D20 Variance in practice

[Haruspex_Pariah]

After over a decade of lurking, reading D&D-related material, and playing D&D video games, I've finally joined an actual gaming group what plays D&D. And I will say there are a lot of things you just don't get from studying theoreticals.

Like how a good roll here and a bad roll there can swing the course of events in ways that can be really fun but also maybe a tad frustrating. A (admittedly low level) party member catching a crit on the first round and dropping immediately, as an example. On the flip side, a party member rolling a 20 on a Strength check and busting down a door so we can escape a danger zone. Certainly these are thrilling, memorable moments, but the results-oriented gamer in me feels a little...put off.

When theorycrafting it's fairly obvious that a +5 modifier is better than a +0 one. Probability favours the one with the higher modifier. That's just basic math. But in a situation where a small number of D20 rolls are made, it sometimes happens that the character with the +0 does better. Because that's how randomness works.

I suppose for gamers who mostly play video games, accepting true randomness is difficult. We take good rolls for granted and act like a bad roll is an offense against us, sometimes. Even though we knew going into it that we were playing a game with RNG-determined outcomes.

Just my thoughts on things. I am enjoying D&D so far, but I am surprised at how much I have to alter my mindset with regards to the gameplay side of it. Rolls in Warhammer 40k, as a comparison, have much lower variance.

[Hytheter]

It's worth noting that the variance of the dice matters not in a vacuum; it's the relative size of the applicable modifiers that really matters. The impact of the d20 is very large when a good modifier is +5 and a bad one is +0, but much smaller when the good modifier is +10, +20, +30.

In D&D the modifiers scale over time so the impact of the d20 will lessen, though the specific edition you're playing will affect the extent of that.

[Haruspex_Pariah]

It's worth noting that the variance of the dice matters not in a vacuum; it's the relative size of the applicable modifiers that really matters. The impact of the d20 is very large when a good modifier is +5 and a bad one is +0, but much smaller when the good modifier is +10, +20, +30.

In D&D the modifiers scale over time so the impact of the d20 will lessen, though the specific edition you're playing will affect the extent of that.

That makes sense. Admittedly I just hit level 5, in my first ever campaign, so I may not have a comprehensive overview of the game as a whole.

Playing D&D 5e, if that matters. Compared to my understanding of 3e, it does seem like the overall range of modifiers is a lot smaller in 5th. A lot of the flat numerical bonuses that the CO sorts hunted for in 3rd seem to be cut out in 5th. I see a lot of advantage and proficiencies, as well as non-numerical bonuses. Ability scores seem harder to boost as well. Correct me if I'm wrong on that.

[Hytheter]

Yeah, the numbers are a lot flatter in 5e. You typically only add your proficiency (+2 to +6) and an ability modifier (up to +5) so your typical roll won't be any higher than +11, though there are a few other modifiers which can apply. Contrast with 3rd edition where your skills and attack bonus can go up every single level and a roll in something you're specialised in likely has tons of other numerical bonuses as well.

[Crake]

3.5 dnd had a bell curve roll variant, where you roll 3d6 instead of 1d20. You have a max of 18 instead or 20, but also a minimum of 3 instead of 1, and an average roll of 10.5 which is actually the same average as a d20. The crit ranges were also sorted out differently, based on similar probabilities. A 16-18 matched a natural 20, 15-18 matched 19-20, 14-18 matched both 18-20 and 17-20 , and 13-18 matched the 15-20 range, though auto fails/successes still only happened on natural 3s/18s.

If you're looking for a lower variance roll system, might be worth trying something like that instead of a flat d20 roll.

[Anonymouswizard]

2d6 and 3d6 are some of the note common basic rolls in the industry for a similar reason (the other really common one is d%). Having more rolls that produce average results rather than ever results tends to make smaller modifiers matter.

Although it's also not modifier size that matters as much as modifier variance and weighting. 3.X had PC offensive and defensive modifiers weighted heavily in favour of offence, which makes the d20 roll relatively meaningless at higher levels. 5e keeps the end have modifiers smaller but also reduces some starting game modifiers (although not the attack or defence ones) which makes the d20 matter a lot more.

Compare this to something like Vortex (used in the Doctor Who game and Rocket Age). Vortex uses 2d6, but also has relatively larger modifiers and now meaningful variance from the get go. Normal characters are rolling at maybe +3 or +4, specialists in their field at +8 or +9, which while a similar numerical difference to 5e is also much larger compared to the values you can roll (which are also weighted towards 7). Being good at something matters a lot more compares to lie level D&D (and about as much at end game).

[warty goblin]

It's worth noting that the variance of the dice matters not in a vacuum; it's the relative size of the applicable modifiers that really matters. The impact of the d20 is very large when a good modifier is +5 and a bad one is +0, but much smaller when the good modifier is +10, +20, +30.

In D&D the modifiers scale over time so the impact of the d20 will lessen, though the specific edition you're playing will affect the extent of that.

It's not even really the modifiers, it's the difference between the DC and the modifier that matters. If X is the d20 roll, m is the modifier, and d the DC, then the math looks like

Pr(success) = Pr(X + m >= d) = Pr( X >= d - m) = 1 - Pr(X < d - m) = 1 - Pr(X <= d - m -1)

This particular equation holds for any game based on comparing a result to a modifier (i.e. not dice pool systems). What changes is the probability function Pr. If you think of modifying the success threshold t = d - m - 1, then you're simply stepping through the cumulative distribution function (CDF)* of a discrete distribution. All CDFS start at zero (here, guaranteed success, you cannot roll under the threshold) to 1 (here guaranteed failure, you always roll under the threshold), the only things that vary are how you move between those extremes. The d20 moves in 5% steps, so as long as the threshold is less than 20, each +1 to your threshold decreases your success probability by 5%. 2d6 is a bit more interesting, increasing the threshold from 2 to 3 moves you from a success probability of 35/36 to 33/36, while a threshold of 6 has a success probability of 21/36, and 7 has 15/36, i.e. that one point increase in DC/1 point decrease in skill costs you a whole 1/6 probability of succeeding in the center, while the 2 - 3 change only costs you 1/18. The effective change when near the center of the distribution is 3 times that at the edge, which is a very different behavior indeed from the d20.

In character terms if Alice needs to roll a 7+ to stab the cultist under 2d6, she'll succeed 15/36 = 5/12, or just under half the time, while if Bob needs a 6+, he will enjoy the benefits of cultist perforation 7/12 of the time. If the cultist chugs a potion of armor so Alice only succeeds on a roll of 12, and Bob on a roll of 11+, Alice will hit 1/36 of the time, and Bob will hit 3/36 of the time. This is what people mean when they say that more skilled characters are more consistent under 2d6 than 1d20. The upside is that, compared to d20, increased skill means more, i.e. Bob's 1 point advantage over Alice is much more substantial under 2d6 than 1d20. The downside is that increasing the difficulty slightly can massively impact the probability of character success; in particularly you're much less likely to do hard/unlikely things in general.

3d6 behaves very similarly to 2d6, but with more probability near the middle of the curve, and I can't run the probabilities in my head anymore.

*If you want to be completely technical, 1 - Pr(X < x) is the survival function not the CDF. If you are familiar enough with this stuff to care, you're also familiar enough to not care.

[Easy e]

Yeah, d20's are swingy as feth. You have a 5% chance of getting any particular number on any given roll.
Even a +% is only changing your odds by 25%, BUT it gives you a 100% of getting 6+.

It can make D&D kind of annoying as character builds that are "optimized" at lower levels to do certain things can still blow it at those things. That is why Advantage and Disadvantage are so critical to 5E.

Games with dice pools really reduce this issue and allow you to focus on success more than rolling.

[Anonymouswizard]

Games with dice pools really reduce this issue and allow you to focus on success more than rolling.

Honestly, past a certain point mitigating the chance of failure doesn't appeal to me.

Personally I've grown to like percentiles, while they're even more swingy than d20s most people seem to be able to grasp how the probability works.

Plus they have a really cool feature compared to most non dice pool in that doubles appear every 11 steps, make them special/critical results and suddenly more skilled characters are less likely to get catastrophic results and more likely to get amazing ones.

[warty goblin]

Honestly, past a certain point mitigating the chance of failure doesn't appeal to me.

Personally I've grown to like percentiles, while they're even more swingy than d20s most people seem to be able to grasp how the probability works.

Plus they have a really cool feature compared to most non dice pool in that doubles appear every 11 steps, make them special/critical results and suddenly more skilled characters are less likely to get catastrophic results and more likely to get amazing ones.

They're not really more swingy, they're just more granular. The underlying distribution is still discrete uniform, so the behavior of the CDF is the same, and that entirely dictates the success probability function.

[Pauly]

D&D in particular has a huge amount of variance for low level characters. Other games such as Warhammer FRP have even more variance.

Generally speaking, regardless of system, as your characters progress, learn skills, acquire magic items and so on the game becomes much less variable. High level characters can mitigate/practically eliminate variance.

A lot of people consider mid level characters (level 5 t0 10 for D&D) to be the sweet spot. You have enough variance for random chance to be a factor, but are not bored by too many foregone conclusions, nor can you be wrecked by one or two bad rolls.

[Cluedrew]

In all my home-brew work I don't think I have ever so much as considered using a d20. At that point you are basically dealing with a percentile system in 5% groups. I usually use dice pools or 2d6 plus stat. Not only because the d6 have a bigger impact on a +1 (and if you need a +2 or +3 to make a notable impact, why use the larger die?) but then you put it on a bell curve so that the middle results are more common. Which you can do by adjusting numbers, but it happening naturally. It also builds in diminishing returns.

Or you can do something like Powered by the Apocalypse or Blades in the Dark and fold in multiple levels of success into a single role.

Which is to say: Yeah 1d20+stat vs. a target number is only a step up from rolling a die in my opinion.

[Anonymouswizard]

In all my home-brew work I don't think I have ever so much as considered using a d20. At that point you are basically dealing with a percentile system in 5% groups. I usually use dice pools or 2d6 plus stat. Not only because the d6 have a bigger impact on a +1 (and if you need a +2 or +3 to make a notable impact, why use the larger die?) but then you put it on a bell curve so that the middle results are more common. Which you can do by adjusting numbers, but it happening naturally. It also builds in diminishing returns.

Or you can do something like Powered by the Apocalypse or Blades in the Dark and fold in multiple levels of success into a single role.

Which is to say: Yeah 1d20+stat vs. a target number is only a step up from rolling a die in my opinion.

I've done stats as die types before, it's interesting. Not had the chance to playtests the system, but seeing as how most rolls are attribute die+skill die versus target number it's down some potential for fun (there's special results on doubles, generally bad things).

[icefractal]

While it's not the best fit for all games, I like the transparency of the d20. Something is DC X and you have a +Y in it? Easy to see what your odds are. How much is a +2 bonus? It'll change the result on 10% of rolls, simple.

Bell curve systems need a look-up table unless you've got it memorized, and systems with "interesting" dice mechanics often make the odds pretty opaque. I think sometimes the intent is "players shouldn't care about the odds", but that doesn't work IMO - players who don't care wouldn't have cared with a simpler system either, and players who do care get extra friction.

[Vahnavoi]

Yeah, trying to one-up people who are obsessed with calculating the odds by making sure the odds are only calculable by the people obsessed with calculating them is a classic game design fib.

Generally speaking, you can turn any independent probability function into % values (and dependent ones too, the values just have to be recalculated turn-by-turn) if you know the relevant math. How the original probability function is set up can make it easy or very hard, but there's little reason to use convoluted dice mechanics that make it very hard unless you think Yathzee is the greatest game ever or like rolling dice for sake of rolling dice.

[MoiMagnus]

While it's not the best fit for all games, I like the transparency of the d20. Something is DC X and you have a +Y in it? Easy to see what your odds are. How much is a +2 bonus? It'll change the result on 10% of rolls, simple.

Bell curve systems need a look-up table unless you've got it memorized, and systems with "interesting" dice mechanics often make the odds pretty opaque. I think sometimes the intent is "players shouldn't care about the odds", but that doesn't work IMO - players who don't care wouldn't have cared with a simpler system either, and players who do care get extra friction.

Simple formula have a clear advantage as they facilitate communication between the game and the player.
And that's why tactical video-game (fire emblem, XCOM) tend to have in appearance very simple mechanics, which allow for players to have fun optimising numbers here and there.

However, they still tend to use more complex curves internally by the subtle but practical method of "lying to the player". E.g, a lot of fire emblem use the 2RN system where instead of making one percentile roll, the game makes two of them and takes the average, which makes attacks with high probability to land quadratically more likely to hit. (Later fire emblems use a system much more complex mathematically). Note that contrary to fudging dice, this lie is "fair" in the sense that it is applied to both side equally.

I believe that some TTRPGs try to rely on opaque systems for a similar reasons: they want to build an illusion of how the odds actually work, different from the actual odds. So that the players can optimise using those illusionary odds (which are tailored to be "interesting to optimise" or "more intuitive to non-mathematicians"), while the system runs with different odds (which are tailored to be "balanced" or "less frustrating"). IME, it's much less effective in TTRPGs than in videogames.

[Vahnavoi]

It's much less effective in tabletop games because some poor human has to read and run the math.

[kyoryu]

I think framing helps, and how you handle skill rolls.

I mean, usually, the issue is less absolute variance and more "big strong guy fails to open the door but weak wizard succeeds".

So, what I like to do is a few things:

1. Don't roll for things that aren't possible.
2. Don't roll for things that are obviously going to succeed
3. For things in the middle, assume that if something's possible, then it will be done eventually, given infinite time and resources
4. Identify what constraint is in play - make the roll into "can you do this before the bad thing happens?" rather than "can you do it?"
5. Alternately, let the roll also inform things about the situation - maybe the lock can't be picked because it's rusted through.

- either way, a straight "let the wizard try" doesn't work, because either we've determined that the door just can't be moved, or you have to deal with The Bad Thing before trying another roll

6. In other cases, turn it into a group roll, using whatever rules your system has for groups.

So, for the door scenario, we'll assume the door is heavy, and is theoretically movable by both the barbarian and the wizard. If we don't want to turn it into a group effort, figure out what will stop the barbarian from trying to open the door... maybe guards might appear? Now a failure doesn't mean that the barbarian "can't", it means he couldn't in time. Okay, so now you can't just have the wizard try, you have to deal with the guards coming... and even then it probably still makes more sense to have the barbarian try again, or figure out a way to get the whole group to work on it together.

While the variance of a d20 is high, and causes these to occur more often, I do think that they're really just inherent problems with how rolls are often framed, and can present in any system.

[Easy e]

I don't make people roll for things if them failing makes no sense or makes the game less interesting.

For example, if the group can not get through a doorway, and therefore can not progress it is not interesting if they fail. Therefore, they should not roll to be able to open the door.

They may have to roll, but it is not to get through the door, it is how long it takes them to get through it.

So, the big guy will eventually batter it down, just does he do it in one go, or does it take him a few minutes? The thief can pick the lock, but does it take a minute or one hour? Gandalf can figure out the riddle, but does he do it immediately or does it take him an hour while everyone sits around the lake with the tentacle monster in it?

[kyoryu]

They may have to roll, but it is not to get through the door, it is how long it takes them to get through it.

And not just time, but other things as well... did you pick the lock before you broke your picks? Did you climb the cliff before the bad guy got away?

There's lots of ways of having stakes on a roll that can make rolling for it interesting.

[Psyren]

3.5 dnd had a bell curve roll variant, where you roll 3d6 instead of 1d20. You have a max of 18 instead or 20, but also a minimum of 3 instead of 1, and an average roll of 10.5 which is actually the same average as a d20. The crit ranges were also sorted out differently, based on similar probabilities. A 16-18 matched a natural 20, 15-18 matched 19-20, 14-18 matched both 18-20 and 17-20 , and 13-18 matched the 15-20 range, though auto fails/successes still only happened on natural 3s/18s.

If you're looking for a lower variance roll system, might be worth trying something like that instead of a flat d20 roll.

Came to suggest this, if a uniform distribution is too swingy then try a bell curve. Dragon Age tabletop uses this as the default and it worked really well, especially with the "stunt" system to replace crits.

[vasilidor]

when it comes to picking a lock, they do not break that often. if there is no rush, don't bother to roll if it is possible to pick the lock with their modifier. only make them roll if it matters if they make it on this turn or if failure can result in a trap going off.

[Martin Greywolf]

I suppose for gamers who mostly play video games, accepting true randomness is difficult. We take good rolls for granted and act like a bad roll is an offense against us, sometimes. Even though we knew going into it that we were playing a game with RNG-determined outcomes.

That's XCOM, baby. A 90% shot doesn't mean you have a high probability of success, it means you have a one in ten chance of getting plasma bolt to the face if you are lucky. If you arent lucky, that was a mind control on your heavy. Who still has his rocket. You did bring a guy with Stasis, right?

And for DnD 5e at least, the low levels play the best if you play them like a game of XCOM: use cover, proceed with caution, deploy consumables when necessary. Don't kick the door in and go swinging, because you are going to get super shot that way.

The issue is, I think, in not communicationg that properly to the players. If you want LotR-like adventure, where the Fellowship when cornered can take down dozens of enemies fairly easily, you need to start at level 5, because level 1 is a bit of a meatgrinder, and there will be at least one, more likely several, deaths in your party.

[Man_Over_Game]

I suppose for gamers who mostly play video games, accepting true randomness is difficult. We take good rolls for granted and act like a bad roll is an offense against us, sometimes. Even though we knew going into it that we were playing a game with RNG-determined outcomes.

Randomness, or the lack of information or expectation, is the enemy of Strategy. The less you know, the less you can plan around it, the less strategy you're allowed to insert into your reaction.

Strategy always favors the expected winner. When two players play Chess, you are rarely surprised at the outcome, because the person you'd expect to win is probably going to win since there aren't any outcomes that a better player can't predict (presumably).

By adding randomness, and destroying expectations of what happens next, you increase the chances of underdogs winning and both sides being surprised.


Unfortunately, this happens to not translate too terribly well at the table, as players are the ones who generally develop actual strategies, not the NPC enemies, and playing those strategies out is what a lot of players find fun. For that reason, you'll find that there are just some moments a DM shouldn't roll the dice, because failing an interesting plan due to a poor die roll may make the story interesting, but it might not make it any more fun.

Players have the most fun when:

• They can play the way they expected to.
• Getting what they want is still a challenge.

And sometimes, it takes a bit of experience to know exactly how to make those two things happen, and usually it doesn't involve the dice making the decision for you.

[Quertus]

And not just time, but other things as well... did you pick the lock before you broke your picks? Did you climb the cliff before the bad guy got away?

There's lots of ways of having stakes on a roll that can make rolling for it interesting.

How long the Orcs on the other side of the door have to prepare (or maybe even call reinforcements, or, Gygax forbid, retreat?)?

[DwarfFighter]

Honestly, past a certain point mitigating the chance of failure doesn't appeal to me.

Personally I've grown to like percentiles, while they're even more swingy than d20s most people seem to be able to grasp how the probability works.

Plus they have a really cool feature compared to most non dice pool in that doubles appear every 11 steps, make them special/critical results and suddenly more skilled characters are less likely to get catastrophic results and more likely to get amazing ones.

As warty goblin points out, no: 1d100 and 1d20 both have linear distribution: Each result has an equal chance of being generated.

Though to be fair, if you have a 64% skill, that is 0 steps removed from figuring out the percentage chance of succeeding, where-as the dominant d20 system of today requires more steps to determine the percentage chance of success.

The bit about doubles appearing every 11 steps, though, seems to skip over the fact that 99 and 100 are next to each other at the extreme end of the scale. This actually skews the chance of special results towards failures, and while skill is is involved, it actually works against you compared to a d20 system's 1 = critical fail and 20 = critical success (usually a GM's house rule, mind you).

1d100: If you have a 10% chance of success, none of your successful rolls will be a double, i.e. no critical successes. However, 11.1% of your failed rolls will be critical failures. Overall you have 0% chance of critical successes, 10% chance of critical failures.

Flip this to 90% chance of success, and 8.9% of your successful rolls are critical successes, 20% of your failed rolls are critical failures. Overall, you have an 8% chance of critical success, and 2% chance of critical failure.

In order to have a equal chance of critical successes and failures, you need a skill of 55%, which you will notice. If you have an equal chance of succeed or fail, i.e. a skill of 50%, there is a higher chance of critical failure than of critical success.

1d20: If you need to roll 19 or 20 to succeed, 50% of your successes will be a critical success. However, 5.6% of your failures will be critical failures. Overall you have a 5% chance of a critical failure, 5% chance of a critical success, which remains equal regardless of skill.

Conclusion: The 1d100 and 1d20 systems are virtually identical, except for granularity and for critical success/failure. If you are keen on scoring critical success, the 1d100 system is the worst option of the two.

-DF

[sreservoir]

The bit about doubles appearing every 11 steps, though, seems to skip over the fact that 99 and 100 are next to each other at the extreme end of the scale. This actually skews the chance of special results towards failures, and while skill is is involved, it actually works against you compared to a d20 system's 1 = critical fail and 20 = critical success (usually a GM's house rule, mind you).

To be fair, this is a problem with reading 00 as 100 and not as 0.

[DwarfFighter]

To be fair, this is a problem with reading 00 as 100 and not as 0.

I know of no percentile system that has the d100 roll range for 0 through 99. Problem or not, it's built into any system using the 1 through 100 range.

[HidesHisEyes]

After over a decade of lurking, reading D&D-related material, and playing D&D video games, I've finally joined an actual gaming group what plays D&D. And I will say there are a lot of things you just don't get from studying theoreticals.

Like how a good roll here and a bad roll there can swing the course of events in ways that can be really fun but also maybe a tad frustrating. A (admittedly low level) party member catching a crit on the first round and dropping immediately, as an example. On the flip side, a party member rolling a 20 on a Strength check and busting down a door so we can escape a danger zone. Certainly these are thrilling, memorable moments, but the results-oriented gamer in me feels a little...put off.

When theorycrafting it's fairly obvious that a +5 modifier is better than a +0 one. Probability favours the one with the higher modifier. That's just basic math. But in a situation where a small number of D20 rolls are made, it sometimes happens that the character with the +0 does better. Because that's how randomness works.

I suppose for gamers who mostly play video games, accepting true randomness is difficult. We take good rolls for granted and act like a bad roll is an offense against us, sometimes. Even though we knew going into it that we were playing a game with RNG-determined outcomes.

Just my thoughts on things. I am enjoying D&D so far, but I am surprised at how much I have to alter my mindset with regards to the gameplay side of it. Rolls in Warhammer 40k, as a comparison, have much lower variance.

There is something really important to understand about 5E in particular, which the game’s text doesn’t explain at all, but once it clicked for me (through playing and DMing the game for several years) the d20 “chaos plateau” became much less of a problem.

It’s all about setting stakes on dice rolls. It’s the most important part of a dice roll, much more important than setting the DC. Essentially, as the DM when you call for an ability check you should be clear on what the player is trying to achieve, what they can achieve, what is the best case scenario and the worst case scenario, *before* you call for the check, choose the stat/proficiency, or set the DC. And you take everything in the current fiction into account when you set the stakes, including the player character themselves. What this means is that the character with the right proficiency and stat for the job probably doesn’t only have a better chance of succeeding on the dice roll, but also a better result on either a success or a failure.

For example:
The party are in a tavern and, wanting to make a good impression on the townsfolk, the bard decides to give a performance. They make a Charisma (Performance) check with their +6 bonus and roll a natural 1. This *does not* mean that they break the e-string on their lute and fall off the stage. They’re a bard, that would be ridiculous. Instead it means they give a technically flawless but uninspired performance and the townsfolk aren’t especially impressed.

Say the party go back the next night and this time the fighter gives it a try, with their -1 modifier, and rolls a natural 20. This doesn’t mean the fighter somehow spontaneously learns how to perform better than the bard. It means, perhaps, that they give a rowdy, spirited, if out-of-key rendition of a popular sea shanty and all the drunk villagers join in and have a great time.

Or when the wizard tries to decipher some runes and fails, they can probably still make an educated guess as to what the runes roughly mean despite not knowing what they actually say, while the barbarian who succeeds perhaps recognises the rune for “fire” because they happen to have seen it before, even if they don’t really know what they’re looking at.

In 5E, everything is contextual. And sometimes this will go all the way up to “there’s no roll, you just succeed” or “there’s no roll, this isn’t possible for you”. Play with this approach and the randomness of the d20 isn’t nearly as problematic. Again, no one can be blamed for not knowing this because the PHB and DMG really don’t explain it at all. But from running and playing the game, I’m convinced that this is how the game’s actual design wants us to approach it.

One very important caveat though: in combat, all of this goes out the window. That’s because in combat the DM doesn’t directly set the stakes. Instead the stakes are set for you by the overdetermining system of hit points, turns, rounds and actions. The stakes are always “if you miss you don’t move any closer to victory and the enemy gets that much more time to damage you.” This definitely makes combat at low levels very swingy. But from about level 3 onward you can use the CR system as a very rough guide to make sure there’s enough leeway and buffers to keep the randomness from producing horrible results.

I hope that’s helpful anyway. Take it with a pinch of salt as I’m sure others will have other perspectives.

[icefractal]

I think that's good GMing advice for 5E, and I think that kind of system can work with the right group, but it does require the GM and players being very much on the same page.

Essentially skills do whatever the GM thinks they do, based off criteria the players may not even know about. If that thinking lines up with the player's thinking, then all's well. If it doesn't, it can feel like you're trying to play poker without being able to see your cards.

I mean, spells could be written that way too, but I wouldn't like them to be:

Fireball
Difficulty: Moderate
A bright streak flashes from your pointing finger to a point you choose within fairly long range, and then blossoms with a low roar into an impressively large explosion of powerful flame.