Valky had a Theory back in Beta Phase 3. His assumption, if he was correct, was that the only stat that scales differently between Auto-Attacks and Ability Attacks was Determination. Our original formula which was made by EasyModeX, as well as the Japanese Formula, were both constructed with this in mind. But, this rule does not fit my data samples what so ever.
Let’s look at the solver I’m using, and yes EMX, this is your original solver, albeit modified slightly.
This is the solver for Auto-Attacks. At the top, you can see all the variables used are exactly the same as my AC formula. The numbers highlighted in Orange are the parts of great significance. DET*E is the scaler which should change for Auto-Attacks, if Valkys assumtpion was correct. The Variance of Errors… Is the variance of Errors. The lower this number, the less errors the formula has. The Ability Formula has total error of over 100, which means it’s absolutely incorrect.
The Japanese assume that Determination scales twice as much with Auto-Attacks, than it does for Ability Attacks and valk has just plain stated “it’s different”. So let’s give it a go. If we solve this formula, only changing the Determination Scaler….
We get a Determination scaling value of 0.0008227, with an error reading of 18.6477. This isn’t accurate at all. So this got me wondering, what if this isn’t the case, and that Auto-Attacks literally have a completely different formula? So I went ahead and tested this again… And would you look at that…
All of the values completely changed. So this brings up a rather interesting topic – Do all of our stats scale differently with Auto-Attacks, or is it just either Strength and Weapon Damage that are different, in additional to Determination? I guess I’ll have to do some searching.
But for now, the current Auto-Attack model which fits my data is:
((AA_DMG*0.0.0388814+1)*(STR*0.0717908+3.5421066)*(DTR*0.0014237+1))*(AA_Delay/3)
Or, a rounded version of
((AA_DMG*0.0.039+1)*(STR*0.072+3.54)*(DTR*0.0014+1))*(AA_Delay/3)
Dervy does Theorycrafting 
I honestly find that formula confusion, and doubt it’s correct. A weapon has 3 stats: WD, AA_D and AA_Delay. These stats have the following relation:
WD/3 * AA_Delay = AA_D or AA_Delay/3 = AA_D/WD.
If you have AA_D * AA_Delay/3 in your formula, that can impossible be reached from the relation of weapon stats. Devs would’ve changed the stat formula sooner or later to fit eachother. (less overhead on calculations)
Valky used AA_D/WD while keeping WD in formula, later AA_D/WD was replaced by AA_Delay/3, which is the same value anyways. Technically it would also be possible to base all damage on Weapon_DPS (==WD/3) and multiply by AA_Delay.
As for DET having a different effect on AA than on the rest: Take the japanese tooltip: Autoattackにはさらに強く影響します。 Or the translation: It strongly affects auto-attacks (At no point at all is potency even mentioned). For physical/magical damage there is no mention of “strongly”, just “affects amount of damage dealt by physic/magic attacks)
On a different topic: Do you have any proof that STR and DET are linear from 0 to BaseValue to CurrentValue?
Taking all other Offensive/Defensive Properties: Determination would be the only outlier, while all 5 remaining properties would work in a similar fashion. (i.e start at their base value, not at 0). It’s relatively unlikely that DET behaves differently to it’s brothers/sisters.
Vitality to HP works as: BaseHP(class) + 14.5* (VIT-202) (people for no obvious reason changed that to VIT*14.5 – Constant(class) – because the notion of BaseHP was so terrifying? )
Piety works in a similar fashion to Vitality, just with a class-specific conversion.
Tendency is that with how many of the other stats are built as “Have some base effect at base value, followed by another effect afterwards”, i find it unlikely DET and STR/DEX/INT would be different.
If you tell your solver to solve for 660 STR and 500 DET, it will solve with these values.
If you tell him to solve woth (660-202) STR and (500-202) DET, it will solve with these values.
Just look how the formula changed when you decided to set DET and WD to +1. The solver solves with the constraints given by you, and that as long as it is possible. Since we’re dealing with multiplication only, by manually reducing one number, the solver will simple adjust other numbers – and as there are still a lot of numbers it can change, it will likely still be able to solve for them. (in your case it was a big increase for the STR terms)
LikeLike
I was actually awaiting your response to see what you have to say. I enjoy your input quite a lot.
I’m still working on how I should calculate Auto-Attacks. In all honesty, I might do what EMX did, and replace AA_Damage with Weapon Damage/3. Would just simplify everything a lot more for me. Not as accurate, but I’m not after complete accuracy. It’s impossible as we cannot get past out base stats at level 50 to do any intensive work at 0 DET/0 Strength. Everything is just an assumption sadly.
What I think you’ve noticed is that this isn’t “the damage formula”, but rather, a model which represents numbers we see in-game within certain constraints. So far, it reflects my data-samples of DET X STR and WD X STR (Need to double check DET X WD however), accurate from 0 WD, 202 DET and 271 Strength, all the way up to the highest values have (58wd, 665 str, 426 DET).
Here’s some minimal data-samples of WD X STR converging at 0 Damage @ 202 Determination btw.
http://puu.sh/hxD63/b45957a7dd.png.
If you have any advice on how I should handle all this data I’ve collected, I’d greatly appreciate it.
LikeLike
Could you give me the exact number where these lines cross eachother (or the x-axis – that seems also to be the point where they cross each other).
Or asking a bit more direct: How close is it to 22.2 ?
LikeLike
550 and 524 met up at 22.8ish, 500 was 22.2. I’ll need a few more data-points I’d say to accurately verify it.
LikeLike
I’m asking that because when i did my naked testing on my WHM (broken weapon, no stats allocated, cleric stance for 211 MND) and then removed all modifier i got a value of 22.11
If in the closest state to naked am looking at a value of 22.11 – it should be relatively easy to see where this value comes from – assuming the baseValue of our Stats begin at their naked value.
We’d need Data from other classes – either naked testing (and divide by potency multipliers, magick and mend type passives) or data gathering similar to yours – and see if all of them end up on a similar value. If that’s the case there a some conclusions you can draw from that (Stats do hardly anything below base value, Same or similar basevalue for classes resulting in STR conversion is class specific)
LikeLike
Yup. I was about to gather extensive Ninja, Warrior and Bard data samples to understand how each job affects damage differently. I assume you’ve already seen my post on Job Coefficients, so that’s what I need to check out.
I’ll post an update once I’ve gathered more data. Also if you have any WHM/SCH stuff, I’d appreciate it.
LikeLike