Crit Suppression discussion

Incoherent · 04/08/09, 11:57 AM

The purpose of this thread is to provide a focus for discussion on the subject of Crit Suppression. The discussion so far has taken place in the Frostfire Bolt Thread and has pulled that thread somewhat off topic, the issue is relevant for all specs.

What is Crit Suppression?

Spell casts on mobs may affected by an unknown adjustment to Critical Strike chance. This has been observed by Hunters and Deathknights, and appears to affect Mages and likely other casters, but the details differ.

Testing on Target Dummies indicate that there is an effect. There is a reduction in Crit chance on Mobs (or at least Target Dummies) that are of a higher level than the caster. Theorycrafting approaches for Boss level mobs may need to take this into account. There are also hints that casts on lower level mobs gain a increase in crit chance.

How much?

Tests so far seem to suggest that the critical chance reduction on Boss level mobs by level 80 casters is about 2.5% but is dependent on gear levels. Extremely low crit chance levels (i.e naked) are not affected as much.

Mechanics of Crit Suppression

To be established

Observations:

It is affected by level difference.
A level difference of 0, e.g. lvl80 vs lvl80 shows negligible suppression.
Large negative differences in level, e.g. lvl73 vs lvl80 show a large crit suppression.
Large positive differences in level, e.g. lvl 80 vs lvl60 show a small crit Inflation
Lvl 80 vs Boss level Target Dummy shows about 2.5% crit suppression for crit chances above ~10%
It does not seem to be affected by Spell Penetration. More testing needed.
Unbuffed level difference of -2 seems to not suffer from Crit Suppression
0 crit rating seems to not suffer from Crit Suppression
A level difference greater than 3 levels experiences some form of Crit Rating dependent "clamping" of crit suppression. The lower the Crit rating, the lower the clamped Crit suppression.
At level differences greater than 3 levels, lower Crit Ratings suffer less suppression than high Crit Ratings.

Theories

Flat modifier
Can’t be. Naked tests with base crit chance of 2% can still crit, significantly more often than a flat modifier would allow.

Flat Modifier reducing (linearly?) below a certain crit chance
Possible. How to factor in level difference? per level?

Adjustment to Crit Rating of ~35 per -ve level difference.
Not fully tested but fits most of the data seen so far. Effect of buffs/talents/debuffs not tested

It is a side effect of Target Dummy testing. There is no Crit suppression on real bosses/mobs.
Possible. The target dummies seem to have odd behaviour, they appear to be level 83 rather than caster+3. Their Spell resistances seem higher than "normal" bosses.

Targets have Resilience
Not likely. At least not the traditional resilience we see on gear. If the target had resilience we would see the effect it had on things like DoT reduction and crit damage reduction.
The effect seems not to be reduced by Spell penetration, more testing needed to verify this.

Crit is calculated as if the caster was the same level as the Target
Possible? The data seen so far does not support this

Diminishing Returns of some sort is in effect
Possible?

Test Requests

These are the tests that need to be performed. If anyone can do these it would be great.

1. Test with 0 Crit rating vs Boss level
I am seeing in the results of 2 samples that 0 crit rating results in NO crit suppression, perhaps even inflation.

2. Test with different values of Crit rating, 0 - normal rating, vs Boss level

Data Collection

There is starting to be a bit of data on this. It would be great if anyone could contribute to the testing, so far the most practical method seems to be Arcane missile spam. I have a request that any data contains the following information (As contained in screenshots of Recount summary and the Character sheet):

Caster Level:  Your level
Target:        Your Target mob, boss, dummy, player
Target Level:  Your target level, boss level    
Spell cast:    Arcane Missiles, Scorch, Frostbolt etc.
Count:         Number of spells cast
Hits:          Number of spells that hit (not crit)
Crits:         Number of spells that crit
Misses:        Number of spells that missed
Hit Rating:    Paper doll hit rating.
Crit Chance:   Paper Doll crit chance. 
Crit Rating:   The total Crit rating points from your gear. (mouseover your paperdoll crit chance.)
Intellect:     Paper doll Intellect for the duration of the test. 

Any other relevant information; buffs, debuffs, talents.
Preferably test without talents or gear that proc Crit affecting effects (FoF, Arcane potency etc)

Data Collected so Far

Who--------	Level	TrgLvl	Count	Hit	Crit	Miss	Hit R	Crit Ch	Crit R	Int	Bf/Tl	GrCr	IntCr
Exactly----	74	80	1177	700	31	446	155	0.0746	87	388	0.00	0.0294	0.0361
Exactly----	75	80	2534	1779	71	684	155	0.0704	87	391	0.00	0.0273	0.0340
Exactly----	76	80	2234	1752	61	421	155	0.0702	100	394	0.00	0.0292	0.0319
Exactly----	77	80	2820	2515	120	185	155	0.066	100	397	0.00	0.0271	0.0298

Exactly----	73	80	3051	1545	53	1453	155	0.1189	198	374	0.00	0.0720	0.0378
Exactly----	74	80	2771	1658	63	1050	155	0.1111	198	377	0.00	0.0669	0.0351
Exactly----	75	80	2688	1916	83	689	155	0.1043	198	380	0.00	0.0622	0.0331
Exactly----	76	80	3039	2372	116	551	155	0.0979	198	383	0.00	0.0578	0.0310
Exactly----	77	79	579	524	55	0	245	0.0917	198	386	0.00	0.0537	0.0290
Exactly----	77	80	3178	2748	202	228	155	0.0917	198	386	0.00	0.0537	0.0290
Exactly----	77	81	888	702	31	155	155	0.0917	198	386	0.00	0.0537	0.0290
Exactly----	77	82	1440	972	30	438	155	0.0917	198	386	0.00	0.0537	0.0290
Incoherent-	80	80	8186	7500	686	0	161	0.0851	198	549	0.00	0.0431	0.0329

Exactly----	73	83	2579	449	17	2113	155	0.1189	198	374	0.00	-10	0.0720	0.0378
Exactly----	74	83	2360	653	33	1674	155	0.1111	198	377	0.00	0.0669	0.0351
Exactly----	75	83	2517	936	40	1541	155	0.1043	198	380	0.00	0.0622	0.0331
Exactly----	76	83	4663	2244	76	2343	155	0.0979	198	383	0.00	0.0578	0.0310
Exactly----	77	83	2545	1511	56	978	155	0.0917	198	386	0.00	0.0537	0.0290
Incoherent-	80	83	7152	6409	410	333	161	0.0851	198	549	0.00	0.0431	0.0329
													
Who--------	Level	TrgLvl	Count	Hit	Crit	Miss	Hit R	Crit Ch	Crit R	Int	Bf/Tl	GrCr	IntCr
Zaldinar---	80	83	557	466	10	81	0	0.0199	0	0	0.00	0.0000	0.0000
Incoherent-	80	83	3960	3647	209	104	223	0.0507	0	694	0.00	0.0000	0.0416
Soulseek---	80	83	5512	5122	390	0	417	0.0848	155*	851*	0.00	0.0338	0.0511
Soulseek---	80	83	6369	5979	390	0	417	0.0848	155*	851*	0.00	0.0338	0.0511
Incoherent-	80	83	7152	6409	410	333	161	0.0851	198	549	0.00	0.0431	0.0329
Incoherent-	80	83	2791	2419	315	57	227	0.1309	106	812	0.05	0.0231	0.0487
Zorb-------	80	83	4000	3385	615	0	374	0.1681	470	944	0.00	0.1024	0.0566
Incoherent-	80	83	3150	2611	478	61	238	0.178	522	919	0.00	0.1137	0.0551
Incoherent-	80	83	2705	2233	432	40	241	0.1858	565	894	0.00	0.1231	0.0536
Shaewyn----	80	83	2501	1988	513	0	350	0.2307	405*	1043*	0.08	0.0882	0.0626
Moonraker--	80	83	5000	3947	1053	0	289	0.2321	599*	1193*	0.03	0.1305	0.0716
													
*guess

Relevant Tables

Miss Chance			Combat ratings. Rating/x=chance					
Lvl D	Miss		Level	Int	Hit	Crit	Haste	
4	0		60	?	800.0	1400.0	1000.0	
3	0.01		70	8000.0	1261.5	2207.7	1576.9	
2	0.02		71	8620.7	1357.4	2375.4	1696.7	
1	0.03		72	9259.3	1460.4	2555.8	1825.6	
0	0.04		73	9901.0	1571.4	2749.9	1964.2	
-1	0.05		74	10752.7	1690.7	2958.8	2113.4	
-2	0.06		75	11494.3	1819.1	3183.5	2273.9	
-3	0.17		76	12345.7	1957.3	3425.3	2446.6	
-4	0.28		77	13333.3	2106.0	3685.4	2632.5	
-5	0.39		78	14285.7	2265.9	3965.4	2832.4	
-6	0.5		79	15384.6	2438.0	4266.5	3047.5	
-7	0.61		80	16666.7	2623.2	4590.6	3279.0	
-8	0.72							
-9	0.83		Mage constant		0.009064			
-10	0.94

Addon Development

Ideas for the development of an Addon that could be used to assist in evaluating effects like Crit Suppression. This is probably deserving of it's own Thread as well, but here goes.

I am not a competent to even say what is possible with lua, nonetheless, here is a wish list.
-An addon that logs and saves an output similar to the /combatlog.
-Rather than logging every event, it only logs Spell casts/misses and DoT ticks/misses by the Player running the addon but contains the same information as a given line in the WoWCombatLog.

So far so good, it's just a filtered /combatlog.

-On the same output line, delimited in the same way, a number of additional values collected and calculated from both the Player Character Info and Buff/Debuff and the Target Debuff stack. Namely, anything and everything that affects; damage done, crit and hit rates/ratings, haste rating, cast speed. Not necessarily the actual debuffs/buffs/procs but the dynamically calculated, current at time of casting, values.

Addons like DrDamage can do this dynamic calculation so I am sure it is possible. It would only need to log on events like SPELL_CAST_SUCCESS, SPELL_MISSED or SPELL_DAMAGE although things like Haste and Cast Speed would need to be calculated on SPELL_CAST_START.

A line in this modified log might look like this:

4/8 02:14:56.175  SPELL_DAMAGE,"Incoherent",80,"Thaddius",Boss,"Frostbolt",28756,0,1,1034,543,2865,321,548,644,1.00,0.83,1.55,3.20

Being:

Date Time  SPELL_DAMAGE,Player,Player Level,Target,Target Level,Spell,Hit for/Missed?,Damage Resisted?,Crit?,Int,Spi,SpellP,HitR,CritR,HasteR,Hit Chance,Crit Chance, Cast speed, Damage Multiplier

The idea here is that if we are in possession of all the variables for every spell cast throughout a raid, we could actually use the information in our Raid logs for proper analysis, we could see why a spell missed, hit for some astronomical amount, not because we see that Dying Curse procced simultaneously with a mana Gem and Potion while standing in 3 Power Spark remains, but we'd be able to at least see the for-that-cast Spellpower, Crit rate etc values - something we cannot do with the current generation of logging addons.
Such a log would be very compact, only logging on Spell release, perhaps a few thousand lines for an entire raid. Maybe some issues with AoE.
Thoughts?

---

Please feel free provide some feedback and suggestions for this, particularly under the Observations and Theories headings. I am taking the liberty of posting as I think it deserves it's own thread.

To do:

Link relevant threads

Changes

12/04/09 Reformatted and updated data table
Added Combat ratings tables.
Added Test Requests

Megaera · 04/08/09, 2:56 PM

Have we considered a solution as simple as crit rating having an undocumented DR formula? I think it's important that we standardize testing methods so that we don't have some people varying expected crit values using talents or Int (which are likely to be off of a DR curve) and others varying expected crit values using rating-based gear/gems.

Imnumber01 · 04/08/09, 3:36 PM

Rather simple thought that may have been overlooked.

Maybe the crit-% on a character sheet is only for a same lvl mob.

And as the mob lvl increases the formula that calculates normal hit vs crit is actually applying our crit rating to a lvl+3 table that would require a higher crit rating to obtain the same percentage of average crits.

*(yes I know that if this worked inversely it would mean you would have 100% crits on some lvl mobs which isn't the case, but its not out of the realm of possibility that Blizz coded it so that it only works one way or has a maximum + or - on either side of it or even that it wouldn't work at all in pvp)

Zerstorung · 04/08/09, 3:58 PM

I believe I sparked this discussion back in the FFB thread when talking about how my RNG was always low. Now, despite possible crit suppression or diminishing returns, this doesn't remove the possibility of getting an unusually high RNG and critting more than your paper doll crit rate, does it? I ask because there have been times where I have crit 70+% on a boss fight, despite having 60% crit.

Enthorn · 04/08/09, 4:18 PM

No Zerstorung, your scenario is not the same thing as crit suppresion. If there is a crit suppression, then over the course of several hundreds of fights (assuming ~100-200 casts per fight), then yes, you could notice a trend, but realistically wouldn't. You would have to document your crit rate on a per cast basis, and even then, your overall crit rate for a fight is an average between all spells. For Frostfire builds, Pyroblasts and Living Bombs have 6% higher crit than Frostfire Bolt. Just as well, the 3% crit on Improved Scorch only affects FB/FFB/Scorch.

Regardless, unusually high RNG or unusually low RNG is not something that crit suppression is influencing short-term (or even long term, one could say, as that's why it's RNG based in the first place). This is why Loatheb is a lucrative fight for crit suppression testing. While maintaining a 100% crit rate for every spell cast could prove difficult, it goes without saying that if a boss has zero crit suppression, and you have 100% crit rate, then every spell cast should crit. At the same time, you could have a 60% crit and still end up with a 100% crit rate, which is one of the reasons thousands of casts are needed for any accuracy.

Althea · 04/08/09, 4:51 PM

Please add an addon section to the main topic, and write down what that addon should do/track. I will start the developing as soon as i have some main points to work with

Canyc · 04/08/09, 5:23 PM

For reference, could someone post a table of crit rating needed for 1% crit at various levels? I'd wager that the player's crit rating is being applied as if the player were of the target's level -- with some sort of normalization factor/caps being applied.

♦ nathanbp · 04/08/09, 5:36 PM

Originally Posted by Canyc

For reference, could someone post a table of crit rating needed for 1% crit at various levels? I'd wager that the player's crit rating is being applied as if the player were of the target's level -- with some sort of normalization factor/caps being applied.

http://elitistjerks.com/f31/t29453-c...gs_level_80_a/ says it should take ~25% more crit rating for 1% crit at level 83 than it does at 80 (57.18 crit rating = 1%). I don't believe that this matches the data so far for crit rates vs. boss level targets, although the data in that thread for rating conversions over level 80 could be wrong. That thread also has the information for determining the conversions at levels 1-80.

Althea · 04/09/09, 4:25 AM

Tonight i will start the development, following the guidelines in the main topic, please suggest an addon name so i can upload it on curse.com and let you all test it; additionally i could use some addon names that track events like we want, in this way i can look at the code and learn something

Sorry for my english, i am ita

Incoherent · 04/09/09, 7:35 AM

Originally Posted by Althea

Tonight i will start the development, following the guidelines in the main topic, please suggest an addon name so i can upload it on curse.com and let you all test it; additionally i could use some addon names that track events like we want, in this way i can look at the code and learn something

Sorry for my english, i am ita

CAnaLog? (Cast Analysis Log) SCALE? (simplified combat/cast analysis logging environment)... I am terrible at these things... With the idiotic tidal wave of offensive anal jokes that periodically flood Trade on my server I'd avoid AnaLog (Analysis Logging) - although that kind of appeals. How about RELEVANCE? REaltime Logging and EVAluatioN of spellCast/(Combat?) performancE. Actually I like that. Someone else suggest something less contrived. Please?

Recount is obviously a good place to start, it's recording damage from the whole environment. DrDamage is able to calculate things like Hit chance and Crit chance for specific spells depending on what buffs/debuffs etc are currently up, but the addon has not been updated for several months and is not accurate- I still use it for quick gear swap rough comparisons though.

Exactly is now 76. I am seeing that the earlier results are not noise - on Boss Level target dummies, between level 74 and 76, Crit suppression is almost constant, at about 6.3% with Crit Rating of 198. At 73 (10 levels delta) it suddenly increases to 8.3%.
Versus a level 80 it is maintaining a constant rate of improvement. The Boss level suppression at 74-76 (delta -9 to -7) is the same as 75 vs 80 target dummy (delta -5).

Althea · 04/09/09, 8:36 AM

RELEVANCE approved, i'll download DrDamage to see how it work and if it can be updated for our purposes

Thegoodmann suggested to

Every 1000 hits/crits, average the crit percentages logged and then store them into a file that has them as [spec,spell,hit/crit, actual,paper,delta].

And i think it's a good pattern to follow to have a set of results divided by files, without having to parse each single relevance log, but other opinions are welcome

Jonny_Monroe · 04/09/09, 9:02 AM

Targets have Resilience
Possible? The effect seems not to be reduced by Spell penetration, more testing needed to verify this.

Not likely. At least not the traditional resilience we see on gear. If the target had resilience we would see the effect it had on things like DoT reduction and crit damage reduction.

Shaewyn · 04/11/09, 1:04 PM

Perhaps one of the effects of this crit-suppression that has been overlooked is its effects on different specs.

A FFB or FB specced mage will suffer far more from any crit suppression than will an arcane mage or affliction warlock. As far as I know, this has not been accounted for in any serious theorycrafting (Rawr or Simulationcraft).

One thing I would also like to point out is that for any addon collecting real boss-fight damage, it's also going to have to collect cast-time crit rates, including buffs, debuffs, and other effects. Paper-doll crit rates are not enough information. Might it be easier to write a combat-log parsing utility? This could look for all of the relevant buffs/debuffs/casts and calculate each cast's crit chance...and with the widely varying crit rates as boss fights go, truly stupendous numbers of casts would be needed to get any meaningful statistical trends.

dedmonwakeen · 04/11/09, 1:57 PM

Originally Posted by Shaewyn

Perhaps one of the effects of this crit-suppression that has been overlooked is its effects on different specs.

A FFB or FB specced mage will suffer far more from any crit suppression than will an arcane mage or affliction warlock. As far as I know, this has not been accounted for in any serious theorycrafting (Rawr or Simulationcraft).

SimulationCraft does not apply crit-suppression to spells (only attacks) because we were unable to find reliable data on the subject.....

Zaldinar · 04/11/09, 6:39 PM

I have a simple combat log parsing script written in PHP that can handle the task provided you supply it with information about your stats. If people would be interested I could write it a simple web interface to upload a /combatlog dump, answer a few questions, and then get a result. The only concern is that we'd need accurate values for the questions, and the upload process would likely be slow.

Incoherent · 04/11/09, 8:21 PM

Originally Posted by Shaewyn

One thing I would also like to point out is that for any addon collecting real boss-fight damage, it's also going to have to collect cast-time crit rates, including buffs, debuffs, and other effects. Paper-doll crit rates are not enough information. Might it be easier to write a combat-log parsing utility? This could look for all of the relevant buffs/debuffs/casts and calculate each cast's crit chance...and with the widely varying crit rates as boss fights go, truly stupendous numbers of casts would be needed to get any meaningful statistical trends.

Absolutely agree. It would have to do some calculations in near realtime, using both current at time of cast paper doll numbers as well as buff/debuffs. The problem with the Combatlog as I see it is information overload. If we could in effect parse and filter as it is being generated it would dramatically reduce the complexity (and file size).

Any parser of the current combat log is going have to track through the log what buffs/debuffs/effects are up, decide if they are relevant and "know" what they mean, Zaldinars parsing script sounds like it might work, with the potential for inaccuracies if the entered stat values are not the same as what they were when the Combat logging was started - that also goes for effects (like say the Onyxia buff) that are up the whole time, if an effect doesn't commence or fade, it won't be logged as far as I am aware. Add to this the sheer size of the WoWCombatLog and I'm discouraged from pursuing this approach.

Today with the assistance of a friendly neighbourhood Paladin I got a bit of data with Exactly at 77 versus some Pustulent Giants of levels 81 and 82 in addition to the normal data vs Target Dummies. The data matches pretty exactly, when plotted against level difference, with the level 80 Target dummy at levels 75 and 76.
I also attempted a small population sample test against the level 79 Library guardians in the Storm Peaks in the same gear. That showed no Crit suppression at all.

I have updated and reformated the data tables in the original post. Also added Crit from Intellect and Crit from gear as I calculate it.

Incoherent · 04/13/09, 8:08 PM

I've been twisting myself inside out looking at this data over the weekend.
I think Crit chance is being affected by chance to miss. This is the bulk of the Crit suppression - but not all of it. For low levels vs high levels there are some other effects... may be also coupled to Hit.
Unless I am being stupid, The logical way to calculate Crit rate is:

Crit Rate = Crits/(Crits + Normal Hits) = Crit Chance, or (Hit Rate * Crit Chance)/Hit Rate if including Hit for clarity (it cancels)

In practice it is more like:

Crit Rate of landed casts = Crit Chance x Hit Rate .

An interesting thing. To me it seems that Crit chance from gear and perhaps Buffs/talents is not benefitting from Hit rating, while Crit from Intellect IS. That has an interesting ramification if it is the case.

I will elaborate tomorrow I hope. Trying to sort some webspace for Charts. The pictures tell it a million times better. I am satisfied that the data collected so far is "good", it is not being swung heavily by RNG. Supporting this is the fact that Experienced Hit rate matches Theoretical Hit Rate very closely for almost all the samples.

Exemplar · 04/14/09, 8:32 AM

This would be a factor of the "two roll system" used for spells. The first "roll" determines whether a spell is a hit, a miss, or a full resist. Only if it is a hit does it proceed to the second "roll" to determine whether normal hit or a crit. Therefore if it doesn't hit, it cannot crit. Below hit cap eats into effective crit chance. Wowwiki has had this data for years - The effect of Hit chance on Critical Hit chance

I'm not sure why the mages are jumping through hoops to come up with complicated mods to measure and assess this. Go glance at the tests performed by Rogues in their thread on Hit. They happened to stumble on melee hit reduction and tested that quite thoroughly while they were at it. Simply hit cap, hit the dummy 10,000 times with the same spell (yes, ten thousand - you need to reduce statistical deviation as much as possible) and unchanging crit (nothing which can proc to alter crit) - measure paperdoll crit and recount. Increase/decrease crit rating while still being at/over hit cap. Rinse and repeat.

Incoherent · 04/14/09, 12:08 PM

Originally Posted by Exemplar

This would be a factor of the "two roll system" used for spells. ...

I perhaps was not clear in my post above. The data I have looked at (not 10,000 casts but 130,000) says that the Gospel according to WoWWiki is not correct. This is not behaving as a two roll system. If it was a two roll system, the Crit rate of casts that land, not total casts, would be the same for any Hit rating.

I am not saying yet that I think this is a one roll system, but so far it sure looks to me like it - or some variant. I think I'll continue my hoop jumping when I get home from the pub.

Zaldinar · 04/14/09, 6:08 PM

Originally Posted by Incoherent

I've been twisting myself inside out looking at this data over the weekend.
I think Crit chance is being affected by chance to miss. This is the bulk of the Crit suppression - but not all of it. For low levels vs high levels there are some other effects... may be also coupled to Hit.
Unless I am being stupid, The logical way to calculate Crit rate is:

Crit Rate = Crits/(Crits + Normal Hits) = Crit Chance, or (Hit Rate * Crit Chance)/Hit Rate if including Hit for clarity (it cancels)

In practice it is more like:

Crit Rate of landed casts = Crit Chance x Hit Rate .

An interesting thing. To me it seems that Crit chance from gear and perhaps Buffs/talents is not benefitting from Hit rating, while Crit from Intellect IS. That has an interesting ramification if it is the case.

I will elaborate tomorrow I hope. Trying to sort some webspace for Charts. The pictures tell it a million times better. I am satisfied that the data collected so far is "good", it is not being swung heavily by RNG. Supporting this is the fact that Experienced Hit rate matches Theoretical Hit Rate very closely for almost all the samples.

So you're saying that given a 10% miss rate and 10% crit rate, your end table looks like:

(100 * 0.1) = 10% miss
90 * 0.1 * 0.9 = 8.1% crit
81.9 % hit

Ive got the next three days off. I'll see if I can get what we talked about into a more public working state, since its mostly capable of isolating all the things you're talking about. Even though I really think theres a more simple solution to all this.

Incoherent · 04/14/09, 7:54 PM

Originally Posted by Zaldinar

So you're saying that given a 10% miss rate and 10% crit rate, your end table looks like:

(100 * 0.1) = 10% miss
90 * 0.1 * 0.9 = 8.1% crit
81.9 % hit

...

Yes, that's the gist of it. It gives a strong correlation with the raw data. One of the better fits I have so far is along the lines of:

Critrate = %CRITchance x %HIT^2

But a little more complex with regards crit from int, gear, talents, buffs and class constant.
The problem is around the kink in the "miss" Table, the lower level un-hitcapped crit is bumping into the unknown of the roll system.

Don't put yourself out regarding addons. Perhaps it's enough to say that a hitcapped level 80 experiences about 2.0% Crit Suppression, increasing to as high as 4% if not hitcapped.
I'm going to keep poking at what I've collected so far, there is some interesting stuff there.

Exemplar · 04/15/09, 8:40 AM

Originally Posted by Incoherent

I perhaps was not clear in my post above. The data I have looked at (not 10,000 casts but 130,000) says that the Gospel according to WoWWiki is not correct. This is not behaving as a two roll system. If it was a two roll system, the Crit rate of casts that land, not total casts, would be the same for any Hit rating.

I am not saying yet that I think this is a one roll system, but so far it sure looks to me like it - or some variant. I think I'll continue my hoop jumping when I get home from the pub.

Indeed, there is 99.999% chance of crit reduction. Melee has it, I'm confident spells have it.
It should be:
Chance to Crit = (Hit% * Crit%) - Crit Suppression

Therefore simply testing in maximum hit gear with varying degrees of crit should (one hopes) show a linear offset. You cannot test two variables simultaneously (hit AND crit). Therefore if it's not linear, you've at least removed hit from the equation.

From your orignal post:
[quote=Incoherent;1184174]Flat modifier
Can’t be. Naked tests with base crit chance of 2% can still crit, significantly more often than a flat modifier would allow.
Naked is not hit capped - of course it will deviate, that's part of my point.

Your table of data is mostly useless - you need WWS for intelligent information - a single application of, say, Heart of the Crusader by a stray Ret paladin for as little as one minute of testing can taint an entire data set.

You don't have a sample of 130,000 - you have multiple 4000 to 8000 tests at varying levels of hit and crit. The statistical devation on each individually is significant. My understanding with 10k test attacks deviation can still be + or - 0.1% - that's huge enough when testing by how much 5% crit is reduced. Because hit and crit aren't the same between the tests it is invalid to combine and treat them as one single set of data. Each individual test with X hit and Y crit should be 10k attacks or more. Preferably more so you can chop out any portions where potentially changing debuffs are present (Misery/Imp FF if under hit cap, Heart of Crusader/Imp SB/Imp Scorch, etc) and still have more than 10k valid attacks for a decent data point.

I can understand your desire to capture large amounts of data rapidly via a mod that will record this from live raids. However the logistics are impractical - you would continue to receive small data sets which are not validly able to be merged. Gothik stat debuffs - int drops, crit changes. Draenei moves out of range or dies - hit changes. Someone gets an upgrade mid-raid, throws it on - hit and crit both change. It's impractical. Rather than spend dozens/hundreds of hours in mod development time, just devote that time directly to pewpew at a target dummy at various levels of crit. Yes, it's mind numbingly boring - but it'll get you the most accurate data possible.

Once you get a number for hit capped crit suppression, then you can go back and run the same tests with less hit and see if results continue linearly or derive your equation if not.

Again - I don't disagree with the premise and I'm personally interested in the end results. I just want to make sure you don't waste time/effort generating useless/misleading data, which is all the first post table contains at this time.

Jonny_Monroe · 04/15/09, 8:48 AM

Just a thought off the top of my head...

Do bosses have an effective defense skill? If they do, it would theoretically be 415, which would create a small reduction in attacker's chance to crit them (and increased chance to miss them, but that would be indistinguishable from the level-based miss chance).

The more I think about it, the less it seems likely. But its one idea to consider.

Telvin · 04/15/09, 9:24 AM

If bosses had defense 415 this would not affect any casts on them.
I doubt an analogon to defense rating exists for melee dps as it is possible for them to never ever miss if geared properly.

Incoherent · 04/15/09, 10:15 AM

Originally Posted by Exemplar

...You cannot test two variables simultaneously (hit AND crit). ...

Don't really agree. You cannot easily analyse tests with two simultaneous variables. Especially not if the sum total of your statistical analysis capability consists of "Add Trendline" to an Excel chart. I don't really want to turn this into a defense of the method argument or anything, there is data there for anyone who wants to look at it. You judge for yourself whether it is useful to you. I'd thank you for not describing it as "useless" though. It's some of the most controlled data you are ever going to see.

Yes the statistical variation for each set is significant, if you are only looking at that set. By looking at many sets you reduce the significance of the errors in those sets - all they contribute to is the noise level. The test you describe is in my opinion producing a population sample of one. Not very useful for statistical analysis. A lot of (sure, inaccurate in themselves) tests becomes progressively more useful as the total number of tests increase. Hence this approach. There are very clear systematics in the data you describe as useless, which tells me that it isn't.

That said, we are working not against physical behaviour, but the whims of Blizzard programming where they can put any conditional, exception, lollypop or banana they want into the equation. In that sense I totally agree with you, removing as many variables as possible is helpful. I just don't think that they are going to spend weeks working out complicated phase of the moon dependent formulae for Crit Suppression. It's going to be dead simple.

There is level dependency in the equation. That may just be the Miss rate. But if we really want to test it we have to run tests at different levels. The difficulty of hit capping against targets that are 4,5,6 or more levels above you make the kind of tests in the first post the only way you're going get any information whatsoever about that. If you think of them as nothing but data points for testing a hypothesis, then they're useful.