The RNG does not affect your crit cap whatsoever. Even if you got unlucky and 100% of your attacks during a short fight were glancing blows that does not change the crit cap.

While it is true that the RNG does not affect the crit cap it is not true that it has no effect on the value of the stat outside what is already modeled in the averages of the sheet. Because the sheets only run calculations on average values of misses, glances, dodges, and criticals it means the effect of the crit cap on DPS at critical chances near the crit cap is larger than what the sheets reflect. According to the sheet if the average miss rate, dodge rate, and glance rates all add up to a number smaller than one minus the average critical rate then there is no negative effect on DPS. This is not the case. Because of the way these values are being approximated as averages there is an underlying assumption that there is an equal chance on any given fight of being 1% over or 1% under the average critical rate, that the critical chance on any given fight is a symmetrical bell curve around the average. Take the following as an example of why this could be problematic.

Let's say a player has the following stats:
24% Glancing
10% Miss
4% Dodge

Which puts the critical cap at 62%

Their tooltip crit is 65.8% which puts their average boss critical rate at 61%. Just under the crit cap so every point of crit they have is working to maximum effect, right? No. A bell curve is predicated on the assumption that the distribution of possible outcomes is symmetrical but this is not the case. What's the chances of, at an average of 61% crit, experiencing 59% crit? It's some number greater than 0 that will depend on the number of attacks that took place in the fight. What are the chances of, at an average of 61% crit, experiencing 62% crit? It's not 0, and it is some number greater than 0 that will depend on the number of attacks that took place in the fight, but it is also demonstrably smaller than the chances of scoring 59% crit. Why? Because the chance of glances, misses, and doges summing to exactly their average is greater than any other which means that, close to the crit cap there's a bigger chance of having less critical hits than the straight average then there is of having more. This is true for every average, tooltip crit rate larger than half the crit cap and the effect gets larger and larger as the average critical chance approaches the crit cap.

Now Ald can correct me if I'm wrong but the sheets, as far as I know, do not account for this interaction, so there is some degree of merit in Tofuu's point about the RNG.

I just want to defend my point once more before I drop this topic: We will use the simple scenario of 100 hits on a target dummy as the baseline for analysis.

Scenario #1: I perform 100 hits on the lvl 83 Dummy, out of 100 hits, all of them glance. The probability of this outcome given infinite number of repetition of this trial is P1. Of course, P1 is understandably very low you probably will never encounter such event in your entire wow life and therefore this outcome has no bearing on our gear choices.

Scenario #2 Similar situation, 100 hits, 27 glancing attacks, probability of this outcome (again given infinite repetition of this trial) will be P2. However, P2 would be a much larger value.

If we make a graph of Time VS [(number of glancing attacks)/(total of number of white attacks performed)] (which is basically time vs percentage of glancing), We will see very large peaks in the very beginning of the time axis, and this peaks will slowly come to rest to reach the value of 0.24 as time goes on.

P1 is very small, and probably we don't need to calculate its actual value.

P2 can be calculated, but I honest don't have the resource on the laptop in my room to do it at this point. I am sure someone can do a simple probability test.

Now, Introduce the following stats weight
4% Glancing
10% Miss
4% Dodge

Which puts the critical cap at 62% and if the average critical chance of the person of interest is 61 percent, the question becomes what is the probability of someone having over 25 percent glancing attacks (we call this P3) and also what is the probability that this same person will crit more than 61 percent (P4). The variable is time because with long enough time, glancing will settle at 24 percent and crit will also settle at its average value. So we basically have to determine P3 and P4 for this particular stats weight and introduce a time value that will output a probability that while being 1 percent under crit cap, the dps loss will be significant.

I really do apologize for not giving concrete numbers, I will try to calculate P2 P3 P4 when I find some free time while I am at my lab. In either case, I hope I've made my point clear, even if these turn out to be incorrect.

Edit: I just saw Tetra's post after posting what I have written down here. I also used his stats weight and agree that the probability distribution will be essentially normal.

Well, given that if you're using agi procs like mongoose or Death's Verdict, or DM:G, your 'average' is really skewed upwards because you get big procs for a short amount of time, then you would run below your average crit rate most of the time, with jumps up way beyond the crit cap (of which all but the AP value of the agi is wasted, or all of the excess proc if it's crit rating).

I think I get what you're trying to say here, the problem is that you have in a sense made two categories on the hit table, one with glancing, misses, and dodges, the other with normal hits and crits. It is true that glances, misses, and dodges have the greatest chance to be at their averages but from what you've explained these three are nearly predetermined and leave a portion of the hit table where crit chances only have a tiny space to move higher then normal and more of a chance to be less then normal which kind of assumes that your chance to hit or crit is based on your chance to glancing, miss, and dodge which of course isn't true. Crit has just a much chance to move that "more" higher.

In a real fight using a bell curve isn't really the best way to envision the chance for five outcomes to occur because you in a sense will base what happens with the other variables around the one in question. Even with a pie chart (which is a good way to look at results and what has happened) you'll tend to think "there's only this much room for this" which is where you start making outcomes dependent on each other. The best way to look at this is to a roll a lop-sided dice or the draw M&Ms from the bag thing (assuming you put them back every time). It's unfortunately intellectually simple but, each roll/draw is independent and none of the sides/draws hold any constraint on the others just like every swing in an encounter. You have the highest chance to be around the average for each of these outcomes, even at or near the crit cap you have a the same chance to x% higher or lower; it'll look higher because you did more crits then normal and/or you did less then normal of some or everything else and vis versa.

Perhaps this is user error on my part but I turned on the "recommend expertise gem" feature to see what it would say. As I expected it said I need expertise, after adding 1 gem I was over the cap but it still recommend expertise gems until I equipped 2 more.

None of you guys talked about the time constraint, which is the key. The reason that there exists Crit cap is due to the fact that glancing attacks holds a higher priority than critical strikes in the hit table. Crit and glancing attacks are not dependent upon each other. Glancing is only one of the variables that defines what crit cap is.

The point I am making here is when the time is short enough, there's a calculable probability that critical strikes will be pushed off the hit table. This probability becomes significant when one's very close to the crit cap and at the same time the fight duration is short.

Therefore, as I mentioned earlier, the variables are the duration (time) and the proximity to crit cap. The one thing I am not sure if everyone agrees on is that given short enough duration, glancing is not always going to be at 24 percent and crit has the probability to hit its upper bound region in the probability distribution. The actual raid crit of 59 percent is not saying 59 percent is the highest it can achieve. 59 percent raid crit is the crit value it will arrive at given long enough duration. The shorter the fight, the closer one is to the crit cap, the more likely the upper region of the crit probability distribution will be wiped off through glancing attack priority over crits.

Just think about the two graphs Time vs glancing percentage and Time vs critical percentage, it's not going to be a straight line sitting at 0.24 for graph 1 or a straight line of whatever the raid actual crit is for graph 2. It's going to fluctuate and eventually settles down given long enough time to reach the expected "average value". The problem is again, when actual crit is so close to crit cap, there's a chance that glancing will become a limiting factor for crit when the duration is short enough.

I don't think you understand the concept of single roll hit table that others have pointed out already. Everything is ALREADY PREDETERMINED, it doesn't matter if you get 100% glancing or 0% glancing in however amount of time, it does NOT affect crit cap. 24% glancing is a SET percentage, it will ALWAYS be 24% on the hit table.

When you say that "The shorter the fight, the closer one is to the crit cap, the more likely the upper region of the crit probability distribution will be wiped off through glancing attack priority over crits." you are indicating a dependency of crit on glancing which is incorrect. Glancing blows do have a higher priority then critical strikes in the sense that if you had no chance to miss or be dodge and had a 100% crit chance from your gear, you would still have a 24% chance to have a glancing blow every individual swing. The only way glancing blows could push critical strikes off your hit table is if you were swinging at an extremely high level mob. As you fight a mob of higher and higher level, you first lose your chance to land any normal hits, and you begin to lose your chance to crit until all of your hits are glancing blows, misses, and dodges. (If you have a ton of time, you could make a level one and swing at the 60, 70, 80 dummies, level up some, swing at them again, etc. you can also just read the first part of Glancing Blows.)

Assuming time is directly related to the number of swings you do, in a fight where you do one swing, there is a 24% chance it will be a glancing blow, 59% chance it will be a critical strike, etc and in a fight where you do 100,000 swings each individual swing has 24% chance to be a glancing blow, 59% chance to be a critical strike, etc. Nothing clicks, changes, or magical happens when you cross the "fine line" between a "short" and "long" fight, the probabilities will all be the same no matter the duration. In a short fight, the results/percentages will seem off like with a fight where you swing one time or ten time or a hundred times, in a long fight, you can expect that the probabilities will closely match the results/percentages simply because there are more swings so you have a higher "confidence" level.

This is exactly my point. Modeling every stat: miss, crit, glance, and dodge as a bell curve where each stat is an independant variable is wrong because it's a one roll system but the spreadsheets treat all of these stats as normally distributed, independant variables. It captures when the average crit rate goes over the crit cap but does not capture the effect that the variance of each stat has on the experienced average crit rate when approaching the crit cap. The crit cap is either everything or it's nothing according to the sheet and in actual play that's simply not 100% true.

This simply is not true. If it were a perfect bell curve it would be true but it is not a perfect bell curve. The chances of going x% higher are always smaller than going x% lower once the tooltip crit exceedes half the crit cap. Crit is not a symetrical bell curve around an average.

Fight duration is also a red herring, there are more constructive avenues of discussion. All fight lenght means is that the stats do not typically have time to converge to their averages before a fight is over, which is obvious. It does not reveal any underlying bias in which direction a stat's bell curve average as derived from tooltip is skewed.

The only thing that would even change is trinket/mongoose procs assuming all other buffs/debuffs are consistent. But it's already modeled in the combat spreadsheet, although not the mutilate one.