Accuracy - Counters the teargets Deflect chance. The game doesn't have miss mehcanics, this stat only related to deflect.
With Deflect chance capped at 50%, this means at 0 Accuracy at worst case (target has deflect capped), on average we will have 50% of our hits deflected. Maximum is no hits deflected, which means maching the targets deflect stat. Currently highest 90,000.

• Maximum effective cap is highet target deflect - 90,000. No effect above this.
• At 0 Accuracy, damage is reduced on average, at most, by 25% (75% of damage goes through). Reducing half the damage, half the time.
• Stat doesn't have diminishing returns on the conversion, always 1:1000.
• Gain to damage depends on targets deflect severity, currently 0.5 (50%).

$${{ \text{damageModifier} = \left(1 - \underbrace{ \left( \frac { nonNegative( \text{targetDeflect} - \text{Accuracy} ) }{100000} \right) }_{\text{capped to [0 - 0.5] (0 - 50%) } } \cdot \text{targetDeflectSeverity} \right) }}$$ For the stat calculation in excel/software:

= (1 - min(max((targetDeflect - accuracy) / 100000, 0), 0.5) * targetDeflectSeverity)

Damage modifier, at different Accuracy values vs  

 

Target Deflect Severity

The no effect gap when we above the cap and do not counter enough, and the no further effect flatline when all of the target deflect was countered.

Important to notice, that at worst we loose 25% damage, as compared to ArPs 50%, and here too wee need to invest 40,000 stats before we see any return.

Lets look at the damage increase graph for adding given number of Accuracy.

vs  

Target Defense

 

Target Deflect Severity

Accuracy behaves exactly like ArP just with much less gain per stat point.