A Call to Arms - MoP Mechanics Testing

[theckhd]

For damage-dealing abilities, that should be fine. Anything that heals, though, will probably be skewed by it. Which means I either need to know how much PvP power you have in each gear set, or we need to use completely different gear sets devoid of PvP gear for heal testing.

[Klaudandus]

the weird thing is that it doesnt seem to affect the healing part of the lvl 90 talents.

i'll get you the pvp power per AP levels later on.

[Klaudandus]

Q:Confirmed that other abilities, like Light's Hammer is not affected by PVP Power -- but Eternal Flame is.

Still would like a confirmation on whether this is a bug or not.


A:It sounds like a bug, but to be clear, PvP Power should do nothing for your healing in a dungeon or raid. When healing someone in a BG, Arena or in the outdoor world, PvP Power should increase your healing. The only exceptions are percent based heals, which does not include Eternal Flame or Light of Dawn. Eternal Flame should work almost exactly like Word of Glory, and I'm not sure what could account for a difference.

Q:Prot paladin DPS relative to other tanks should be examined. Most changes to ret, and some changes to holy, affect Protection Paladins directly. EDIT: The bugfix to Eternal Flame make it somewhat unattractive to Protection. I'm a little concerned prot is going to gravitate towards Sacred Shield exclusively, especially because we have better ways to spend our Holy Power.

A:Prot paladin DPS is in a good place for us internally. The Ret nerfs didn't seem to drop Prot down below other tanks. Eternal Flame should feel like a slightly better Word of Glory for Prot, which is definitely a spell you do want to use. You should be able to figure out if the absorb over time from Sacred Shield provides as much defensive benefit as the hot from Eternal Flame.

[Klaudandus]

quick pvp power finding... ~330 pvp power adds 5 HPS to the EF tick

[Fetzie]

Was still listed on the first page, so I thought I'd check it quickly:

Test #37: Sanctity of Battle

Level: any
Gear: naked
Target: none
Goal: find a friend that can supply the 5% spell haste raid buff. Compare the tooltip of Crusader Strike with and without the buff active.
Report: Any change to the CS cooldown due to the buff.

No change in cooldown for crusader strike, Hammer of the Righteous, Avenger's Shield et. al. when a moonkin druid joined my group. Cooldown of Crusader Strike and HotR remained at 4.5 seconds.

See screenshot: http://i.imgur.com/j7MUJ.jpg

[theckhd]

quick pvp power finding... ~330 pvp power adds 5 HPS to the EF tick

That still doesn't tell us what the scaling function is.

For example, I can fit the data in the spreadsheet two ways. Using the minimum WoG data:

Code: Select all

General model:
     f(x) = 3694+301*0.377+a*x+b
Coefficients (with 95% confidence bounds):
       a =      0.1573  (0.1557, 0.1589)
       b =       336.5  (334.4, 338.7)

Goodness of fit:
  SSE: 11.3
  R-square: 0.9999
  Adjusted R-square: 0.9999
  RMSE: 1.373

Code: Select all

General model:
     f(x) = (3694+301*0.377+b)*(1+x/a)
Coefficients (with 95% confidence bounds):
       a =  2.634e+004  (2.606e+004, 2.662e+004)
       b =       336.5  (334.4, 338.7)

Goodness of fit:
  SSE: 11.3
  R-square: 0.9999
  Adjusted R-square: 0.9999
  RMSE: 1.373

This tells us two things:
1) the base heal value of WoG is off by around 337 damage compared to the tooltip. The max seems off by 364 as well from other fits.
2) PvP power could add healing linearly (like AP) or multiplicatively (as in, X pvp power increases WoG by Y%).

To discriminate between the two potential scalings, we need WoG data from gear sets 6A-6G. No need to repeat that for EF, we can reasonably assume the scaling function is similar between the two (i.e. EF isn't linear if WoG is %-based).

[Klaudandus]

yeah, I'm already working on it -- I had to stop for a while, went to eat some steak =P

Halfway done with EF, after that will do WoG6A thru 6G, just for confirmation.

[theckhd]

Well, this is interesting.

So far, ES and LH match their tooltips quite well:

Code: Select all

es_fit =

     General model:
     es_fit(x) = a*x+b+12000
     Coefficients (with 95% confidence bounds):
       a =       5.936  (5.936, 5.936)
       b =         989  (987, 991)

es_gof =

           sse: 0.5734
       rsquare: 1.0000
           dfe: 2
    adjrsquare: 1.0000
          rmse: 0.5354

lh_fit =

     General model:
     lh_fit(x) = (a*x+b)
     Coefficients (with 95% confidence bounds):
       a =      0.3225  (0.3182, 0.3268)
       b =        3295  (3269, 3321)

lh_gof =

           sse: 95.5817
       rsquare: 1.0000
           dfe: 2
    adjrsquare: 1.0000
          rmse: 6.9131


lh_fit2 =

     General model:
     lh_fit2(x) = (a*x+b)
     Coefficients (with 95% confidence bounds):
       a =      0.3192  (0.3113, 0.3271)
       b =        4005  (3958, 4053)

lh_gof2 =

           sse: 320.7961
       rsquare: 0.9999
           dfe: 2
    adjrsquare: 0.9999
          rmse: 12.6648

Holy Prism is another beast though. For offensively cast, the data is a bit noisy, but still a reasonable (if imprecise) fit:

Code: Select all

hpr1_fit =

     General model:
     hpr1_fit(x) = a*x+14523
     Coefficients (with 95% confidence bounds):
       a =       1.478  (1.323, 1.633)

hpr1_gof =

           sse: 8.8204e+005
       rsquare: 0.9889
           dfe: 3
    adjrsquare: 0.9889
          rmse: 542.2308


hpr1_fit2 =

     General model:
     hpr1_fit2(x) = a*x+17750
     Coefficients (with 95% confidence bounds):
       a =       1.378  (1.264, 1.492)

hpr1_gof2 =

           sse: 4.8083e+005
       rsquare: 0.9943
           dfe: 3
    adjrsquare: 0.9943
          rmse: 400.3441

However, it's pretty clear that the healing portion of HPr cast offensively is not quite right:

Code: Select all

hpr1h_fit =

     General model:
     hpr1h_fit(x) = a*x+9793
     Coefficients (with 95% confidence bounds):
       a =       1.348  (1.318, 1.378)

hpr1h_gof =

           sse: 3.2594e+004
       rsquare: 0.9995
           dfe: 3
    adjrsquare: 0.9995
          rmse: 104.2343

It's a really good fit... except that we're getting 134.8% SP scaling instead of 98.2%. Further investigation reveals that PvP power is probably the culprit:

Code: Select all

     General model:
     fr(x,y) = 9793+0.962*x+a*y
     Coefficients (with 95% confidence bounds):
       a =      0.7794  (0.7284, 0.8305)


gof =

           sse: 2.3485e+004
       rsquare: 0.9997
           dfe: 3
    adjrsquare: 0.9997
          rmse: 88.4771

     General model:
     hpr1h_sfit2(x,y) = 11970+0.962*x+a*y
     Coefficients (with 95% confidence bounds):
       a =      0.7029  (0.5517, 0.8541)


hpr1h_sgof2 =

           sse: 2.0611e+005
       rsquare: 0.9972
           dfe: 3
    adjrsquare: 0.9972
          rmse: 262.1118

So it looks like HPr is getting ~75% PvP power scaling. The multiplicative PvP power formula gives worse fits (though not that much worse, admittedly, given how limited the data set is).

I think we're simply going to need more AP/PVPP/healing data for WoG and Holy Prism at vastly different AP and PvP power values (i.e. outside the scope of our current gear sets).

I'm going to take a look at the self-cast data now to see how that compares.

[Klaudandus]

ok, will redo those then.

just to verify

[theckhd]

HPr self-cast healing is also affected by PvP power

Code: Select all

General model:
     f(x,y) = (14523+1.428*x+a*y)
Coefficients (with 95% confidence bounds):
       a =      0.9791  (0.8187, 1.14)

Goodness of fit:
  SSE: 2.322e+005
  R-square: 0.9985
  Adjusted R-square: 0.9985
  RMSE: 278.2

General model:
     f(x,y) = (14523+1.428*x)*(1+0.01*y/a)
Coefficients (with 95% confidence bounds):
       a =         257  (235.4, 278.6)

Goodness of fit:
  SSE: 6.124e+004
  R-square: 0.9996
  Adjusted R-square: 0.9996
  RMSE: 142.9


Curiously, the %-based scaling fit works better here. Though again, both are so close that it's really not possible to conclusively say which is correct. That said, compare this to the offensively-cast fits:

Code: Select all

General model:
     f(x,y) = (9793+0.962*x+a*y)
Coefficients (with 95% confidence bounds):
       a =      0.7794  (0.7284, 0.8305)

Goodness of fit:
  SSE: 2.348e+004
  R-square: 0.9997
  Adjusted R-square: 0.9997
  RMSE: 88.48


General model:
     f(x,y) = (9793+0.962*x)*(1+0.01*y/a)
Coefficients (with 95% confidence bounds):
       a =       219.4  (175, 263.9)

Goodness of fit:
  SSE: 2.225e+005
  R-square: 0.9969
  Adjusted R-square: 0.9969
  RMSE: 272.3


Now, it's entirely possible that the spell has different PvP power scaling for each version. However, the fact that the percentage-based version gives nearly the same conversion factor (219-257 PvP power per 1% healing increase) is suspicious. Especially since we got the same rough conversion factor from WoG:

Code: Select all

General model:
     f(x) = (3694+0.377*301+b)*(1+0.01*x/a)
Coefficients (with 95% confidence bounds):
       a =       263.4  (260.6, 266.2)
       b =       336.5  (334.4, 338.7)

Goodness of fit:
  SSE: 11.3
  R-square: 0.9999
  Adjusted R-square: 0.9999
  RMSE: 1.373

It could be coincidence, but I suspect that it indicates that pvp power is multiplicative.

For further evidence, let's look at EF:

Code: Select all

General model:
     f(x) = (4030+0.377*301)*(1+0.01*x/a)
Coefficients (with 95% confidence bounds):
       a =       259.5  (254.5, 264.5)

Goodness of fit:
  SSE: 172.3
  R-square: 0.9983
  Adjusted R-square: 0.9983
  RMSE: 4.961

On a lark, I tried using this fit for WoG as well:

Code: Select all

General model:
     f(x) = (4030+0.377*301)*(1+0.01*x/a)
Coefficients (with 95% confidence bounds):
       a =       262.8  (261.5, 264.2)

Goodness of fit:
  SSE: 11.99
  R-square: 0.9999
  Adjusted R-square: 0.9999
  RMSE: 1.309

That explains where the extra base heal on WoG was coming from. EF's base heal and WoG share the exact same formula now: (4030+0.377*SP) per holy power. WoG's tooltip just hasn't been updated to reflect that buff. EF simply adds the HoT.

Speaking of the HoT:

Code: Select all

General model:
     f(x) = (391+0.045*301)*(1+0.01*x/a)
Coefficients (with 95% confidence bounds):
       a =       269.2  (263.8, 274.6)

Goodness of fit:
  SSE: 1.633
  R-square: 0.9984
  Adjusted R-square: 0.9984
  RMSE: 0.4829

So again, EF's HoT component is consistent with (391+0.045*SP) and the newly-discovered PvP power scaling (1% per every ~260 PvP Power).

The only thing that remains at this point, I think, is to nail down the exact scaling factor more exactly. It might be faster to just ask the Devs for that in the beta thread, but I'll see what I can do in the meantime.

[Klaudandus]

Done with EF6A-G, halfway done with WoG6A-G -- will update the spreadsheet as soon as I finish, then I'll move in to redo self-cast HPr

[Klaudandus]

Question, do you need the damage data out of HPrism and the others or just the healing portion?

All raid level dummies are busy at the moment =(

[theckhd]

Just the healing portion. Damage seems fine.

[Klaudandus]

Just the healing portion. Damage seems fine.

Ok, i'm putting in the data for WoG6 and EF6 -- will do HPr and HPrSelf as soon as I'm done with this...

Should I do LH's healing portion as well?

[theckhd]

Don't bother, I think we're pretty well set on L90 talents now. I'm mostly interested in nailing down the PvP power scaling factor, and convincing myself that EF and WoG both have the same base heal formula.