The formula seems quite accurate but I'm not sure it is complete.
I do have very strange values when I'm hitting for very low values.

I sampled some hits and crits at different levels. values from 1400sh to 1700sh normal and crits are in line or just different by 1 (so if it is rounded probabilistically we're in line).
3/8 vs 1/2 is a 25% nerf on rage generation than previously was (24% if you have endless rage).

Some tests that don't match are:
- unarmed ungeared with ER: hit (70 dmg) - rage gained 2, rage expected 3.87
- unarmed ungeared with ER: crit (140 dmg) - rage gained 4, rage expected 7.74

- 9dps 3.2s 2h axe ungeared no ER: hit (130-132 dmg) - rage gained 3, rage expected 5.35
- 9dps 3.2s 2h axe ungeared no ER: hit (135-137 dmg) - rage gained 4, rage expected 5.37
- 9dps 3.2s 2h axe ungeared no ER: crit (259-263 dmg) - rage gained 9, rage expected 10.68

- 9dps 3.2s 2h axe ungeared with ER: hit (139-143 dmg)- rage gained 4, rage expected 6.8
- 9dps 3.2s 2h axe ungeared with ER: crit (298-300 dmg)- rage gained 9, rage expected 13.7

Now consider that at 0 dmg done you should deal 3/8*fs so minimum rage gained is
- for unarmed 1.89s --> hit: 2.48 - crit: 4.96
- for 3.2s --> hit: 4.2 - crit: 8.4

If ER boosted the whole rage generated, an unarmed crit should have produced at least 6 rage.
Anyway if ER only boosts the damage part of the formula (7.5d/s) there are still some discrepancies and the values are never in the 1 range. At this point there are a few things that I would like to know, and these are:

1) How does ER boost rage?
2) Is there a damage treshold before which you don't produce any rage? ex: 3/8*(fs + 7.5*(d-X)/c)
3) Is rage rounded probabilistically?



For the issue of losing 1 rage I'm inclined to think it is due to latency and rage decay. At least I'm seeing this problem occur very often on first hit on a dummy while I didn't notice it happening while in combat. I also noticed this loss AND rage decay don't trigger a UNIT_RAGE event every second.

UNIT_RAGE is an event sent theorically every time your rage changes. If you use Bloodrage for ex, you get a UNIT_RAGE event every second and you gain 1 rage. When you exit combat and rage decay starts, even if your bar lose 1 rage point at a time you only get 1 UNIT_RAGE event every 2-3 points of decay. I guess the rage decay only triggers the event every nth seconds thus producing a sort on last tick as soon as you enter combat.

For what I have understood once in combat the rage decay doesn't stop, but you start gaining rage at the same rate.
If this is true this "rage gain" doesn't trigger at all a UNIT_RAGE event.

Good point. I had forgot about parry haste. However, I still think it uses base weapon speed because it would be hard to determine how to grant rage based on haste buffs falling off mid swing. I am thinking along the lines of WF totem, Flurry, haste procs or on use from trinkets, the Ret Pally dying or something similar. How would Blizz grant rage in those situations? Would they simply just use the current weapon speed when the attack lands or do they "pro-rate" the rage granted based on how long you were at a weapon speed during the swing before it lands? I really think Blizz took the easy way out and just decided to use base weapon speed to avoid overly complex rage generation equations.

Edit: Haste would still be relevant even if the rage equation used instantaneous weapon speed for the rage equation. It would be a dps increase, but it wouldn't be a rage generation stat as well.

Rage is at least tracked with one decimal, perhaps rounded after that.
Great Rage - Spell - World of Warcraft
Energize - Spell - World of Warcraft
There you can see that rage is at least tracked with 1/10th accuracy internally.

Ok, I have a few consideration to do after having reviewed rage generation.
One and probably the most important is about miss and expertise. While hit is still as valuable as before expertise is now imho much less important.
When you get dodges or parries you seem to generate the same rage as a normal hit. This devalues expertise contribution to rage generation by quite a lot.

Considering this, an arms warrior on a white dodged or parried swing can benefit of a "free OP" while for a dodged/parried move replaces a ms/slam/exe with a 2.5s OP that is a slight loss of dps.
The priority list would put OP first in the list so that you have more chances to not overwrite t4b on dodges.
Also OP itself can't be dodged nor parried.

My average damage consists of 20%sh white, 16%sh OP, 14%sh MS and DW, the rest is split among HS, Rend and WW.
Now:
- 16% of my damage isn't dependant on expertise at all.
- DW damage will lose a % due to "dodged crits" (ie possible dodged special hits that could have been crits; this suppose a 2 dice roll and chances should be around dodge%*crit%. An arms has 5.5% cap baseline (considering SoA) so with 40% crit you should have around 2% chances to get this kind of occurrence. You anyway replace that swing or move with an OP (if it doesn't overwrite t4b) that will most likely proc a DW stack.
- Rend dodged only lose around 1/12th due to delay and you waste 1s gcd (1.5s wasted, -0.5s bonus replacing next gcd with OP). I estimate this loss being around 2/3rd of a slam - (OP-slam), if you can avoid t4b overwrite.
- 20% of my damage if dodged can give me a white - (OP-slam) loss.

For every example consider OP-Slam as the benefit from both damage and gcd reduction.

In the end there is a loss but I'm not gonna try to estimate the entity of hit, since it's more tied to the number of occurrance in a given time rather than a fixed % value. However the loss is very lower than for a fury warrior that has 0 base expertise from talents and 85+% of his damage is affected by it.

Fury "just" devalues expertise for the loss of dps that can somehow be averaged (since I doubt fury warriors will stancedance to use OP) as (1-dodge)*damage max.

I highly doubt it could be worth gemming for expertise as arms, and I think there is a breakpoint in fury aswell.
At low gear STR could pull slightly ahead, at high gear you are already with Arp>STR and it gives better returns per point.
What you would lose is the reliability of interrupts, that in some fight is a must, but it's situational.