Yuki/Gekko/Kasha Attack bonus Testing

[Hakamaru]

I actually did a similar post on the KI forums regarding this subject. My argument was that Y/G/K ignores a certain amount of defense rather than an attack bonus. The math came out similar to yarkos. I got flamed for it tho.

yuki/kasha/gekko ignore defense trait? - Killing Ifrit

This? You didn't get flamed for it. People disagreed with your claim because it was too extreme. You claimed that Y/G/K ignored defense completely, and that's certainly not true.

Nice work Yarko. I'm glad someone finally went out and got their hands dirty.

[Valyana]

And Max was saying earlier that Steel Cyclone has the same bonus as Ground Strike. How about other weapons?

Shark Bite? Dragon Kick? Spiral Hell? Full Swing/Retribution?
Impulse Drive? How does this effect compare to the "ignores defense" property of Wheeling Thrust?

Calamity/Mistral Axe? Judgment? Blade: Ten? Savage Blade?

[Yarko]

And Max was saying earlier that Steel Cyclone has the same bonus as Ground Strike. How about other weapons?

Shark Bite? Dragon Kick? Spiral Hell? Full Swing/Retribution?
Impulse Drive? How does this effect compare to the "ignores defense" property of Wheeling Thrust?

Mistral Axe? Judgment? Blade: Ten? Savage Blade?

I can't help you with that because I don't have those jobs leveled, nor do I feel like spending more time WSing crabs!

Anyway, since I got some PMs regarding this and there is even a new thread in this section about Osode vs Haub+1, I'll repost the maths for "when attack is enough for Gekko" after the new findings.
Basically, attack is going to be capped when:

(ATK/DEF - dLV*0.05) * 2 > 2,2 - dLV*0.05

which means

Atk > (1,1 + dLV*0.025) * Mob's Def

where dLV is the difference between mob's level and sam's level (only if the mob is higher level of course).

Even though this formula is different from the one on Page 2, where I thought the bonus was actually applied to Ratio, the calculations come out to be the same. Thus I will just quote the post on page 2, but I'll also correct some of the values because I think I underestimated the Defense of HNMs (it seems to be in the ~550 range rather than ~450).

- merits/limbus/dynamis/einherjar/etc mobs are around LV 83 and have about 330 defense ~430 attack will put you to the cap;

- HNMs like Cerberus/Khimaira/Ixion/Fafnir/Gods are around LV87 and have about 500-550 defense 700-770 attack will put you to the cap;

- Wyrms/Kirin/JoL are around LV95 and have about 600 defense ~950 attack will be enough to hit the cap.

So yeah, this means that if you reach those attack values, you are better off using Osode and Smilo+1 for Gekko, if you don't, keep using Haub+1/Cerb(+1).

So guys, don't toss your Byrnie+1/Haub+1/Askar Bodies, you still want to use them if your attack isn't that high on HNMs.


As for the accuracy bonus discussion in the other thread, I don't remember if I already said this or not but I did a quick test about YGK acc bonus using a stack of blinding potions on Lesser Colibris with Soboro.
My Hit Rate was ~30% while my WS Acc was 80-85% (only 30 WSs performed, but enough to give a rough estimation) which seems to be a 50% hit rate bonus or an Accuracy+100 bonus.... imo, enough not to worry about acc in almost all situations.

[Enedin]

Very nice info Yarko! Doing the same math for SS is depressing lol, guess I'll stack attack ftw.

[Araelus]

Wise fwom your gwave!

*ahem*

There's a debate going on over in the gear+maths thread about Yarko's findings and what is going on with Gekko, so here we go again.

Based on this post:

Basically, attack is going to be capped when:
(ATK/DEF - dLV*0.05) * 2 > 2.2 - dLV*0.05
which means
Atk > (1.1 + dLV*0.025) * Mob's Def

where dLV is the difference between mob's level and sam's level (only if the mob is higher level of course).

and this post:

- merits/limbus/dynamis/einherjar/etc mobs are around LV 83 and have about 330 defense
- HNMs like Cerberus/Khimaira/Ixion/Fafnir/Gods are around LV 87 and have about 500-550 defense
- Wyrms/Kirin/JoL are around LV 95 and have about 600 defense

it seems that level 82, monster DEF/LVL is 3.99, but that ramps up to 6.33 at level 87 and then stays constant out to at least level 95.
Using the known levels and defense values and extrapolating other level ranges from the known monsters' DEF/LVL we get

http://img12.imageshack.us/img12/1075/defstuffcopy.png

The argument in the gear thread is whether this cap can be reached using half of the required attack, plus Gekko, or if the multiplier has diminished returns as dLVL rises. I made the mistake of claiming that Gekko ignores 100% of a monster's defense, instead of ignoring the first page of this thread, so we turn to:

Even though this formula is different from the one on Page 2, where I thought the bonus was actually applied to Ratio, the calculations come out to be the same. Thus I will just quote the post on page 2, but I'll also correct some of the values because I think I underestimated the Defense of HNMs (it seems to be in the ~550 range rather than ~450).

If the effect is the same, basically, as cRatio cap is factored into the required attack values to hit cap, then can we ignore that treating Gekko as doubling attack is wrong in principle so long as it is correct in practice, and just treat it as doubling attack? It's purely a pragmatic response to the game not displaying monster defense or player cRatio in the equipment window.

While I retract my previous statement that Gekko ignores 100% of a monster's defense, I stand by the results of that statement, so long as Yarko's math works out the same in either case.

My question, which he does not clearly answer, is when the WS multiplier is applied, but if the result is the same whether it is an att+, def-, or cRatio multiplier, then all we have to do is figure out how, or even if, the effect scales down as dLVL increases. Or is the Gekko 2*multiplier being included in the ATT-cap formula I quoted and used to build the chart above?

In which case the "required attack" is the value needed in addition to Gekko, not before Gekko is applied.

[shaddix]

Well I don't have sam so I can't go test for you, but surely it is fine to work with the idea that gekko simply doubles attack if the numbers end up right. Obviously curious people(such as myself) would keep chasing a more complete explanation.

edit, did anyone ever test my rana+oni idea? Merit with at least 1 minuet and meat, whenever zerk is up switch to oni and rana and see how it fares vs gekko and hagun.

[Araelus]

OK, I wanted to find a way to summarize all this data for ease of use. Reviewing tests and making a nice chart will be useful for accessibility purposes.

Definition of Variables

Spoiler: show

ATT = player's attack value
DEF = monster's defense value
pLV = player's level
mLV = monster's level
dLV = mLV - pLV when mLV > pLV
LC = "level correction" = dLV*0.05
Ratio = ATT/DEF
cRatio = Ratio - LC
wsM = "weapon skill multiplier"

Known Value Caps

Spoiler: show

cRatio caps at 2.2
wsM*cRatio also caps at 2.2
after this cap of 2.2 a level correction is applied

Known WS Multipliers

Spoiler: show

Tachi: Gekko        2.0
Tachi: Yukikaze    ~1.5
Tachi: Kasha       ~1.6-1.7
Spinning Slash      1.5
Ground Strike       1.6~1.75

Defense Values

Spoiler: show

We know Greater Colibri at level 82 have 327 defense.
We have approximate mLV and DEF values from Yarko of HNMs with lv87 550 DEF; lv95 600 DEF.

That means a normal monster of level 82 has 3.99 DEF/mLV, and HNMs of both level 87 and 95 have ~6.32 DEF/mLV.

The range between lv82 and lv87 can be considered the transition from normal monsters to HNM fare, since HNM have considerably higher defense than would otherwise be expected based on their level alone. But once we are in HNM territory, there is nothing to show DEF/mLV should increase any more than ~6.32 so this value will be used for all unknown level HNMs.

Calculations used for ATT cap

Spoiler: show

We know cRatio caps at 2.2 even after wsM is applied.
So when speaking of an "attack cap" what we mean is the ATT value which raises cRatio as high as possible. At the "attack cap" any more attack will not raise cRatio.

So first we take ATT/DEF = Ratio. Then Ratio is reduced by the level correction of dLV*0.05 to get cRatio. Then cRatio is multiplied by wsM, and capped back to 2.2 and then level corrected again.

This final value of (cRatio * wsM) thus has a maximum value of 2.2 - LC

(cRatio)*wsM > 2.2 - LC
(Ratio - LC)*wsM > 2.2 - LC
(Ratio - LC) > (2.2 - LC)/wsM
Ratio > [(2.2 - LC)/wsM] + LC
ATT/DEF > [(2.2 - LC)/wsM] + LC
ATT > {[(2.2 - LC)/wsM] + LC } * DEF

recall that LC = dLV * 0.05
which we kept shortened because the whole equation is rather messy
plugging in back in results in (eventually)

ATT > { (2.2/wsM) + [dLV*(0.05/wsM)] } * DEF

we can now see that Yarko's equation of ATT > (1.1 + dLV*0.025) * DEF
was designed for Gekko (wsM = 2) but not for other weaponskills
instead of being unsure of what situation it was describing

Let's take our equation, plug in the lower bounds of the three Tachi:Y/G/K multipliers, and see how they chart against monsters!

http://img21.imageshack.us/img21/1075/defstuffcopy.png

While ATT can be raised above 999, it's clear that on many HNM the only way to reach attack cap is by using Tachi: Gekko.

[Zoolander]

Bumping this because this isn't really a "done" conversation yet, and I mainly had one big question to ask. I see all this math under assumptions that the formulas and everything are correct, which can be difficult to replicate with 100% accuracy.

My question is, has anyone done any extensive parsing on different level mobs with a STR setup (name osode and smilodin+1 mantle) versus the tradtional attack and STR setup (namely hauby+1 and forager's/cerb). I'm just wondering what the real effects are of going with the osode + smilodon versus the hauby+1 + foragers on WS. You can do all the numbers in the world, but the only way to know for sure is to go out there and parse/test the two setups against eachother on merit mobs, HNMs, etc.

So yea, just wondering if anyone has done this or seen some parses of WS DMG comparing these two setups.

[VZX]

With the assumption pDIF = cRatio - LC

We can easily compare it to:

 ΔBD       ΔcRatio
----- vs ---------
  BD        Ratio

[Araelus]

The problem is that I didn't think pDIF is considered to be equal to cRatio - LC. pDIF is a random output within a certain range as currently defined by

Spoiler: show

http://images1.wikia.nocookie.net/ff...573px-PDIF.gif

and even this chart is inaccurate, because 2-handed weapons no longer cap cRatio at 2.

(cRatio - LC) with (cRatio = 2.2) simply gives us an upper bound on what pDIF(max) will be.

My concern is that all the data from Studio Gobli that you thankfully collated and analyzed is no longer accurate. Is the notion (pDIF = cRatio - LC) now accurate?
I'm interested in setting up and performing tests if we have the DEF/VIT values for different known level mobs.

[Yarko]

I would like to update the information in this thread for the people that were interested, as there are some things that I found out to be wrong in these last months:

- the fights against SW Serket led me to believe YGK multiplier takes effect after the LV correction, but later I found out that Serket has a slight (Slashing/Piercing only?) DMG-% which altered the data and the estimation of its defense; the right formula for WS cRatio is the original one:

cRatio = Ratio - Lv Correction

where Ratio = ATK * Multiplier / DEF and seems to be the one that actually caps at 2.3;

- Ratio cap is actually 2.3 and not 2.2. I'm not sure why I assumed it to be 2.2 when (1.85+2,75)/2 = 2.3, but yeah I went to hit some Colibris outside WG with a 2.25 Ratio and it wasn't capped;


- HNM's defense is not linear with their LV:
Khimaira/Cerberus are only LV85 but they seem to have 470-500 defense (Cerberus also gets some huge DMG-% at low health though);
Dark Ixion/Fafnir are LV88-90 but they seem to have much lower defense, 370-380;
JoL is LV83 but has a fuckton of defense, 780-820;
Tiamat is LV95 and only seems to have 350-370 defense.

- At this time it's unknown how Sneak Attack affects the cRatio equation for those WSs, but I guess it can be conjectured it's something like cRatio = Ratio - LvCorr + 1. Unknown where/how a cap applies though.

The equation to find out how much ATK caps Gekko is

ATK*2/DEF > 2,3

ATK > 2,3*DEF/2

I've seen a lot of "smilo and osode onry for gekko!" following these tests, but keep in mind that if ATK isn't capped, it helps Gekko a lot as it gets doubled. So don't toss those cerb+1s and byrnie+1s.

[Masamune]

I'm currently doing extensive parsing @ different attack values, but fixed str on Lesser colis, in the Ratio range of 1.05-2.09, for 1H and 2H weapons. These parses are just basic smacking, 0 WSs, no Berserk/Defender or any buffs. Didnot use any buffs as i suspect the game flooring at least once for display dmg values on log, which complicates greatly interpretation of parses, and buffs like Berserk/Defender would amplify greatly the error from that hidden floating point. For same reason, i didnot use Daggers or Polearm.

Parses have still missing suspected values, but shows already that Studio Gobli formulas are wrong when saying all pDIF min/max/critmin/critmax have same slope of 1.2. So far i get different slopes for each pDIFs linear going from 1.15 for min2H, up to 0.9 for Critmax2H.
They wrong also when showing a "floor" for CritMin in Ratio1.25-1.5: there s none.
On topic, Critmax2H caps @ Ratio=1.65; Critmin2H/Max2H/Min2H seems to caps around 2.25, from the current slopes i parsed since i couldnot get an attack high enough for this Ratio.

Concerning the caps, maybe it would help this topic to redirect to the tests i did a few months ago on ERonfaure bunnies: results posted @ wiki "pDIF" discussion page.

I'm curious also how you guys get HNMs' def values ? only "accurate" method is checking LowDef/HighDef with attack, can't be done on "Impossible to gauge" mobs o.O. Can't use pDIF formulas either since they are wrong, so which method did you use?

I'm also curious how you guys parse your damage on Steelshells, knowing those mobs have TPmoves that can fuck your parsing ? if by hand, i can guess as human testers are, samples won't exceed 500...

And finally, i'm curious at how much EXACT TP value each of your WSs have been done (for the test have a minimum accuracy)? (since even 1% TP affects the WSmod)

All those questions makes me very sceptical about the validity of any results in this topic :/ Though i /bow the effort done.

[Nameless]

Yarko's testing again confirms what I had believed all along that multiplier is applied to the ratio level, and ratio itself is capped at 2.2 or 2.3 (and you believe 2.3 is the answer, which actually fits even better with my data), and then clvl applies. Making it impossible to reach that 2.3 on mobs who are lvl higher than you.

We can use the current known formula to back track HNM's defense.

[Yarko]

I'm currently doing extensive parsing @ different attack values, but fixed str on Lesser colis, in the Ratio range of 1.05-2.09, for 1H and 2H weapons. These parses are just basic smacking, 0 WSs, no Berserk/Defender or any buffs. Didnot use any buffs as i suspect the game flooring at least once for display dmg values on log, which complicates greatly interpretation of parses, and buffs like Berserk/Defender would amplify greatly the error from that hidden floating point. For same reason, i didnot use Daggers or Polearm.

Parses have still missing suspected values, but shows already that Studio Gobli formulas are wrong when saying all pDIF min/max/critmin/critmax have same slope of 1.2. So far i get different slopes for each pDIFs linear going from 1.15 for min2H, up to 0.9 for Critmax2H.
They wrong also when showing a "floor" for CritMin in Ratio1.25-1.5: there s none.
On topic, Critmax2H caps @ Ratio=1.65; Critmin2H/Max2H/Min2H seems to caps around 2.25, from the current slopes i parsed since i couldnot get an attack high enough for this Ratio.

Concerning the caps, maybe it would help this topic to redirect to the tests i did a few months ago on ERonfaure bunnies: results posted @ wiki "pDIF" discussion page.

If you'll ever manage to find out the pDIF equation post right away cause that would be quite interesting to have.

I'm curious also how you guys get HNMs' def values ? only "accurate" method is checking LowDef/HighDef with attack, can't be done on "Impossible to gauge" mobs o.O. Can't use pDIF formulas either since they are wrong, so which method did you use?

I've never claimed them to be completely accurate. I used parses, known ATK and DMG, I estimated cRatio using the avg melee hit. This is assuming pDIF min and max have cRatio as mean, which should be true or very close in most part of the pDIF equation (except maybe for when pDIF min is 0).Sure this can easily be off by 20-30 defense, but not by 100. That was my goal and is good enough for me.

I'm also curious how you guys parse your damage on Steelshells, knowing those mobs have TPmoves that can fuck your parsing ? if by hand, i can guess as human testers are, samples won't exceed 500...

And finally, i'm curious at how much EXACT TP value each of your WSs have been done (for the test have a minimum accuracy)? (since even 1% TP affects the WSmod)

Yes I calculated the exact fTP every time, based on how many TP I had during WS (never more than 110, btw)
And yes, parsed by hand. Are you crazy, 500 tries? Each test has 30-40 data points at most, but have been replicated multiple times. Why would I need so many samples just to check if this model works? Let's make an example, last test I ran and of which I can still find the results (they're on paper) is Spinning Slash on Lesser Colibris.

Test on Colibris:
DMG 85
ATK 316
STR 74
DEF 231/235/241
fSTR 5/6/6

Lowest pDIF WS: 500 (it happened on max lv colibri) = 1,597 pDIF
Highest pDIF WS: 734 (again happened on max lv colibri) = 2,345 pDIF
Avg pDIF = (1,597+2,345)/2 = 1,971

This was after 33 WSs. I was unlucky and pDIF didn't fluctuate as much as I wanted, but:

In my sample pDIF only varies by 2,345 - 1,597 = 0,748. Assuming in this area of the pDIF equation it can vary as much as when it's at cap (2,75-1,85=0,9), the maximum error would be

Lowest pDIF with error = 2,345 - 0,9 = 1,445; Avg pDIF (Ratio) would then be (2,345+1,445)/2 = 1,895
Highest pDIF with error = 1,597 + 0,9 = 2,497; Avg pDIF (Ratio) (2,497+1,597)/2 = 2,047

This means the avg pDIF (Ratio) could actually vary from 1,895 to 2,047.

If my model is right, the expected Ratio on max lv colibri is

ATK * 1,5 / DEF = 316 * 1,5 / 241 = 1,966

which comes close to the avg pDIF I parsed, and is inside the max error limits.

Can I say Spinning Slash multiplier is exactly 1,5 with so few data points? no
Can I say it's between 1,45 and 1,56? yes
Do I care to run 500 samples just to say it's exactly 1,5 and not 1,52, 1,48 or 189/128? no

You "500 samples or this test means nothing!" people drive me crazy, use your head.

[Masamune]

the problem is only way to overcome that freaking randomizer is to get extreme super luck, or high sample size.

"Using yur head" like you say sure is convenient to get an approximation, which may sound ok for you and some other poeple to just know if it's between 1.45 and 1.56.
BUT
that approximation, when coupled with the pDIF and ratio truncation to 3rd digit + the fSTR flooring and finally the flooring just before the result display on log results, could also add Berserk or any buffs you used... in the end that's not a range of 1.45 - 1.56 but more like 1.4 - 1.6, which means almost +/- 10 dmg above/below real parsed dmg for normal hits. Personnally, i don't find that error neglectable when it gets multiplied during WS calculations.
So yea most poeple might be tempted to just "use their head" and stop there...

Regarding the parses i did so far problem is indeed sample size: i didnot get lucky enough to overcome that freaking randomizer after ~1k hits/300crits on each lvl of Lcolis to get ALL normal/crits min/max extreme dmg values. If you have a better idea to actually get an ACCURATE analyzis without being crazy parsing, im all listening for sure. So do you still think i should post my incomplete parses ?

This is not a real phenomenon we trying to modelize here, but a computer game, there is a way somewhere to get exact expected dmg values, just gotta overcome that randomizer + find a matching model, not an approximation. That's my opinion in answer to yours, no prob.

[Nameless]

This is specificly for Yarko:

I know this is probably related to pDIF, which we have a more current thread than this one. But this is more specified toward WS.

I know the theory of ratio having a 2.25 cap and ws multiplier is applied to ratio and then the whole thing is lvl corrected. I know that because the whole idea was based off my parse result on bird. However, I recently came across a thread on alla saying that Spinning slash might have only a 1.25 multiplier instead of 1.5, and that ratio doesn't have a cap. If that is the case, then it would explain the low number at bird, which is caused by not having enough attack to cap cratio. And it would also explain some of the parse result on Kirin on this thread http://www.bluegartr.com/forum/showthread.php?t=57830

I have to say that I believe those are SAM/THF with sneak attack since only two WS were fired by each SAM (the high and low averaged to be the average it is shown), and that the SAM/WAR on post 27 result looks like a perfect match with what our current theory is that against high lvl mob, you are just not going to hit that 2.25 cap.

Since I did not test SS's multiplier, I don't know if Yarko is 100% positive of it being 1.5. So I am bumping this thread to make sure that we are not using the wrong data for all these.

Thank you.

[Yarko]

Well look at the test posted 2 posts above, 1,25 wouldnt fit at all. Also I'm not sure at which parse and ws dmg being weird you're referring to in that thread you linked

[Masamune]

So I am bumping this thread to make sure that we are not using the wrong data for all these.

Indeed, that's an important and right doubt you got there in my humble opinion. And that change radically how you look at your own parses (how confident in their results i meant).

I wonder also how those WS multipliers can be tested "correctly", maybe using the floor @ pDIF=1 (1.25<ratio<1.5) using min known normal damage (at least that pDIF part are known 100% proven, not like the other curves...) ?

[Nameless]

Well look at the test posted 2 posts above, 1,25 wouldnt fit at all. Also I'm not sure at which parse and ws dmg being weird you're referring to in that thread you linked

The second post on that kirin parse thread. The first two players did pretty high WS that seems to suggest that if both weren't /thf then it would exceed the 1.4 cratio cap against kirin (17 lvls higher). Tho as I said, I believe both were /thf cuz only two WSs were used on each.

[Nightfyre]

fSTR would be uncapped in this case; 11 for the level 73 and 10 for 74 and 75. You used 360 defense there, so these were level 74?

I'll play around with this a bit and see what fits. Thanks for testing!