Open magey opened 3 years ago
Beanna pointed out some testing may be more difficult with the removal of weapon skill gear, let's check if fishing poles can be used as a proxy like in vanilla
Note: this research and comment were made before the glance chance and expertise bugs were fixed; I'm leaving it up for archival purposes. See updated results a couple posts below
A few observations from the beta as of today, with supporting data following it:
Player | Skill | Hit | Crit | Expertise | Target | Defense | Attacks |
---|---|---|---|---|---|---|---|
60 Orc Warrior | 300 | 0% | 17% | 5 (1.25%) | 63 Boss Dummy | 315 | 23834 |
Notes:
5% + (315-300)*0.2% = 8%
8.02% ±0.34%
5% + (315-300)*0.6% = 14%
12.53% + 1.25% = 13.78% ±0.42%
5% + (315-300)*0.1% = 6.5%
and measured 4.44% + 1.25% = 5.69% ±0.26%
. Initially we thought it might be using the player dodge formula instead (0.04% per point instead of 0.1% which would result in 5.6% dodge) but as you will see in the other testing setups this is not the casePlayer | Skill | Hit | Crit | Expertise | Target | Defense | Attacks |
---|---|---|---|---|---|---|---|
61 Orc Warrior | 300 | 0% | 17.24% | 5 (1.25%) | 70 Dummy | 350 | 7579 |
60 Orc Warrior | 300 | 0% | 25.34% | 5 (1.25%) | 70 Dummy | 350 | 8608 |
Player | Skill | Hit | Crit | Expertise | Target | Defense | Attacks |
---|---|---|---|---|---|---|---|
60 Human Warrior | 300 | 0% | 28.54% | 0 | 70 Dummy | 350 | 9213 |
60 Undead Warrior | 300 | 0% | 8.04% | 0 | 70 Dummy | 350 | 1749 |
Notes:
5% + (350-300)*0.2% = 15%
15.11% ±0.55%
and 15.71% ±0.68%
5% + (350-300)*0.6% = 35%
33.30% + 1.25% = 34.55 ±0.73%
and 34.60% ±0.89%
5% + (350-300)*0.1% = 10%
, measured 8.70% + 1.25% = 9.95% ±0.43%
and 10.33% ±0.57%
Player | Skill | Hit | Crit | Expertise | Target | Defense | Attacks |
---|---|---|---|---|---|---|---|
58 Human Warrior | 290 | 0% | 15.89% | 0 | 63 Boss Dummy | 315 | 19561 |
Notes:
5% + (315-290)*0.2% = 10%
10.00% ±0.42%
5% + (315-290)*0.6% = 20%
20.12% ±0.56%
5% + (315-290)*0.1% = 7.5%
but measured 6.47% ±0.34%
which is not close to either formulaRegarding glance chance being wrong; recently came across this extensive collection of WWS parses from original TBC where you can clearly see melee has a ~25% glancing blow chance against bosses.
After the glance chance and racial expertise bugs were fixed, we did another round of attack table tests. All tests were done with level 60 characters wielding a single Poniard with 300 daggers skill vs a level 63 boss dummy, with no hit or expertise from talents or gear, and no buffs that alter stats which can affect the collected data if they expire mid-session like crit or hit .
Player | Crit | Attacks |
---|---|---|
Orc Warrior | 18.35% | 71518 |
Orc Warrior | 30.09% | 9747 |
Human Rogue | 10.44% | 11451 |
Human Paladin | 8.98% | 17741 |
Orc Warrior | 31.59% | 8136 |
Summary | 18.06% | 118593 |
Outcome | Chance | 95% CI | Count |
---|---|---|---|
Miss | 7.97% | ±0.15% | 9449 |
Dodge | 5.93% | ±0.13% | 7036 |
Parry | 14.06% | ±0.20% | 16676 |
Glance | 23.81% | ±0.24% | 28237 |
Block | 4.96% | ±0.12% | 5883 |
Crit | 13.32% | ±0.19% | 15801 |
Hit | 29.94% | ±0.26% | 35511 |
As can be seen, the error margins are pretty tight thanks to the high sample count, and indeed most table outcomes are extremely close to their theoretical values. The major standouts here are dodge and glance chance, and these have been consistent across the different logs aggregated in the table above. The major standout here is dodge which has been consistent across the different logs aggregated in the table above.
Dodge being all over the place fits with our previous test results before these bug fixes; we had hoped that by fixing these bugs dodge would be corrected as well but that does not seem to be the case. Additionally, after the glance chance bug fix which switched to using the BC formula it does not converge to its theoretical value (25% in this case); it's always less, ending up around 23.5-23.8%, in all the logs we've analyzed.
After doing some equal-level tests to disprove #53, I ran into a curious result; hitting the level 60 dummy with my level 60 warrior resulted in 6% glance chance. This is way off the 'known' TBC numbers on old wowwiki entries and EJ guides - we were supposed to see 10% glance chance vs. a same level mob given by the formula on wowwiki:
The new formula seems to be: glancing blow chance = 10 + mob defense - player weapon skill, weapon skill being capped at 5 * level. For a level 73 mob, that's a 25% chance of glancing blow.
Digging in deeper into WotLK-era tests conducted on EJ I ended up in the famous 'Retesting hit table assumptions' thread which had this to say regarding glancing blow chance:
Moving on, I have the following stats from all swings performed on the boss dummy with 0 expertise:
29141 swings 7102 glances 1979 dodges
This is an observed glance rate of 24.37% and an observed dodge rate of 6.79%. The 95% confidence interval for glances is 23.88-24.86%, leading us to believe that the 24% glance rate we have been working with since 2.1 is probably still accurate. The 95% confidence interval for dodges is 6.50-7.08%. Even the 99% confidence interval is 6.41-7.17%, which doesn't overlap the 6.25% value that we've seen a lot lately. 6.50% still seems reasonable.
They mention the TBC glance chance post-2.1 to be 24%, which matches with what we got in our tests. With same level glance chance being 6% and +3 being 24% I had a theory that the formula is something similar to:
glance_chance = max(0, 6% + (defense_skill - weapon_skill) * 1.2%)
This made me initiate a series of tests with varying weapon skill values to verify the theory, which resulted in the following table (thanks to everyone from Fight Club who contributed logs!):
ΔSkill | Theorized | Measured | 95% CI |
---|---|---|---|
-5 | 0% | 0% | ±0.00% |
-4 | 1.2% | 1.10% | ±0.33% |
-3 | 2.4% | 2.43% | ±0.28% |
-2 | 3.6% | 3.55% | ±0.46% |
0 | 6% | 6.07% | ±0.21% |
5 | 12% | 12.12% | ±0.66% |
10 | 18% | 18.03% | ±0.41% |
15 | 24% | 23.81% | ±0.24% |
We're still collecting more samples to minimize the errors and to test more defense/skill points, but so far it would appear the formula is correct.
One other thing I ran into in the equal-level tests is that dodge was again way off; attacking a level 60 dummy with a level 60 warrior dual wielding with 300 dagger skill should result in 5% dodge chance; instead I got the following table (crit chance was 18.35%):
Outcome | Chance | 95% CI | Count |
---|---|---|---|
Miss | 23.80% | ±0.37% | 12017 |
Dodge | 4.04% | ±0.17% | 2038 |
Parry | 0.00% | ±0.00% | 0 |
Glance | 6.07% | ±0.21% | 3064 |
Block | 0.00% | ±0.00% | 0 |
Crit | 18.38% | ±0.34% | 9281 |
Hit | 47.72% | ±0.44% | 24101 |
Similar to previous tests we see that every outcome matches the theoretical value - except for dodge.
Gonna try to get more samples, but this was my initial findings for pets (mostly verifying the hit cap is 9% even for pets, and that everything else behaves the same as players)
I have a somewhat small sample size (comparatively) but these results are interesting...
Unfortunately it looks like Beta ate my homework, but I managed to upload to WCL for analysis before the log file disappeared. https://classic.warcraftlogs.com/reports/yY8ZBfcJ4Q6jwRPa#fight=2&source=1&type=damage-done and https://classic.warcraftlogs.com/reports/WZ6ab2pcLzdnPRFB#fight=1&source=4&type=damage-done
Pet Melee hits 70 vs 73 combined of the 2 logs, 14.99% Crit chance (4.99% spellbook, 10% talent), 4% hit talent:
Result | Chance | 95% CI | Count |
---|---|---|---|
Parry | 13.80% | ±0.61% | 1669 |
Dodge | 6.65% | ±0.44% | 804 |
Miss | 4.94% | ±0.39% | 597 |
Glance | 24.03% | ±0.76% | 2906 |
Crit | 9.93% | ±0.53% | 1201 |
Block | 4.89% | ±0.38% | 591 |
Hit | 35.75% | ±0.85% | 4323 |
Player Melee Hits 70 vs 73 combined of the 2 logs, 16.16% Crit chance (spellbook), 5.33% hit (84 hit rating):
Result | Chance | 95% CI | Count |
---|---|---|---|
Parry | 12.22% | ±0.76% | 879 |
Dodge | 5.30% | ±0.52% | 381 |
Miss | 3.35% | ±0.42% | 241 |
Glance | 23.96% | ±0.99% | 1723 |
Crit | 11.96% | ±0.75% | 860 |
Block | 4.56% | ±0.48% | 328 |
Hit | 38.65% | ±1.13% | 2779 |
Preliminary conclusions show that pets aren't plagued by the issues that we're seeing with dodge? It also appears pets do have crit suppression (4.8%) along with a 9% hit cap, 14% parry chance, 24% glance chance.
Will update with more data later when I get time, or if someone wants to contribute pet data themselves.
The linked extensive collection of WWS parses doesn't seem to support a 35% glance reduction: On average, I'd guess that "Avg Glance" / "Avg Hit" is around 0.75, for a 25% glance reduction.
Manually extracted data from those parses. For around 30k swings, glance reduction is very close to 25%.
The linked extensive collection of WWS parses doesn't seem to support a 35% glance reduction: On average, I'd guess that "Avg Glance" / "Avg Hit" is around 0.75, for a 25% glance reduction.
Manually extracted data from those parses. For around 30k swings, glance reduction is very close to 25%.
Very nice detective work! this surely merits a look by Blizzard. I wonder what it implies for lower level glance damage reduction (vs. +0/+1/+2)
The last TBC version of Vulajin's Roguecraft spreadsheet is using
Mob Level | %Chance | %Reduction |
---|---|---|
73/Boss | 24% | 25% |
72 | 16% | 25% |
71 | 8% | 25% |
70 | 0% | 25% |
I'd reason that only the boss level values were exhaustively tested.
Our finnish friends have used Ember tanks on Al'ar (e.g. "Dahako" here, or "Guntah" and "Dahako" here). Various sources agree that Embers of Al'ar are level 70 elite mobs, but the logs clearly show glancing blows against them. Average glance rate for six exclusive or near-exclusive Ember tanks is 56 / 983 = 5.7% glances/swings*. There's no tendency with respect to glance reduction other than that it's very low.
) for swings, I've used (#Hits + #Crits + #Glances) (1 + %Miss + %Resist), rounding up, so actual swings might be up to ~10% lower.
Various sources agree that Embers of Al'ar are level 70 elite mobs, but the logs clearly show glancing blows against them. Average glance rate for six exclusive or near-exclusive Ember tanks is 56 / 983 = 5.7% glances/swings*.
That seems to match with the testing we did on the beta WRT the glancing chance formula; 6% for equal-levels mobs is what we're expecting. Nice work!
We did another round of tests, this time 70 vs 73 (350 vs 365) using various classes with 0 hit% and 0 expertise totalling 105K attacks:
Outcome | Chance | 95% CI | Count |
---|---|---|---|
Miss | 7.95% | ±0.16% | 8379 |
Dodge | 6.48% | ±0.15% | 6826 |
Parry | 14.12% | ±0.21% | 14887 |
Glance | 24.22% | ±0.26% | 25528 |
Block | 5.06% | ±0.13% | 5337 |
Crit | 14.53% | ±0.21% | 15315 |
Hit | 27.64% | ±0.27% | 29133 |
This time dodge came out as expected. This leads me to suspect there's a specific problem with the dodge formula involving lower level mobs/players.
We did some tests for glancing blow damage reduction after the fix in #57 to determine the new formula used. After analyzing the results it seems that if the difference between the attacker weapon skill and target defense skill is less than or equal 10, it still uses the old Beaza formula for determining the damage reduction. The new formula only takes effect for skill differences 11 and up. Further analyzing the 11+ results let us derive the new formula. To summarize:
For ∆-Skill 10 and below, the old vanilla/classic Beaza formula:
low_end = max(0.01, min(1.3 - 0.05*skill_diff, 0.91))
high_end = max(0.2, min(1.2 - 0.03*skill_diff, 0.99))
For ∆-Skill 11 and above, the new TBC formula:
low_end = max(0.01, min(1.4 - 0.05*skill_diff, 0.91))
high_end = max(0.2, min(1.3 - 0.03*skill_diff, 0.99))
avg_reduction = (low_end + high_end) / 2
The test results:
∆-Skill | Hits | Theoretical | Measured |
---|---|---|---|
15 | 9403 | 25.00% | 25.19% |
14 | 7268 | 21.00% | 21.20% |
13 | 5234 | 17.00% | 16.89% |
12 | 5189 | 13.00% | 12.79% |
11 | 5248 | 9.00% | 9.01% |
10 | 14901 | 15.00% | 15.03% |
9 | 7302 | 11.00% | 11.11% |
8 | 4326 | 7.00% | 7.07% |
7 | 3933 | 5.00% | 5.00% |
6 | 5075 | 5.00% | 4.99% |
Now, the question which arises after looking at this table is whether this is how it behaved in original TBC or is it some kind of quirk of how this glance damage change was implemented in BCC?
Differences from classic
Testing emphasis