How do the three elements affect curiosity multiplier?

[naosnule]

Some data:

Bloated Belote
10,10,10 => 1600 lp
10,28,28 => 2457 lp

Toy Chariot
10, 10, 10 => 1400 lp
26, 10, 10 => 1830 lp


Ladybug
10, 10, 10 => 400 lp
10, 10, 17 => 452 lp
10, 10, 14 => 429 lp

From seeing the data it is evident that it isn't as simple as the multiplier being sqrt ("your highest element"/10)

[Sevenless]

No it's probably something like cuberoot( E*S*V )

But I'm not enough of a mathboy to figure that one out without knowing some more hints about how the new quality into stat formulas work.

[MagicManICT]

From what I've seen posted and looked at myself, it's avg of 3 as the Q, and then the sq root of Q / 10 as usual. (Thought maybe they each had a factor on XP, time, and LP, but meh.. I was wrong).

[endiron]

No it's probably something like cuberoot( E*S*V )

But I'm not enough of a mathboy to figure that one out without knowing some more hints about how the new quality into stat formulas work.

Certain curios I've seen have qualities like (10, 12, 10) and (12, 10, 10) for instance, and they produce the same LP amount.
So each individual attribute provides an equal amount of extra LP as any other attribute, and that seems to match up with what your cube root formula, at least as far as I can tell.

[naosnule]

From what I've seen posted and looked at myself, it's avg of 3 as the Q, and then the sq root of Q / 10 as usual. (Thought maybe they each had a factor on XP, time, and LP, but meh.. I was wrong).

For the Bloated Belote case, that wouldn't work though:

sqrt ([28 + 28 + 10]/30)* 1600 = 2373.18 but the actual result is 2457.

[MagicManICT]

It's definitely not the cube root, either, unless something changed with the quality calculations. It would lower the Q even lower than averaging them would. A low value has a significantly greater impact on a geometric mean than an arithmetic mean, where as an abnormally high value has less impact in a geo mean than in arith means.

math I used sqrt ( (28 * 28 * 10) ^ (1/3) ) /10 ) * 1600 = 2255 approx

to reverse engineer it, assuming the quality scaling is the same--sqrt ( Q / 10 )-- you get approx q23.581441. If I do this: ( (10 + 28) / 2) + 28) / 2 = 23.5

edit: doing some follow up on this, and that math doesn't seem to work on the other listed curios...

[mobtown]

what would happen if you factor int or another stat into it?

[SovietUnion]

From my observations, it seems that the third quality parameter is more valued in the formula than the first two. No direct proof yet. I'm still trying to wrap my head around the whole quality concept.

Here's some data to confuse you all:

Still dream:
Presumed LP gain at 10/10/10 quality: 450

My 4 exhibits:
10/10/13 -> 474 LP Total: 33 Avg.q: 11
15/16/21 -> 596 LP Total: 52 Avg.q: 17.3
15/14/24 -> 608 LP Total: 53 Avg.q: 17.6
17/16/21 -> 606 LP Total: 54 Avg.q: 18

This example shows that the 3rd quality parameter seems to be more valued

Another curio supports this theory.

Boar tusk:
22/41/22 -> 1293 LP Total: 85 Avg.q: 28.3
53/31/43 -> 1560 LP Total: 127 Avg.q: 42.3
40/25/61 -> 1582 LP Total: 126 Avg.q: 42

This shows that it is no linear, square or cubic mean. At least from these 2 examples.

I really do wonder what out of this world formula the devs chose for calculating quality vs. LP. Will continue to work on this. Hope somebody finds this data useful

[SovietUnion]

what would happen if you factor int or another stat into it?

Nothing. My alt sees the same values as my main, so it is not stat / character dependent.

[majnao]

It's quite easy to figure out the effect of the single variables using the design of experiments (DoE). Another approach is to use simple multivariable regression. It just takes time, which I don't have right now