As a follow-up to my quick note on combat, here’s the idea of the Crit/Fumble Die. I had a discussion about crits with my players, and where I like having a cit-confirmation system, they almost all preferred a natural-20-always-crits system.
My preference stems from the math and what we end up modeling. With cit-confirmation, when you roll a 20 you get an auto-hit and then roll again to confirm, hitting the creatures AC again (doesn’t need to be another 20) means you get a critical hit. This means that there’s always a 5% chance to hit your opponent (auto-hit on a natural 20), and additionally that 5% of all your hits are going to be criticals. You’ll hit weaker enemies more, and thus get more crits on them, and tougher enemies will be hit less and have proportionally fewer crits. I’m building a similar system for fumbles because I like using them, but “always fumble on a natural 1” just adds in too much chaos (5% of all your swings are dismal failures).
Counter-wise, a 20-always-crits system means that 5% of all your swings (not your hits) will be critical hits. You will crit as much on strong enemies as you do on weak enemies, and if you have auto-hit on 20 as well you will *only* crit on tough enemies. That means 95% of the time you can’t touch the guy, and the other 5% you’re landing devastating blows. That just feels wrong.
But all my players see is that they roll a 20 and then I “rob” them of their crit when they fail to confirm. And I can see the logic in that. The Crit/Fumble die is my proposed solution, divorcing the “did I hit him” roll from the “did I crit him” roll. Each attack rolls 2d20, with one designated as the hit-die and one as the crit-die. If the hit-die beats the target’s AC, you hit and deal damage; if the hit-die is a natural 20, you auto-hit regardless of AC, but it has nothing to do with a critical strike. If you hit and the crit-die is a 20 (regardless of what the hit-die was), then it’s a critical strike. If you roll a 20 on the crit-die but miss with the hit-die, it was a good swing that just didn’t connect. And of course, if you miss with the hit-die and the crit-die is a 1, you just fumbled and something bad happens.
You’ll have 5% of your hits be critical, 5% of you misses will be fumbles, and 5% of your attacks with be auto-hits and auto-misses. But hopefully the perception that failing to confirm a crit “robs” the player of anything.
Jack's Toolbox 
I always hated confirming crits. One rule I did try was that whenever someone rolls a natural 20, they get to decide what happens (within reason); but when a natural 1 is rolled, their opponent decides what happens, something bad. Worked reasonably well.
Yeah, that’s the way many/most people I know do it, but it means that a full 10% of all rolls are exceptional and no circumstance changes that; the math just doesn’t sit well with me. It’s a matter of taste and if auto-crit/fumble (whatever that means in your game) is fun for you, go with it. I’m just looking for something more nuanced, I guess.
Which is to say, I think you should fumble more on hard rolls or against hard opponents, and crit more against easy rolls and easy opponents. Auto-crit/fumble doesn’t allow for that. Crit confirmation does, and it does it in a rather simple way.
Check this out: save versus paralysis to avoid a critical hit.
http://deltasdnd.blogspot.com/2012/07/critical-hits.html
Isn’t that a great idea? I much prefer it to the confirmation die. It serves the same purpose, while allowing the target a last chance to get out of the way. It also makes high level fighters (and monsters) slightly less susceptible to critical hits.
Your method does reduce the chance of a crit in all situations. If normally a to-hit roll succeeds on an 11 and crits on a 20, that’s a 10% chance that a successful roll will be a crit. Rolling again to hit when the player rolls a 20 reduces the chance by half. So your players are being robbed of some crits.
But I agree that a crit on every 20 makes no sense, because it makes crits more common on successful rolls for low-level characters and less common on successful rolls by high-level characters.
You can flatten the chance of a crit by rolling a “crit die” with the d20. Say a d10 is rolled too, and if it comes up a 1, AND the d20 roll hits, you get a crit. That’s a flat 10% of all successful rolls being crits, matching the average chance above. Conversely, a 1 on the d10 and a failed d20 roll could be a fumble.
You could also have players roll a different crit die depending on class or level to increase or decrease the chance of a crit. But then you can’t use the same die for fumbles or you’ll increase the chance of them as well.
Your method does reduce the chance of a crit in all situations. If normally a to-hit roll succeeds on an 11 and crits on a 20, that’s a 10% chance that a successful roll will be a crit. Rolling again to hit when the player rolls a 20 reduces the chance by half. So your players are being robbed of some crits.
I think we need to check what it is we’re discussing. Yes, confirming crits means you’ll have fewer crits: that’s a feature not a bug, and in fact is exactly what I’m aiming to do. Auto-crit on a 20 means that every roll has a 5% chance to be a devastating blow (a crit) regardless of their AC or yout to-hit. I don’t like that because it makes things too swingy. It’s only “robbing” them if you feel that auto-crit on 20 should be the norm.
Actually, you’re “crit die” concept is no different than mine, except for die size, and my crit die is no different from rolling to confirm. Your varies the ability of a class to confirm a crit, which is interesting, but I’d need to look at the math to really see how I feel. Should a Fighter (or rogue or whatever) really have a 10% crit chance? Should *anyone* have a 17% (d6) or 25% (d4) crit chance?
If only 5% of successful rolls are a crit, that does reduce the chance of a crit in all cases. Say I need an 11 on d20 to hit. A successful roll will also be a 20 10% of the time, so 10% of my successful rolls are crits. Of course, the better my chance to hit, the lower the percentage of hits that are also crits, which is weird.
For the crit die, I would say, fumble on a 1 and a miss, crit on a 19 or 20 and a hit. Possibly increasing it to an 18-20 or even 17-20 for higher level characters.
More concise than the above, but it’s all the same material. Actually, auto-crit has a flat 5% of rolls, confirm-crit has a flat 5% of hits, which means that if you want character build to matter (AC and to-hit), you want to confirm crits.
I think you’re off about the better chances/lower percentage bit, too. Suppose Monster M has and AC 11, so a base +0 to-hit has a 50% chance to strike. With auto-crit he will have 5% (p[20]) chance to crit and with confirm-crit he will have a 2.5% (p[20,11+]) chance to crit. If you have a +2 to hit that give you a 60% chance to hit, a 5% chance to auto-crit, and a 3% (p[20,9+]) chance to confirm a crit. At +5 to-hit you have a 75% chance to hit, a 5% chance to auto-crit and a 3.75% p[20,6+]) chance to confirm crit.
If you do the math of hits/crits, they all come out to be 20%, but that 20% of hits is bigger (in raw numbers) for the guy with +5 to hit than the guy with +0 to hit. (I’m actually a bit shocked that the math says 1 in 5 hits is a crit… someone check my math on this?)
It could be interesting to vary the range on the crit die; I would do so based on weapon choice rather than class or level, to keep in line with 3.X/Pathfinder as-written.