that's similar to or at least would match up with fraps, she'd then be able to derive an increase/decrease in speed, although it's not as reliable a method as for spellcasts.

[Tricen]

If you have to watch every frame to see if there is a haste bonus, I don't think it'd be worth it, or would matter at all <_<?

[Eraphiel]

I don't remember Kirs doing many animation tests outside of caps for max slow (dw stage1 relic sword/dagger on rdm/nin), or testing slow2's potency (staring at small bees attacking and looking for a slowdown....I commend her patience). So yeah...testing haste as far as it affects recast timers is far easier, and generally where more questions come in as far as being able to optimize and shave another 1s more off a needed spell with minimal gear.

Melee-wise...more = better as far as haste, the testing was mostly to nail down certain quantities and limits. How much MND does it take to cap certain levels of slow1/2, curiosities like the max slow on melee speed, etc.

Given how much of a pain in the ass she's said it is to do...I'm not sure how much time I'd be willing to devote out of the sake of curiosity.

[Asmoranomar]

Okay, that helps, but my next question is -

How do you account for inconsistancies? For instance - When I have the same setup with hasso/seigan + dread spikes, I've run into cases where the recast 'flip-flops' between two different recast values. Do you just take the lowest recast value?

Also, I found my hangup on my testing. It's not that my tests were wrong, we were just calculating the same value from two different limits (ie: XXX.99999 and XXX.0000 floored is still XXX) and in the specific case i was doing - both were true (Kirschy's being the exact value while mine being approx). It wasn't until I went back to hasso+dread that I was able to see this, and subsequently smashed my face into my desk.