Extensive Desynthesis Rate Research

[Amele]

Been a long time folks; ただいま!

I'm posting this on behalf of a retail friend who I know by the name of "Gloin", who undertook to research Desynthesis rate information on current Retail.

tl;dr: Basically he reconfirmed the information that the Japanese wiki has: here - 60% failure, 40% success, given success: 40% NQ, 30% HQ1, 20% HQ2, 10% HQ3 (given HQ: 3:2:1 ratio); given failure 50% chance of losing the material.
And no tiering, obviously, although as you can see he looked for it to be sure.


Tier                1(+23)   2(+43)   3(+63)   4(+83)    Total
Fails                 2530     2438     2328     6382    13678
Material Loss         1267     1206     1146     3691     6845
NQ                     742      662      598     1689     3691
HQ1                    499      461      454     1305     2719
HQ2                    327      328      338      844     1837
HQ3                    169      197      164      398      928
Fail Rate           59.29%   59.67%   59.97%   60.11%   59.85%
Material Loss Rate  50.08%   49.57%   49.23%   50.55%   50.04%
Success Rate        40.71%   40.33%   40.03%   39.89%   40.15%
NQ Rate             42.72%   40.17%   38.48%   39.87%   40.23%
HQ1 Rate            28.73%   27.97%   29.21%   30.81%   29.63%
HQ2 Rate            18.83%   19.90%   21.75%   19.92%   20.02%
HQ3 Rate             9.73%   11.95%   10.55%    9.40%   10.11%


All desynthesis attempts were performed on "Leather Highboots" at the listed skill deltas, with no moghancement active (to ensure no aura effects.), no food active, and no synthesis support.

He also has begun investigation/reconfirmation of the effect of Moghancement: Lightning. (Overwhelming so far, but based on conventional wisdom we don't expect that the aura strength will matter.)


With 5730 failures, 2331 resulted in a lost material (still leather boots.) which gives a prevalence of 40.68% and a wilson binomial confidence interval (95%) of 39.42 to 41.96%.


With a baseline of 50% for leather boots material loss (determined from the initial testing above) we can conclude that the rate change is either 10% absolute or 20% relative, which is a higher magnitude than I think most of us believed it was.

I will post more information as it comes, but wanted to get the initial findings somewhere public instead of on a discord since we agreed it was worth sharing with the wider community.

If anyone else is interested in helping us further investigate material break rates, please feel free to DM me but don't expect an immediate response, I'm not very active in the FFXI community these days.

[Spicyryan]

Thank you for taking the time to test. This is important information.

[Amele]

Some more findings; the big surprise here was that "Moghancement: Desynthesis" appears to impact success chance by quite a bit.

Moghancement lighting, Powerful aura:

With 6215 failures, 2460 resulted in a lost material, which gives a prevalence of 39.58% and a wilson binomial confidence interval (95%) or 38.37 to 40.80%

So probably no effect from aura strength.

Moghancement Desynthesis, Powerful aura :

with 7388 attempts, 3210 successes, for a prevalance of 43.45% success (~4% higher than anticipated) with a wilson binomial CI (95%) of 42.32 to 44.58%. My Guess is that it's a relative 10% increase in success applied to the base success rate of 40%

Moghancement Desynthesis, Overwhelming aura:

with 6302 attempts, 2822 successes, for a prevalance of 44.78% success (~4% higher than anticipated) with a wilson binomial CI (95%) of 43.56 to 46.01%. the 44% value (10% increase of 40%) is in the CI for both, so aura strength probably doesn't matter here either

Comparing the two, both increase yield. Mog: Lightning increases yield more by about 2.5%, Mog: Desynth reduces crafting time for a given quantity of materials by about 12.5%.

[Byrthnoth]

So powerful vs overwhelming matters for desynthesis moghancement but not lightning?

Eh, probably not based on the confidence intervals

[Amele]

I could ask him for p-values to double check... but I think the differences were not significant. Null Hypothesis was that aura strength did not matter and we definitely didn't get a result that would enable us to reject that null.

I don't know if these were enough tests to distinguish between whether it's multiplying by 10% exactly or +/- some fractional X/1024, but that's probably only of interest to the folks working on emulation?

[Byrthnoth]

Yeah, not of any practical value (0.0006 different) and the sample size you would need would be stupid.

I imagine that wind, water, etc. do the same things for their respective desynth recipes.

Do things that reduce item loss (food, ring, etc.) or increase success rate affect desynths?

[Amele]

Sorry, Took a Break for the American Holiday and then we took a few days to get back to it.

Do things that reduce item loss (food, ring, etc.) or increase success rate affect desynths?

Coconut Rusk (-6% material loss, 3% HQ increase) + Moghancement lightning

Material Loss:

Losses: 2634, Attempts: 7546, Rate: 34.91% Wilson CI (95%) : 33.84%-35.99%

HQ Rate:

HQs: 3128, Successes (NQ+HQ) : 5148, Rate: 60.76%, Wilson CI (95%) : 59.42%-62.09%

The observed Material loss rate seems to be consistent with a function like: "(Base - Food {- Gear?} ) * (1 - moghancement)". where moghancement lightning is 0.2 (so base only 50% * 0.8 = 40% observed in the original trial)

It is not consistent with either a 16% (10+6%) absolute or 26% (20+6%) relative so other possibilities include a misprint on the item (seems improbable) or a cap at 30% reduction in material loss which is not consistent with the Sept 11, 2017 Patch note ( JP Note - NA note ) linked both since they say different things (weird).

The observed HQ Rate fits within a marginal +3% (or +3/128 (?)) multiplier on HQ rates. (i.e. we're getting .6*1.03, not .6+.03 ) - which I believe is consistent with the testing for success synth on coconut rusk? Craftymath v2 But it is also possible that HQ adjustment for Rusk does not impact desynthesis (that's also within the 95% CI)


Still need to test Gear, I think that's next. Trying to find someone with a Shield since we suspect that's going to be a necessary part of the puzzle for investigating the absolute cap alluded to in the patch notes....