A Call to Arms - Cataclysm Mechanics testing

[theckhd]

Assuming the inferences we made with the Dwarf data are correct, here's the revised output from the Human model:

Code: Select all

       Model   Ras   Diff  %Err
ShoR    1225  1231      6     0
CS      1483  1476     -7    -0
JoT      520   520      0     0
AS      3615  3605    -10    -0
HW      2600  2601      1     0
HoW     1522     0  -1522  -100
Exor    1354  1354      0     0
SoT      203   204      1     0
SoR       54    57      3     6
SoJ       25    24     -1    -4
Cens     835   838      3     0
Cons    1645  1645      0     0
HotR     232   231     -1    -0
HaNova   847   842     -5    -1
Melee    772   776      4     1


I need to see if the differences in SoR and SoJ are significant, but that's a low priority. And we need some data on HoW to check the model there. I'd also like another, independent confirmation from another tester on another pre-made, possibly with a slight variation on the gear set just for a sanity check.

Other than that, the next stop will be to get level 80 data to iron out the base damages of abilities at level 80.

[Kierly]

I'm just finishing up a log of attacks made using the 0/38/0 spec with no glyphs. The only gear I'm using is the premade holy sword and shield which should help with testing ap and sp coefficient differences. The character was human for the tests.

Item Stats: (in case you don't have them)

Code: Select all

Weapon: Blade of the Burning Sun
172sta
115int
41.7dps
1.6speed
1-132dmg
crit rating 82
haste 67
sp 1532

Shield:  Zora's Ward
11504 arm
224 sta
149 int
haste 107
mastery 87


Here is the log:
http://www.worldoflogs.com/reports/j7z92d68km8cgkxs/

The overall avg for SoTR might be slightly off since I think one of them was only 2 Holy Power.

[Neziah]

Pre-made L85 Human Default
http://www.worldoflogs.com/reports/rt-0 ... etails/30/

Image with stats
http://img.photobucket.com/albums/v655/ ... 223800.jpg
Image with Build for reference
http://img.photobucket.com/albums/v655/ ... 224920.jpg

[theckhd]

I'm just finishing up a log of attacks made using the 0/38/0 spec with no glyphs. The only gear I'm using is the premade holy sword and shield which should help with testing ap and sp coefficient differences. The character was human for the tests.

Item Stats: (in case you don't have them)

Code: Select all

Weapon: Blade of the Burning Sun
172sta
115int
41.7dps
1.6speed
1-132dmg
crit rating 82
haste 67
sp 1532


Are you sure about the damage on that? Sigrie has outdated and ridiculous numbers (409.7 dps), but 1-132 seems like a very low damage range. Also, could you post your AP and SP in that setup (I get 563 AP and 1990 SP).

This is what the data gives me:

Code: Select all

ShoR     676   627    -49    -7
CS       171   215     44    26
JoT     1242  1140   -102    -8
AS      3787  3764    -23    -1
HW      3098  1612  -1486   -48
HoW     1702     0  -1702  -100
Exor    1500  1527     27     2
SoT       23    23     -1    -2
SoR       90    95      5     6
SoJ       41    41      0     0
Cens    1118  1080    -38    -3
Cons    2100  2100      0     0
HotR      27    26     -1    -4
HaNova  1441  1452     11     1
Melee     89    90      1     2

Everything seems fairly reasonable. ShoR is low because of the 2-pointer. I'd guess that CS is probably off due to the base damage range being wrong, but HotR's physical component seems correct. Holy Wrath data will be unreliable since you were hitting multiple dummies at a time (it looks like 2, hence why it's off by 50%). I'm not sure what happened with JoT damage, but there are two separate entries for it in the log ("Judgement of Truth" and "Judgement"), but they're both low.

However, this gives us reasonably good confirmation of Consecration, Censure, and HammerNova damage, so that's good.

I'm beginning to suspect that the SoR coefficients are a tiny bit off, if we get another data set or two with 6-8% error I'll take a closer look at it.

[theckhd]

Pre-made L85 Human Default
http://www.worldoflogs.com/reports/rt-0 ... etails/30/

Image with stats
http://img.photobucket.com/albums/v655/ ... 223800.jpg
Image with Build for reference
http://img.photobucket.com/albums/v655/ ... 224920.jpg

I'm not sure I'll be able to make out much from this log. There's clearly other effects going on. As an example, your Consecration damage jumps around from 385 to 600 and several values in between, presumably due to other players' debuffs on the dummy. That screws up a lot of the values.

At the very least CS, SoT, Censure ticks, and Consecration ticks all seem to be right, confirming that Crusade works and the haste scaling of Censure with JotP.

[Kierly]

Yeah those are the values listed on the tooltip for the sword. With the sword equipped my dmg on the char sheet is listed as 65-197 with 83 dps.

ap is at 563, sp is at 1990

Yeah Holy wrath was hitting 3 targets most of the time. If you need some with holy wrath hitting just one I can do more testing later today.

I'm not sure what happened with JoT damage, but there are two separate entries for it in the log ("Judgement of Truth" and "Judgement"), but they're both low.

Judgement is when I had insight on to get some more mana back. And My guess on the JoT is that not all judgements had 5 stacks. I tried to keep a 5 stack up for it but some of the initial ones might not have had the full 5 stack.

[Neziah]

I'm not sure I'll be able to make out much from this log. There's clearly other effects going on. As an example, your Consecration damage jumps around from 385 to 600 and several values in between, presumably due to other players' debuffs on the dummy. That screws up a lot of the values.

At the very least CS, SoT, Censure ticks, and Consecration ticks all seem to be right, confirming that Crusade works and the haste scaling of Censure with JotP.

Yeah I saw all of that extra info in the log afterwards. I was the only one hitting the dummy I was directly attacking. So I logged back in to look around a bit closer and there was one behind me with the same name. So I was not sure if it was splashing, or if it was picking up the damage on that other one as well. I'll try and find one somewhere else tonight. That was in Stormwind and the dummies are pretty close to each other. Boss dummy was pounded by tons of people so could not even get close to that one. Need to find a less busy place or time.

[Bastien]

...snip...

Boss dummy was pounded by tons of people so could not even get close to that one. Need to find a less busy place or time.

Try Theramore. It's usually pretty deserted.

[Neziah]

...snip...

Boss dummy was pounded by tons of people so could not even get close to that one. Need to find a less busy place or time.

Try Theramore. It's usually pretty deserted.

Good idea, thanks!

[theckhd]

Yeah those are the values listed on the tooltip for the sword. With the sword equipped my dmg on the char sheet is listed as 65-197 with 83 dps.

ap is at 563, sp is at 1990

That's very strange. CS damage is about 25% higher than it should be, despite having the same inputs. With a sample size of 64, we'd expect the standard error to be only about 7% of the range. The only thing I can think of is that we're looking at a statistical outlier. We'll have to take some more data with this set-up to see what's going on.

[Kierly]

Okay I did another test set of 400 crusader strikes, 100 with each seal on. Same gear set as before.

http://www.worldoflogs.com/reports/vtaexe1l8rpd1u74/details/0/

From looking over the data from the test it looks like it is still at the 218 amount.

Also Judgement of Truth is better lined up in this test.

Something is off somewhere...

[tlitp]

OK, time to test some things while beta is still up.

• race : any
• level : ideally 83/84, but 85 will do as well
• spec : holy/prot, with 3/3 JotP
• gear : 0 hit/exp, high Strength/haste (crit/mastery are irrelevant)
• weapon : Tomahawk (low-lvl vendors)
• seal : Insight
• buffs : none (including self-buffs)
• target : level 85 dummy, attacking from the front

How-to :

• keep haste and total AP constant (no relevant procs/enchants)
• keep Insight active
• try to use a less popular dummy ("overkill" values should be avoided, as much as possible; the same for target debuffs)
• autoattack the dummy for at least an hour (bare minimum); more is better

What to report :

• level, weapon damage (character panel), haste rating, total phys_crit (character panel)
• upload the log on WoL
• upload the raw CL (or a portion of it, say 20%) on a file-sharing service of your choice

Thanks.

[theckhd]

Okay I did another test set of 400 crusader strikes, 100 with each seal on. Same gear set as before.

http://www.worldoflogs.com/reports/vtaexe1l8rpd1u74/details/0/

From looking over the data from the test it looks like it is still at the 218 amount.

Also Judgement of Truth is better lined up in this test.

Something is off somewhere...

OK, that log confirms the JoT code is correct. In fact, everything in that log lines up well except SoR (still about 6% high) and CS (26-27% high). There's definitely something fishy going on with CS.

Looking more closely, here's the code for CS:

Code: Select all

raw.CrusaderStrike= 1.2.*player.wdamage.*mdf.phdmg.*mdf.Crus.*(1+2.*mdf.WotL).*mdf.t11x2;

mdf.Crus=1 because you're not specced into Crusade
mdf.WotL=0.3, for a net bonus from that term of 1.6 (Wrath of the Lightbringer)
mdf.t11x2=1 since I'm assuming you haven't farmed T11 on your premade
mdf.phdmg=0.6818 based on the dummy's armor and your level (lvl 85 armor mitigation factor)
player.wdamage=131 (average weapon damage from paper doll)

mdf.phdmg should be correct, since your HotR and melee hits are correct (player.wdamage*mdf.phdmg=89). If we omit this term (i.e. if CS was coded as magic damage), we get an expected CS of 251, which is too high.

mdf.WotL should be correct, since your Judgement damage is correct. That said, it's possible that it's bugged and giving too large a bonus to CS. If it were giving an extra 30%, it would give us an expected damage of 203 - a little low, and a fairly unlikely source of the error.

mdf.Crus should be irrelevant since you're not specced into it. However, if you had been specced into it, and it were giving a "phantom" bonus of 30% due to some sort of a bug, that would give us an expected damage of 223, which is fairly close to what you're seeing.

player.wdamage should be correct based on melee data. On the other hand, what if CS were coded to be speed-normalized? If it were coded as 2.0*player.wdps (player.wdamage is closer to 1.52*player.wdps), it would give us an expectation of 225, again close to what we're seeing.

The baffling part is that Ras's data is incredibly accurate to the model. On the other hand, Neziah's data shows a 29% difference as well, but since there's so much stuff going on in his log it's hard to tell if that's reliable.

For the moment, let's assume Ras's data is correct, and thus the model is correct for that configuration. What's different between Ras's situation and Kierly's?

-Different weapon (faster, different damage range)
-Different AP/SP (probably not significant in this case)
-Potentially different spec histories (if there was some sort of phantom talent bug, for example)

While it's certainly possible that they gave CS a spellpower scaling coefficient (i.e. to help Holy leveling), it's fairly unlikely. The most likely explanations in my mind are a phantom talent bug or some sort of weapon normalization below 2.0 speed (i.e. it could calculate damage via min([2.0 player.wspeed])*player.wdps instead of player.wdamage).

So here's what we'll do.

Test #1: Rule out phantom talents - solved

Step 1: Roll a new pre-made (i.e. no spec history whatsoever). Spec 0/38/0, get naked except for weapon/shield, use Seal of Righteousness, and log ~200 CS attacks. Either the prot or holy weapon/shield combination will be fine for this exercise.

Step 2: Spec your final 3 points into Crusade (0/38/3). log ~100-200 CS attacks, enough to determine a statistically significant difference.

Step 3: Unlearn talents, spec back into 0/38/0. Log ~200 CS attacks again.

This should tell us whether Crusade is giving us a phantom effect or not. Please do this all within the same talent tab (i.e. don't use dual-spec), in case it doesn't work when changing specs. Make sure to post your AP and SP from the character sheet as a consistency check.

Test #2: Rule out weapon normalization - Solved

Step 1: Find an array of weapons of different speeds between 1.4 (Dalaran Sword) and 2.6. This list of vendor weapons might be useful (note that if you clear the "Sold by Vendor" filter, it will list a speed column for easy sorting - I have no idea why the column disappears when that filter is applied). One weapon at each speed might be overkill, but increments of 0.2 speed would be good. If we get 4 or 5 people involved in this, it'll go faster, as each person needs to collect fewer data points.

Step 2: Spec 0/38/0, get naked except for the weapon (no shield for simplicity). Use Seal of Righteousness. Log about 100-200 CS attacks with each weapon. Please try and keep the logs separate (i.e. stop logging and upload after each weapon). Post a link to each log along with the weapon used and paper doll AP, SP, weapon damage range (i.e. X-Y), and swing speed.

This should give us enough data to determine if there's something funny going on with the base weapon damage component of CS. It should also give me an abundance of data with which to figure out where the 6% discrepancy in Seal of Righteousness is coming from.

[theckhd]

Also, has anyone observed whether Grand Crusader procs off of a miss/dodge/parry? The wording of the talent certainly seems to indicate that it does not, but since it should have a significant effect on the simulations, I want to be sure.

[Sabindeus]

Theck: shouldn't CS contribution from AP be normalized since it's an instant attack?