Based on the description of the attack roll calculation, it seems like accuracy/evasion are to be balanced by absolute delta (since the relative advantage of 100 accuracy to 80 evasion is equivalent to 20 accuracy to 0 evasion). This means in order to keep things roughly analogous at high level versus low level, the modifiers to accuracy/evasion probably should be a fixed value relative to class rather than level based. 3rd edition D&D is sort of a counterexample; their hit roll is basically the same system except with class-based level dependent scaling on the hit roll, which means the nature of the hit roll changes dramatically as level increases (at low levels, wizards can potentially contribute in combat with weapons; at high levels it becomes pointless since the absolute difference becomes so great). This also means if you want to keep the nature of the hit roll similar as level increases/game progresses, you would want relatively tight control on the variance of accuracy and evasion.
Yes, the plan is to balance it by absolute delta. Actually the problem in D&D3.5 is more that attack bonus grows far faster than AC. So a high AC becomes pretty meaningless. That's why character optimization usually goes for things like damage reduction or miss chance.
Note though that there's nothing inherently wrong with having the nature of the hit roll change throughout the game, though for balance purposes those who lose out in the longer run should get something else to compensate for it. In effect, having the hit roll be scaled with level/game progression should result in characters becoming more specialized in tasks besides the hit roll (e.g. spell progression in the D&D analogy).
Well there is the way D&D4 did it, with simply adding half the level to the hit roll. Of course for the most part that was equivalent to simply not increasing to hit and defences per level. A system using raises, like TDE or L5R do, would change that of course. Since the higher accuracy would allow more room for raises in the first place.
I agree with you that a constant class specific accuracy value, only modified by actions, weapons and effects/situation, would make a lot of sense. If we want to keep the dynamic the same that is. Of course that would mean loosing the advancement aspect that is so inherent in RPGs for this element. Which could be remedied by a level dependent modifier to both accuracy and evasion.
Of course that would also depend on the nature of the character advancement system we decide on. At this point I'm not sure myself what would be the best course of action and will probably change my mind during later playtests anyway. Some things look unfortunately only good on paper.
The second part of the mechanic seems to value reduction somewhat highly. The way hit/clean hit are valued seems to imply that in general, reduction values are going to be much smaller than the damage value (or the existence of the regular 'hit' won't mean much). This means in absolute terms reductions are going to have to scale slower than damage as you probably want net damage to scale roughly with health values (not doing so is perfectly reasonable, too, but it means again the early game has a different feel than the late game). The value of reduction also increases as you have more of it, making it tough to assign a value to (a la Armor Pen from old WoW). This formulation is inherently unstable; this is not necessarily a bad thing, it just means you have different regimes and you can't necessarily assign a single value to a statistic (finding corner cases where something becomes suddenly far more valuable than expected is part of the fun of playing non-generic systems for mechanics monkeys, after all).
Yes, reduction values should scale slower then damage values. A reduction of 0.5 average damage would indicate a really armoured foe. One of the design requirements I was given was "elemental" strength and weaknesses, so the reduction value will also be different for specific attacks. That way higher reduction values will also be possible. The next post will be about the way attributes work, where I wanted to explain some more about the system.
If you want reductions to scale more closely with damage (hardly a requirement, but may be a design goal), you probably want to make the effectiveness parameters tighter, depending on where you want to balance with respect to reduction. 0.5 - 1.5 doesn't really work well unless reduction is smaller than 0.5 of damage, and it can easily result in crits that are many times the amount of a regular hit (maybe this is a feature). 1.0 - 1.2ish would probably work better if you wanted to put damage and reduction on a more even footing. Moving crits out of the hit roll is also another option (more parameters generally means more options for exact tuning, though also more ways for things to break). Another possibility is to have the hit roll modulate the reduction rather than the damage; this can make damage reduction scaling more manageable.
Actually if we scale reduction to a maximum of 0.5 damage a critical hit will at most deal two times the damage a clean hit would. Which makes me wonder if "Hit" was the proper descriptor for that level, it might cause problems.
If the scale would be 1.0 - 1.2 we would have the problem again that the only choice for a bad hit roll would be an outright miss and a critical hit with only 20% damage increase would be a bit underwhelming. (Probably welcome if you end up on the receiving end.) Making critical hits a separate check would of course also be an option, it wouldn't change the dynamic we currently have though. And you could end with a mere "Hit" being a critical. The main idea was the accuracy wouldn't be just a binary choice, either you hit or you don't. In the first design accuracy had a direct effect on damage (basically we simply added the delta to the damage), but that simply made raising accuracy better then raising damage. Since the expected damage increased by one point for every point of accuracy plus the fact that you wouldn't risk a miss. Having two attributes that do the same thing doesn't make that much sense.
Moving the hit modifier to the reduction side of the equation is an interesting idea. It obviously breaks down for 0 reduction. Still it would create a very different dynamic, I'll definitely think a bit more about it.
Of course when we talk about average damage we should always keep in mind that there will be actions/weapons with a higher base damage but a lower expected damage. For example a heavy weapon that has 150% base damage, but takes twice as long to fire. Usually that weapon wouldn't be used, but against high reduction enemies it becomes a great option.