Prime WS Information, Testing, Discussion

Something feels weird about these WSs or SE stopped using nice numbers. I get Lorg Mor as a ~38% STR/MND mod (or 40% STR or MND, 35% MND or STR) and 4.25 at 1000TP scaling to 14.875 at 3000 TP.

You can use your gorget tests to determine that you have between 467 and 477 base damage. Stage 4 Lorg should be 260 with fSTR, so that's 207 to 217 WSC. The weapon gives +30 STR/MND, so it seems unlikely it would be uneven. Did SE give it 45% STR/MND mods and reintroduce the WSC level scaling factor (iirc 0.83 the last time I dealt with it)?

Anyway, using that base damage range you end up with something near the fTPs I proposed above.

Byrth,

Welcome back, I have been following your posts for years. Glad to have you helping with this.

One thing I notice is that everyone's approach is slightly different. We all get very similar answers but dialing in the final answer seems out of reach.

Also, I think we may be using slightly different inputs unknowingly. For example, in your WSC calculation for Maru kala you got 282 WSC with 40% STR/DEX. I actual use the equation WSC = floor (floor((A x A%) + (B x B%)) x α) which gives me 283! I don't know if there is a source that has figured out which is right but I bet there could be several discrepancies like this between all the contributors. If anyone knows if there was in detail testing on where the flooring happens in WSC calc please link it, otherwise I can try collecting some data.

Finally, it would probably really help to have a centralized spreadsheet that has a lot of these calculations and an approach that we all can agree on. Some of the JP "X" posts show good examples of how visualization can help. Probably google sheets is best since most people don't pay for excel. Then its just a matter of collecting the data. I'm not volunteering you for this... just thought I'd put it out there.

The biggest problem I have with it is that sharing google sheets shares identity. I don't want the top two hits when someone searches my e-mail to be linkedin and a video game forum spreadsheet.

Something like this technically works to hide identity, but loses formula access and thus is useless.

I think alpha doesn't exist anymore, right? Didn't they remove it in the WS revamp of 2014 when they declared that the Aeonic WSs cap out at 85% (instead of 100%)?

I don't really know where all the flooring steps happen anymore, particularly with PDL. I'm also not completely sure that our numbers for all those factors are precise. That's why I've been a bit wishy-washy and favored parsimony over a better (but still imperfect) match to the data.

Yes, the alpha should be gone. I have done a few calculations on older and newer ws recently and it all suggests alpha is still gone. I doubt it would be any different for prime. I am collecting data right now for the WSC calculation using shockwave and will post the findings which should also confirm it.

Yes, the alpha should be gone. I have done a few calculations on older and newer ws recently and it all suggests alpha is still gone. I doubt it would be any different for prime. I am collecting data right now for the WSC calculation using shockwave and will post the findings which should also confirm it.

I should have a more reliable sample for maru kala coming for you guys coming this week. I don't expect to finish it tonight, but I've managed to get the 1 and 2k tp both with and without fotia values to a 1.0494 or greater spread, and the 3k no fotia to a 1.0497 spread. I'm still working on the 3k tp with fotia. The 3k is the longest to do because of the time to build tp. I'll go back and do the str and dex values afterward and present my findings when I have a full data set. So keep an eye out for it. It'll probably be up within the next few days.

I'm actually extremely happy with how good the spreads are turning out to be. The 1k no tp is a full 1.05 spread so I think that set is actually perfect. Little bit more time and we'll have something.

Following up on the WSC calc confirmation I went and killed rabbits in Ronfaure as 70WAR/NIN with a "Greatsword". And I chose Shockwave for this test because the mods are 30% STR/MND and they were confirmed in this post: https://forum.square-enix.com/ffxi/threads/42588

Here are the inputs:
Gsword pDIF = 3.75 + 26/256(war pdl+ trait) = 3.851563
Wpn base = 55
fSTR capped = floor(55/9 + 8) = 14
STR = 105
MND = 62
fTP = 1

I took a look at melee swing first to make sure the basic values for pDIF etc were sound.
min = floor((55+14)*3.851563) = 265
max = floor(floor((55+14)*3.851563)*1.05) = 278
After around 80 swings my range was 265-278 so I know att/fSTR was capped and pDIF/pDL+ were accounted for correctly.

Now lets calc Shockwave with WSC = floor(0.3*105)+floor(0.3*62) = 49
min = floor(floor((55+14+49)*3.851563)*1) = 454
max = floor(floor(floor((55+14+49)*3.851563)*1)*1.05) = 476

And with WSC = floor(0.3*105 +0.3*62) = 50
min = floor(floor((55+14+50)*3.851563)*1) = 458
max = floor(floor(floor((55+14+50)*3.851563)*1)*1.05) = 480

It only took about 10 WS and my range was 459-480. So WSC = floor((A x A%) + (B x B%)).

Also, this was done without the alpha modifier to WSC and the math all seemed to workout so that's some solid data showing alpha wasn't brought back at any point.

Ok, so I've spent... a LOT more time gathering numbers for maru kala. I really hope this is enough information to nail down the specifics of the weaponskill. For the tp scaling test the weaponskill set I used is what I listed a few posts up in the thread; 5 pieces of mummu to ensure my fSTR is capped along with stage 5 varga. For the tp scaling test I performed a series of weaponskills exactly at 1 2 and 3k, noting the highest and lowest values for this. Also of note, for the tp tests done with fotia, as suggested by Byrth I have updated the data to include both fotia gorget AND fotia belt. I threw out the old 1k, 2k, and 3k TP data with fotia and started brand new for those three tests.

The spread I ended up getting isn't perfect, but it looks pretty good. The 3k with fotia test started taking an excessively long time and I wasn't getting anywhere further at that point so I moved onto the stat mod testing. I really don't want to have to keep redoing that one if it isn't necessary, so I'd like to submit my findings for review as they are.



So Simon, Byrth, or anyone else. Care to see if you can narrow the mods a bit with this updated information?

The coverage is amazing, but the results I get aren't parsimonious enough to pass the smell test for me. I think something is wrong with my damage formula or constants, still. Your results are incredibly precise at this point, so maybe it will give us some hint about where I'm wrong.

Using:
Base D = 331 = 3 + FLOOR(0.11*(482+277)) + FLOOR(8 + (3+216)/9) + 216
WSC = FLOOR(mod1 * stat1) + FLOOR(mod2 * stat2)
D = Base D + WSC
fTP = waypoint + Fotia Count * 25/256 (must be divisible by 256)
Prime AM 1k TP = 1 + 15/256 multiplier
Prime AM 2k TP = 1 + 23/256 multiplier
Prime AM 3k TP = 1 + 31/256 multiplier
MNK H2H pDIF cap = (3.5 + 77/256)*Prime AM
Secondary randomizer range = [1,1.05]

Damage = FLOOR(D*fTP*pDIF cap*Secondary randomizer)


Anyway, using the above equation, I get 41.75% STR and DEX mods with waypoints of 3 + 33/256, 7 + 70/256, 11 + 106/256. That just seems too unlikely to me, and the results don't consistently fall into your actual damage range so something about the flooring or math must be off. For example, your 1000 TP range is 7851 to 8244, but 7851*1.05 is 8243.5, which implies there is no flooring before the secondary randomizer.

Anyway, if you're looking to predict damage using the above formula then the mods and waypoint I listed above are probably good enough, but it leaves me dissatisfied from a reverse engineering perspective. Perhaps I'll get out COR and give it a look at some point!

You know, those numbers are shockingly close to what the japaneese test reports.


I think there's some rounding going on here somewhere that's throwing numbers off and making it hard to pinpoint the exact values. When I did my Sarv testing the results were just a little bit off too, and the reverse engineering didn't feel quite right either. I'd like to hear what simon has to say about this as well

From what you both say I think it would be safe to pinpoint the weaponskill at

40% STR/40% DEX
1000 - 3.0
2000 - 7.0
3000 - 11.0

If this isn't exactly correct, it's going to be very very close; extremely so to the point that for the wiki reference it should be good enough. Also, I want to add on to your thought line here


I have suspected for a very long time that our numbers on the wiki of all our weaponskills are "extremely close" but almost none, if any are "dead on". I remember a very long time ago now when S-E actually GAVE US the modifiers to a specific few weaponskills. They outright told us what the mods were. And the mods were... unusual. Very non-even. I believe it's a byproduct of how the flooring and fractional match works in this game. It's also possible that for some stupid reason at some point in time the devs decided to intentionally give very random numbers for mods for "balance reasons".

That s-e official information comes from this post right here

The March 2015 Version Update Notes



Just look at those stat mods. They're not even at all. They don't follow any rhyme or reason and the logic behind them feels more like "well this sounds cool" or "for BALANCE!" as the reason

10.19?? 6.09? 7.11?? 8.19?? 4.81?? 8.81?

What are these numbers? Are we just throwing darts at a board now and seeing where they stick? Because 10.19 is perfectly fine but 10.2 is completely overpowered right? Like... whyyyyyyy?!?!? But this is the official numbers that were handed to us by the people who actually have access to the code. So if those numbers are so wonky, then something similar is probably going on here. We try to make our logic even and neat, but s-e doesn't give a ***about that from the looks of it.

Byrth,

For WSC use WSC = FLOOR(mod1*stat1 + mod2*stat2)

Finally, try flooring here:
Damage = FLOOR(FLOOR(D*fTP*pDIF cap)*Secondary randomizer)

My money is on fTP being represented as x/2048 and they just left it at 2 digits after the decimal place.

0.9 = 185/2048
0.19 = 390/2048
0.79 = 1618/2048
0.81 = 1660/2048

Not that this answers the question of why they'd pick odd numbers. I'm guessing

Yeah, my criticism was about why they would make the math so jaggedy rather than using nice even numbers, not really how they would do it. I know everything is done through integer math via a system of bits that was friendly for the PS2 back in the day. Numbers are all in the form of XX/16, XX/128, XX/256, xx/512, XX/1024, XX/2048 etc etc. I know this, and I eluded to that being the reason that I think my sarv data also fit "very close" to what could be nice numbers but was odd in the same way these numbers are.

The point I was getting at is that on the wiki almost every weaponskill is laid out pretty neatly with documented numbers for TP scaling in even increments of 0.5, and stat mods scaling in 5% intervals. And I suspect that this is universally wrong on all counts, but only by small and insignificant margins. And for wiki purposes how much different is a stat mod of 44 versus 45%, or 41.5 versus 40%, or a tp scaling of 3.0, 7.0, 11.0 versus 3.06, 7.11, 11.17? Yes the wiki formula will vary slightly from actual game results, but the difference will be so minimal it wouldn't change anything. We'd still gear the same and the actual performance we see would be close enough to the projected values that the difference would be irrelevant. But the look and feel of the numbers being presented is so much cleaner the way the wiki currently looks. I much prefer the wiki reading the way it currently does on most pages for that reason.


Nail on head. That's s-e for ya.

Least dagger WS' were mostly accurate, ever since they put out the automaton WS numbers I often wonder if they're correct at all when they put out ftp/wsc

I agree we are close enough on values for practical use. But I would like to understand a bit more why we can’t nail it down.

One thought I have is we aren’t using the right aftermath value. If you look at the table that says 12% at 3k it’s rounded. Take the actual value 3k dmg and add .9 and divide and you don’t even get 12%. Ex: 1142.9/1021 =1.119 or if you did it the other way 1021x1.12=1143.5 which also doesn’t fit. so I am thinking maybe it’s much more likely 30/256 or maybe an out of 1024 value.

Honestly, I'm sure this would go nowhere but, it would be nice to ask the devs on the official forums after all these years. Maybe they'd be nice enough to just share the actual equations or at least list all the actual values. They did that for treasure hunter which they refused to do for centuries.

Obviously someone has to ask in Japanese though. I have a Japanese account, maybe it'll trick them into thinking it's a Japanese player

I mean. It's a waste of a question. Does it matter if its 41% or 47% or 43.28% more str more better.

If the mods are 40/40 or 41/41 or 44/44 or 47/47 it doesn't change the gear set

This was my reasoning for my advocating using the nice even numbers on the wiki. If the stat mods are 40% or 41% or 41.5 or some random in between 41.28% kinda number because s-e used a weird XX/1024 value that doesn't math out clean you're still gonna gear the same. What's BiS will still be BiS. You wouldn't swap out a regal ring for some other ring because of such a small percentage change, and the actual numbers that happen in game will be extremely close to the formula posted. It's just so much cleaner and easier to read this way.

Another example. The true value of the ftp bonus for fotia gorget and belt is 25/256, but for mathing purposes we generally just use 0.1. We know that 25/256 does not equal 0.1, it actually comes out to 0.9765xxxx, but who wants to throw THAT number into a field when doing calculations? The difference between a 48,572 calculated result being off from the actual 48,534 number you might see in game is pretty insignificant.

Exact values matter for sims :) Especially if you compere against different WSs or even different jobs.

Checking pdl would be probably easy done with something like Jokushunoibuki. It has 300 damage, so it would have a very good round-down precision and throwing doesn't have 5% randomizer, so you would only need to throw it once for each AM levels and once without AM. So you would only need 4 of them.

I know. I'm aware some of what I stated is just my personal preferences. Did you get a chance to look at the updated numbers I posted Simon? I'm curious to know what you would put the mods at with the new values.

I think it would be great if someone could do a verification of that table. The four main points: no aftermath, 1k, 2k, and 3k for stage 5. Kill outside a starter city, post the job, subjob, weapon/ammo? And of course don't use any other pdl+ gear. Even if it isn't with Jukoshunoibuki a confirmatory source would be nice.

Also, was there a post at any point that definitely tells us if pdl+% is implemented after or before the trait?

Dmg = Base*((pdif+pdltrait)*pdl%)
or
Dmg = Base*((pdif*pdl%)+pdltrait)

Even with only a tier II trait and tokko bow + beetle arrow you should be able to tell with 5%+ pdl gear on Ronfaure rabbits.

Dmg=(136+12+48)*((3.25+(51/256))*1.05)=702.6
vs
Dmg=(136+12+48)*((3.25*1.05)+(51/256))=700.7

I just came back to the game a few months ago so I am still super slowing working my way up to get some of this gear or I would do it myself.

Ight guys, I managed to get a 3%+ pdl DRK JSE neck super easy and so I did the 2nd test I suggested. Conclusion : pdl%+ is most likely an out of x/1024 value and is applied after the pdl+ trait. See below tests using DRK/NIN and a Phulax bow with various arrows on rabbits in Ronfaure.

Now we just need to figure out the exact prime values at each fTP tier.

chapouli arrow
trait only Dmg = (126+48+44)*(3.25+0.5) = 817.5
trait first 1.03 exact Dmg = (126+48+44)*((3.25+0.5)*1.03) = 842.025
trait first 263/256 Dmg = (126+48+44)*((3.25+0.5)*1.02734375) = 839.853515625
trait first 264/256 Dmg = (126+48+44)*((3.25+0.5)*1.03125) = 843.046875
trait first 1054/1024 Dmg = (126+48+44)*((3.25+0.5)*1.029296875) = 841.4501953125
trait first 1055/1024 Dmg = (126+48+44)*((3.25+0.5)*1.0302734375) = 842.24853515625
pdl% first 1.03 exact Dmg = (126+48+44)*((3.25*1.03)+0.5) = 838.755
pdl% first 263/256 Dmg = (126+48+44)*((3.25*1.02734375)+0.5) = 836.873046875
pdl% first 264/256 Dmg = (126+48+44)*((3.25*1.03125)+0.5) = 839.640625
pdl% first 1054/1024 Dmg = (126+48+44)*((3.25*1.029296875)+0.5) = 838.2568359375
pdl% first 1055/1024 Dmg = (126+48+44)*((3.25*1.0302734375)+0.5) = 838.94873046875
Actual was 817 and 841

beetle arrow
trait only Dmg = (126+12+44)*(3.25+0.5) = 682.5
trait first 1.03 exact Dmg = (126+12+44)*((3.25+0.5)*1.03) = 702.975
trait first 263/256 Dmg = (126+12+44)*((3.25+0.5)*1.02734375) = 701.162109375
trait first 264/256 Dmg = (126+12+44)*((3.25+0.5)*1.03125) = 703.828125
trait first 1054/1024 Dmg = (126+12+44)*((3.25+0.5)*1.029296875) = 702.4951171875
trait first 1055/1024 Dmg = (126+12+44)*((3.25+0.5)*1.0302734375) = 703.16162109375
pdl% first 1.03 exact Dmg = (126+12+44)*((3.25*1.03)+0.5) = 700.245
pdl% first 263/256 Dmg = (126+12+44)*((3.25*1.02734375)+0.5) = 698.673828125
pdl% first 264/256 Dmg = (126+12+44)*((3.25*1.03125)+0.5) = 700.984375
pdl% first 1054/1024 Dmg = (126+12+44)*((3.25*1.029296875)+0.5) = 699.8291015625
pdl% first 1055/1024 Dmg = (126+12+44)*((3.25*1.0302734375)+0.5) = 700.40673828125
Actual was 682 and 702

iron arrow
trait only Dmg = (126+14+44)*(3.25+0.5) = 690
trait first 1.03 exact Dmg = (126+14+44)*((3.25+0.5)*1.03) = 710.7
trait first 263/256 Dmg = (126+14+44)*((3.25+0.5)*1.02734375) = 708.8671875
trait first 264/256 Dmg = (126+14+44)*((3.25+0.5)*1.03125) = 711.5625
trait first 1054/1024 Dmg = (126+14+44)*((3.25+0.5)*1.029296875) = 710.21484375
trait first 1055/1024 Dmg = (126+14+44)*((3.25+0.5)*1.0302734375) = 710.888671875
pdl% first 1.03 exact Dmg = (126+14+44)*((3.25*1.03)+0.5) = 707.94
pdl% first 263/256 Dmg = (126+14+44)*((3.25*1.02734375)+0.5) = 706.3515625
pdl% first 264/256 Dmg = (126+14+44)*((3.25*1.03125)+0.5) = 708.6875
pdl% first 1054/1024 Dmg = (126+14+44)*((3.25*1.029296875)+0.5) = 707.51953125
pdl% first 1055/1024 Dmg = (126+14+44)*((3.25*1.0302734375)+0.5) = 708.103515625
Actual was 690 and 710

bone arrow
trait only Dmg = (126+9+44)*(3.25+0.5) = 671.25
trait first 1.03 exact Dmg = (126+9+44)*((3.25+0.5)*1.03) = 691.3875
trait first 263/256 Dmg = (126+9+44)*((3.25+0.5)*1.02734375) = 689.6044921875
trait first 264/256 Dmg = (126+9+44)*((3.25+0.5)*1.03125) = 692.2265625
trait first 1054/1024 Dmg = (126+9+44)*((3.25+0.5)*1.029296875) = 690.91552734375
trait first 1055/1024 Dmg = (126+9+44)*((3.25+0.5)*1.0302734375) = 691.571044921875
pdl% first 1.03 exact Dmg = (126+9+44)*((3.25*1.03)+0.5) = 688.7025
pdl% first 263/256 Dmg = (126+9+44)*((3.25*1.02734375)+0.5) = 687.1572265625
pdl% first 264/256 Dmg = (126+9+44)*((3.25*1.03125)+0.5) = 689.4296875
pdl% first 1054/1024 Dmg = (126+9+44)*((3.25*1.029296875)+0.5) = 688.29345703125
pdl% first 1055/1024 Dmg = (126+9+44)*((3.25*1.0302734375)+0.5) = 688.861572265625
Actual was 671 and 690

stone arrow
trait only Dmg = (126+5+44)*(3.25+0.5) = 656.25
trait first 1.03 exact Dmg = (126+5+44)*((3.25+0.5)*1.03) = 675.9375
trait first 263/256 Dmg = (126+5+44)*((3.25+0.5)*1.02734375) = 674.1943359375
trait first 264/256 Dmg = (126+5+44)*((3.25+0.5)*1.03125) = 676.7578125
trait first 1054/1024 Dmg = (126+5+44)*((3.25+0.5)*1.029296875) = 675.47607421875
trait first 1055/1024 Dmg = (126+5+44)*((3.25+0.5)*1.0302734375) = 676.116943359375
pdl% first 1.03 exact Dmg = (126+5+44)*((3.25*1.03)+0.5) = 673.3125
pdl% first 263/256 Dmg = (126+5+44)*((3.25*1.02734375)+0.5) = 671.8017578125
pdl% first 264/256 Dmg = (126+5+44)*((3.25*1.03125)+0.5) = 674.0234375
pdl% first 1054/1024 Dmg = (126+5+44)*((3.25*1.029296875)+0.5) = 672.91259765625
pdl% first 1055/1024 Dmg = (126+5+44)*((3.25*1.0302734375)+0.5) = 673.468017578125
Actual was 656 and 675

Virtually every multiple in this game uses /1024, likely cause that is what the original devs knew from coding in quasi ASM for older consoles like the PS2.

Updated but still haven't managed a full 5% spread in any of the data yet :(