magey
commented
2 years ago Testing on build 44930 it looks like avoidance DR has been implemented but now this mechanism has other critical bugs.
First let's cover what works correctly. If you only have defense rating from gear and no parry/dodge rating at all (so this can't be tested on a DK) then the character sheet total avoidance stats shown under the 'Defenses' tab and the values returned from APIs like GetParryChance() and GetDodgeChance() give the diminished values for dodge/parry, and the values are correct and match Whitetooth's formula. Indeed, this is how it worked in original WotLK and is how they collected all the data points back then to estimate the formula by using regression analysis.
Once you add parry/dodge rating into the mix things go horribly wrong, though. Let's take parry rating as an example, although dodge rating has the same problem. I ran a little experiment on my warrior that has 10% naked parry chance (5% base + 5% from Deflection); starting with a naked character, I incrementally equipped more and more gear with defense bonuses and documented the expected vs. in-game total parry chance. For the last 3 pieces of gear I equipped gear with parry rating on it as well. The experiment is summarized below, where the Total expected column is calculated as 10% + TotalBonusDR and Total bonus (DR) is calculated per the formula above:
| Rating | Rating Bonus | Defense | Total bonus | Total bonus (DR) | Total expected | Total actual | Base + BonusDR + RatingBonus |
|---|---|---|---|---|---|---|---|
| 0 | 0.00% | 400 | 0.00% | 0.00% | 10.00% | 10.00% | 10.00% |
| 0 | 0.00% | 411 | 0.44% | 0.46% | 10.46% | 10.46% | 10.46% |
| 0 | 0.00% | 421 | 0.84% | 0.86% | 10.86% | 10.86% | 10.86% |
| 0 | 0.00% | 437 | 1.48% | 1.50% | 11.50% | 11.50% | 11.50% |
| 0 | 0.00% | 445 | 1.80% | 1.81% | 11.81% | 11.81% | 11.81% |
| 0 | 0.00% | 451 | 2.04% | 2.04% | 12.04% | 12.04% | 12.04% |
| 0 | 0.00% | 458 | 2.32% | 2.31% | 12.31% | 12.31% | 12.31% |
| 0 | 0.00% | 470 | 2.80% | 2.76% | 12.76% | 12.76% | 12.76% |
| 0 | 0.00% | 479 | 3.16% | 3.09% | 13.09% | 13.09% | 13.09% |
| 0 | 0.00% | 484 | 3.36% | 3.27% | 13.27% | 13.27% | 13.27% |
| 0 | 0.00% | 491 | 3.64% | 3.52% | 13.52% | 13.52% | 13.52% |
| 0 | 0.00% | 497 | 3.88% | 3.74% | 13.74% | 13.74% | 13.74% |
| 0 | 0.00% | 501 | 4.04% | 3.88% | 13.88% | 13.88% | 13.88% |
| 40 | 0.88% | 509 | 5.24% | 4.91% | 14.91% | 15.80% | 15.79% |
| 96 | 2.12% | 519 | 6.88% | 6.24% | 16.24% | 18.36% | 18.36% |
| 145 | 3.20% | 525 | 8.20% | 7.25% | 17.25% | 20.46% | 20.45% |
As can be seen in the table, the numbers match up exactly until we start adding parry rating. Once we do that, the total value given in-game by the character sheet and the APIs seems to be 10% + TotalBonusDR + RatingBonus, meaning we are gaining parry from rating twice - once correctly using the total diminished value of total parry from defense and rating, and once again using the un-diminished value of parry from rating only.
I also performed a test for dodge rating and the results were similar.
CharWS
WotLK introduced diminishing returns on the parry, dodge, and miss avoidance stats. The specifics of how these diminishing returns work and what is the formula to calculate them was the subject of a big research effort on EJ by Whitetooth that's documented here.
He then gives a nice table that summarizes the $\large k$ , $\large C_p$ , $\large C_d$ , and $\large C_m$ values for all classes.
We[^contributors] went to test avoidance DR on the beta (build 44644) to see how accurate Whitetooth's formula and coefficients are, and to find any quirks in DR behavior. We used the exact same race / class / gear/ talent combination and attacked the same target on all the collected logs so that we can easily combine the data.
We used a level 70 Orc DK template and were attacked by a level 57 Servant of Razelikh. Due to the natural weapon skill / defense difference we had an additional
(70*5 - 57*5) * 0.04% = 2.6%to parry, dodge, and chance to be missed that is not diminished by the DR formula. Fully geared the character had:147 * 0.04% = 5.88%bonus parry, dodge, and chance to be missedThe breakdown and summary of these stats in table form (note that only the Bonus column is subject to diminishing return):
This is verified by the character sheet (note these screenshots naturally do not include the 2.60% from level difference):
Calculating the diminished values of these stats with Whitetooth's formula:[^precision]
$\large k = 0.956$ $\large C_m = 16$ $\large C_d = 88.129021$ $\large C_p = 47.003525$
$\Huge Miss_{DR} = \frac{1}{16} + \frac{0.956}{5.88\%} = \frac{1}{0.225085} = 4.442765\%$
$\Huge Dodge_{DR} = \frac{1}{88.129021} + \frac{0.956}{10.05\%} = \frac{1}{0.106471} = 9.392196\%$
$\Huge Parry_{DR} = \frac{1}{47.003525} + \frac{0.956}{9.92\%} = \frac{1}{0.117646} = 8.500079\%$
Using these diminished values in place of the raw bonus values from the table above, our expected avoidance values from the test are:
And finally our test results from combined logs of 17572 incoming attacks:
As this table demonstrates, avoidance DR simply isn't working at all; our measured values are within error margin of our raw total values and nowhere near the diminished values.
[^contributors]: Thanks to Char, deance, and Kilo for helping collect data for this test [^precision]: Rounded to 6 decimal digits for easy reading