Read the source and upvote this guy for his work:
http://www.reddit.co...chanics_enmity/
Introduction
Understanding the mechanics of melee damage is the most sought after knowledge among every endgame LS...
Methods
The first step in assessing the damage mechanics with the new battle algorithms involved establishing which type of number generation system FFXIV:ARR used. In FFXI they used a normal distribution (think bell curve) where the majority of damage followed a midpoint with slight variations of increased and decreased damage. In FFXIV 1.20, we established that there existed a random nonnormal distribution. This means that the damage numbers are drawn at random with equal probability from a predetermined set. The predetermined set included an 8% variation from a midpoint value. For example, I am averaging 100 damage. The minimum damage I will do is 92. The maximum damage I will do is 108. The odds of doing damage for any number between 92 and 108 is equal. FFXIV:ARR follows the FFXIV 1.0 system. The only change I noticed is that they reduced the variation to +/ 5% instead of 8%. It is because of the equal probability and fixed set that allows my testing method to work.
The method I used to test the stats is similar to the method I used in my 1.20 testing. I first established a base statline while naked. Based off of the beta testing manual information, only Strength (STR) and Determination (DTR) affect the damage output of PLD, WAR, MNK, and DRG. To test the influence of these stats, I held one stat constant while I increased the others. I did not hold any other stat constant, and I believe this did not alter the results (I will explain later).
For WAR, I tested on a Level 50 Water Sprite. For DRG, I tested on 2 mobs, a Level 50 Water Sprite for only dDMG/dSTR, and a Level 40 Earth Sprite for dDMG/dSTR and dDMG/dDTR.
Results
 Dragoon Testing with Mogfork
Influence of Strength on AutoAttack (AA), True Thrust (150 Pot), and Vorpal Thrust (100 Pot) on a Level 50 Water Sprite
http://i.imgur.com/IZTCFl3.png * STR ranged from 268378. The slope of the line indicates how many points of damage are added per point of strength added (dDMG/dSTR). The Rsquared value indicates the linearity and how well a linear line "fits" the data.
Raw Data for DRG on Level 50 Water Sprite
http://i.imgur.com/PDm6MUl.png
Influence of Determination and Strength on AutoAttack(AA) on a Level 40 Earth Sprite http://i.imgur.com/vVkDYJe.png *DTR ranged from 202266. STR ranged from 302344.
 Warrior Testing with Moogle Axe
Influence of Strength on AutoAttack (AA), Maim Buff with AA, Maim Buff/Storm's Eye Debuff AA, Heavy Swing (150 Pot), and Skull Sunder (100 Pot) on a Level 50 Water Sprite http://i.imgur.com/wEBgIIW.png *STR ranged from 254354. The +10% indicates the 10% damage buff from Maim. The +10/10% buff indicates the +10% Maim buff and the 10% Slashing Debuff.
Raw Data for WAR on a Level 50 Water Sprite http://i.imgur.com/nvdVu8n.png *Maim % refers to the buff on Autoattack, not the weaponskill damage. Similarly for Maim+Storm's Eye %.
Discussion
First lets begin by talking about the linear nature of all of the graphs. This is an important concept to grasp. Stat scaling in FFXIV:ARR is linear. That implies that your return on investment will be equal at any given stat value. There is no diminishing return on stat investment until you hit the cap. Also as I mentioned before, I only held STR and DTR constant. Since I had such a high degree of linearity with all the other stats varying greatly, we can safely assume that all other stats have no influence on damage output for autoattack and weaponskills.
Next let's discuss the difference in efficiency between Strength and Determination. As you can see from the Dragoon Testing graph on the Level 40 Earth Sprite, Strength was nearly twice as efficient in increasing damage pointperpoint. While this wasn't tested on a higher level mob, I would expect the results to translate fairly well, although I will retest this on a Level 50 mob this weekend. This makes sense considering Determination influences every class's damage and healing potency. So to make it on par with or greater than STR/DEX/INT would be unbalanced.
 Take home points: Strength has nearly twice the return on investment as compared to Determination.
Next let's take a look at the raw data charts. Looking at the difference in the Average Damage across AutoAttack, a 150 Potency WS, and a 100 Potency WS, we can notice several trends. First off, Autoattack is nearly equivalent to the 100 Potency WS. This should translate to Relic weapons and future content weapons. Also notice the difference in Average Damage for the 150 potency WS and the 100 potency WS for both WAR and DRG. For both jobs the ratio of WS damage for the 150pot and 100pot WSs is ~1.5. This implies that potency is a linear scaling modifer. A 200 potency WS will do EXACTLY twice the damage as a 100 pot weaponskill. There is no exponential scaling.
 Take home points: Autoattack is roughly equivalent to 100 potency for endgame gear (WAR ~110115, DRG ~95100). Weaponskill potency scaling is linear.
dLVL
dLVL is an important concept to grasp. The term simply refers to the "difference in level" between a user and a target. I use the terms dSTR and dDMG a lot. All the little "d" means is "difference between 2 values (subtraction)". You might also see the Greek symbol delta to refer to this.
dLVL in 1.0 played the biggest role in determining damage. And many of you are intuitively familiar with this concept. Consider Thundara doing 6k+ on a L1 Marmot as to doing only 500 on a L50 Amal' Ja. dLVL afffected every aspect of the game from defense, crit damage, to the influence of every WS and spell. It also made stat testing a pain as each difference in level produced entirely different results when testing. However, it appears that FFXIV:ARR will NOT have dLVL be a determinant in stat algorithms. Here is the evidence as to why we think this:
You can no longer oneshot many low level mobs.
If dLVL played any type of significant role in damage calculations, your damage output on a lvl 8 mob as compared to a level 50 mob should be very noticeable. The very fact that you can no longer oneshot these lower level mobs is strong evidence that dLVL is no longer in the game and that damage instead is based solely on attack/defense modifiers.
My testing on L40 and L50 Elemental Sprites
I only did brief stat testing, but what I found on L40 and L50 elemental sprites was quite surprising. Not only are the rate of return of dDMG/dSTR almost equal, but the actual damage numbers are incredibly close. For 323 STR and 202 DTR with Mogfork, I did 89.5 dmg on average to a L40 Earth Sprite. For 316 STR and 202 DTR with Mogfork, I did 88.5 dmg on average to a L50 Water Sprite. You wouldn't find numbers this close with a Lvl 49 and Lvl 50 mob in 1.0, and we're talking a dLVL of 10 here!
Why is this important? Consider the Critical Damage stat. In 1.0, this stat became slightly useless because dLVL often caused crit damage to floor (opposite of cap) at 15%. Your crit rate also floored. This made investing in the stats have an incredibly painful return on investment in 1.0. However with dLVL no longer being a strong factor, Critical Damage builds are now a real possibility. I don't have many hard numbers on this, but luckily I did record some of the Critical Damage values. On an L40 Earth Sprite I did 85.5 damage on average, while I Crit'd for 130 damage on average. This is roughly a 1.5x return. Now it is yet to be seen how this translates to higher level mobs and there could be a "hidden" stat that affects Critical Damage taken. One piece of evidence for this is the GLA ability awareness that reduces Critical Damage taken by 25%. This could either be increasing a stat or simply be an external modifier (if attack = crit, then dmg = attack x 0.75. Or something like that). The point remains that Critical Damage builds could be very powerful builds. We will have to do some more testing this weekend.
Damage Formula
Much of what we discussed were hypotheticals, however we already have a lot of data that could back our assertions. First let's talk about
 AutoAttack Potency
As I have mentioned several times before, autoattack seemed to have around the 100 potency mark for DRG and WAR. Now we realized that there might be a very simple determinant for this. Consider the autoattack stat and the physical damage stat on weapons. Take for example the Mogaxe and the Mogfork
If you divide the autoattack value by the physical damage value you get 40.45/41 or 98.6% for the DRG Mogfork and 45.92/41 or 112% for the WAR Mogaxe. Compare this to the hard numbers I got from my WAR and DRG testing. For DRG's Mogfork I got anywhere from 9799% and for the WAR I got anywhere from 111.8113% when I compared autoattack damage to a 100 potency WS. This seems to be too close to be a coincidence. Thus I think we are at the point where our current belief is that AutoAttack potency is equal to the AutoAttack stat on weapons divided by the Physical Damage, then multiplied by 100.
 The important modifiers
When thinking about a formula it is important to keep in mind which modifiers are contant, which are linear, and which are external to an equation. Here are some of our current thoughts.
Potency is a linear modifier. 200 potency will do 2x 100 potency in terms of damage
Stats are linear modifiers. They follow a slope equation of dDMG/dSTAT
The slope for stats is a nonconstant modifier. It appears to be influenced by Physical/Magical Damage.
The +10% damage type of stats appear to be external modifiers. They multiply the final damage calculation
So we can come up with
Damage = [(Potency * some scaling modifer(?)) + (STR * dDMG/dSTR) + (DTR * dDMG/dDTR)] * External Modifiers
(I believe the scaling modifier is determined by the weapon's Physical Damage) (Where dDMG/dSTR and dDMG/dDTR is dependent on Physical Damage of the weapon)
What we really need to see is how Potency scales with Physical Damage on weapons. This should give us a much better idea on the formula. Now we also need to understand that there is no way to influence the potency of a WS outside of choosing a different weapon; however, depending on the influence of the weapon's Physical Damage stat, there could arise a situation where a P DMG 100 weapon with 20 STR is worse than a P DMG 95 weapon with +50 STR. It really depends on the scaling. *Continuing to write. Just want to submit so I don't lose anything~
The Importance of Stats
This is more of a subjective topic, but it is important nonetheless. Compare my data for DRG while fighting a L50 Water Sprite:

Naked at 267STR, 202DTR = 77.5 dmg AutoAttack

Geared 378 STR, 202 DTR = 102.5 dmg AutoAttack
That is a 102.5/77.5 or 32% increase by just adding 111 STR. I know that I am undergeared compared to other players and I think Miko said he could get his STR up to ~410. This would come out as a 42% increase compared to naked gear. Now keep translating this with materia and future dungeon drops and I believe we are going to start seeing up to 500 STR (assuming it doesn't cap). Also keep in mind that I head DTR to its base value of 202 for my testing. This stat will also be increasing. All in all, I can imagine a fully geared player doing somewhere between 1.501.75x more damage than a naked player with the same weapon. The reason this turns out to be the case is because SE decided to scale down damage to smaller values and the use of linear stat modifiers. Adding 1 more point of damage when your base is 100 is much more potent than adding 1 damage when your base is 500.
I think this is a reflection on the complaint that stats were "meaningless" in 1.0 (whether that is a true statement or not). So SE decided to go with a system where players fully geared are going to be doing 1.51.75x more damage (and with faster GCD) making the difference much more appreciable. While there is still a lot of speculation in what I have wrote, I believe I can justify it in my mind considering what SE's goals were towards altering the damage/stats algorithms.
Final Note: Please don't take the damage formula I proposed as 100% correct. There are tons of things I left out for simplicity such as defense of the enemy mob, elemental resistances, slashing/piercing/blunt resistance, etc. This is just a more simple formula for things that we as players can change to increase our damage.
Buffs
Shield Oath is a PLD buff that grants +enmity to all actions. This is preliminary, but it appears that Shield Oath doubles the enmity generated from each action.
2 Flashes (no Shield Oath) = 1 Flash (Shield Oath)
Buffs themselves give hate.
Roughly 5.5 uses of Sword Oath will equal the hate generated from Flash (without Shield Oath) Roughly 11 uses of Sword Oath will equal the hate generated from Flash (with Shield Oath)
Weaponskills
The +enmity attribute present on weaponskills such as Skull Sunder and Savage Blade appear to be multipliers of hate damage and not just a strict + X amount of hate.
Savage Blade = ~2x Riot Blade (no combo, no shield oath) with Weathered Gladius
Savage Blade = ~2x Riot Blade (no combo, no shield oath) with Moogle Blade.
(note both Savage Blade and Riot Blade are 100 potency weaponskills).
Damage over Time (DoT) abilities continue to generate hate with every tick. A 20 potency 30s DoT will generate 200 potency worth of hate over the 30 seconds and not all at once.
Abilities
Provoke is a GLA ability that instantly ties you with the player who has the highest enmity and places aggro on you. As long as the person you are tied with does no further actions, he will forever remain below you on the enmity list. The skill essentially "catches you up"
Example: Aki the BLM has 5000000 points of enmity on Ifrit
Miko the PLD died and got raised. He currently only has 1000 points of Ifrit. Miko uses Provoke. Miko now has 5000001 points of enmity and is at the top of the hate list. (Note: This is permanent hate)
Cure Hate
Overcuring does not appear to generate hate. Also
1 Flash (no Shield Oath) = ~750 Cure (Aque healed on WHM 300/heal, took 3 heals to steal hate)
1 Flash (Sheld Oath) = ~1500 Cure (Aque healed about 1500 on 5 cures)
Skill Speed and the Special 341 Stat
Based on this I would expect the formula to actually be this:
Global Cooldown = 2.5  0.01 x roundup(speed/10)
But would obviously have to verify the rounding part. Easiest way to do this is equip +1 skill speed and see if the GCD changes to 2.49. If so, can fairly safely assume it's a rounding change. This would be somewhat important to minmaxing as you would attempt to always get a value where the 1's digit was 1 (no point in say getting 9 skillspeed). This formula may only apply to R50 though.
The testing on greased lightning would conclude:  Greased Lightnining is flat 5% current GCD after gear enhancement (stacking up to 15%)  not sure on rounding error but didn't take that deep a look
Interesting question here is for spell speed, does this decrease the casting speed similarly but subtracting 0.01 second increments for every 10 speed or is there some proportional effect. Like if you have 200 spell skill and compare Raise (8 sec) vs Cure (2 sec), do you shave more seconds off the casting time of Raise than Cure b/c it's longer casting time? Very easy to look in game next time.
The other obvious question here is is there a cap to this effect. That may be hard to show though.
Based on this person's testing, skill speed is obviously shaving 0.01 second increments in a linear fashion and looks to be a static return; however, it is technically an increasing return only noticeable at extremely high amounts of speed. Since for every 10 more you add, you are shaving 0.01 off a progressively smaller base number. For instance shaving 0.01 off 2.5 is a 2.5/2.49 x100%  100% = 0.4% increase in efficiency. As you increase to say 500 skill speed and the GCD is 2.0, adding another 10 is now a 2.0/1.9 x100%100%= 0.5% increase in efficiency. To use an extreme example, say you have 2480 skill speed w/o presence of a cap and your GCD is 0.02 seconds. Adding another 10 speed brings your GCD to 0.01 seconds, or a 50% increase in efficiency. Here's a graphical version that illustrates this well.
Essentially this is an increasing return mechanic, but might as well be linear. The increasing return is only seen at extreme amounts of speed. You're likely to hit a supposed gap long before this.
I was talking to Valk last night about the significance of the Spell Speed, Skill Speed, Accuracy, Critical Hit Rating, and Parry skills all having the same basic value (with a 341 value at 50). I ended up just looking up the beta Lodestone and found a character of every level 150 documenting what the base amount of this was. I then graphed it and got this:
This had a very good best fit with polynomial even just eyeballing it. The interesting thing here is that if you just take the best fit curve excel gives you and plug in level 99, you end up with 1037.97. This is actually really close to 1024 which holds the significance of being 2^{10} in coding. I'm not sure if this was coincidence or not. Anyways, I fiddled with the curve a bit to try to account for rounding error a bit more and ultimately got the equation of:
Base Statistic = rounddown{0.083625x[Level]^{2} + 1.5x[Level] + 55.4}
With this, you can predict how the statistic will scale at future levels:
This is pretty much it for the actual testing, but what's really potentially interesting here is the fact that so many viewable stats have this same number. What I really think is going on here is that this 341 value at R50 is the new representation of dLVL in 2.0. The only reason the game is allowing you to view it in these 5 instances is that these 5 stats can actually be modified by gear; however, this value being tied to your level likely plays an additional role in a number of calculations.
Looking at critical hit rating, there has to be a critical hit defense stat to compare things to. This is supported by the fact that awful GLD ability that increases critical hit defense exists. We were asking well if it exists what is it? It's likely just this basic value tied to your level (aka at R50 341). So if you are a naked R50 w/ critical hit rating 341 and fight an R50 mob, they cancel and the critical hit % is some baseline value. You fight an R1 mob with level stat 56, you get a huge critical hit bonus of 285.
Going with the idea of accuracy and evasion, Soge and Will were saying how in WoW, every mob of some level has the exact same evasion. In fact there's like a table of accuracy benchmarks you have to hit to ensure 100% accuracy and you just slot enough accuracy to hit the necessary level of the mob you're fighting and then ignore accuracy after you hit it. Compare this to XI where you can have 2 enemies of the same level with wildly different evasion rates. If the 2.0 system goes with evasion as a stat that cannot be modified and is simply tied to the level, then a similar table of accuracy benchmarks could technically be made eventually. If true, this system would be far more simplistic than FFXI, and essentially mimic WoW.
Basically to test this, would need to fight an R50 mob at +0 accuracy and collect a large # of swings to get a % hit rate. This establishes a set hit % when accuracy and a supposedly hidden evasion stat tied solely to your level are the exact same. We could then repeat with a lower level mob, say randomly an R45 w/ level stat 292 (49 less). Fight different level mobs and keep repeating and eventually you will develop a curve that shows how changing difference in ACCbase stat affects the hit rate. You could alternatively increase your own accuracy and keep the R50 mob as well.
Spell and skill speed would be more interesting. It appeared odd that your GCD was 2.5 regardless of level, yet your skill and spell speed kept increasing. What I suspect is happening is when you're talking about criticals and accuracy, you are comparing to mob level, but for these 2 stats, you're actually just comparing it to your own. So at level 50, your skill speed is 341, and your level statistic is also 341. So you end up with 341341 = 0 meaning you're stuck at the basic GCD of 2.5 seconds. This would imply though, that the rate or return for spell speed / skill speed is static throughout all levels. This would require validation with a nonR50 character to see if the rate of return is greater than 10 speed for 0.01 seconds. If it actually is greater, then the balance check may involve division.