Right... but if it's 5% and there's a penalty for failure, why would I want to rule that possibility out through fiat? Why is the skill system failing to support normal ranges of numbers?
If you want the player's action to have a 5% chance of failure, set the DC accordingly. Maybe it'll be 9 or 7 instead of 10. It's entirely in your purview to do so.
Except the peasant's arrow only hits the dragon through a special "20s always succeed" rule that doesn't apply outside attack rolls.
Depends on the dragon, they're not all AC 23+
Also - if anything, this is one place where you don't actually want this to happen, because it means that you only actually need a modest army to defeat the biggest baddest dragon in a single round.
I recall seeing a designer quoted as saying that was exactly what they had in mind. You could toe-to-toe or cast a dragon to death if you were high enough level, but if you could gather enough determined archers together, you could also defeat it that way.
It's the exact same place where the advice for impossible and automatic successes come from.
Narrating success or failure or call for a check is right there in the most basic take on the resolution system in the DMG.
If we're going to say "it's optional"
Everything's optional. The passage you quoted was advice.
Right... so our bounded accuracy is thrown out the window, replaced with DM fiat.
No, DM fiat is removed, at the DM's discretion, and replaced by BA - when he calls for a check.
Semantics!
Seriously, though, bounded accuracy is about checks, and checks happen when the DM calls for them. Not calling for them when you don't want at least a 5% chance of failure (or success) isn't overriding them, it's just DMing.
It's also why the example I brought up is a problem. Because the DM now has to rule different levels of possible/impossible depending on the character attempting the task.
He can if he wishes.
At that point the skill system is failing - it's not giving the DM meaningful guidance, and it's not communicating chances of success or even correct expectations of capability to players.
The system simply doesn't begin to work until the DM begins making rulings. Calling for a check is a ruling, success/failure, is a ruling. No DM, no ruling, no system. The DM is integral to resolution this time around.
You're saying that like the DM wasn't allowed to rule something impossible in a prior edition that he would rule impossible in the current edition
If you're OK ruling something impossible in the face of a table giving a fixed DC for the action in question, and a player who's bonus alone overwhelms that DC, why would you have an issue with it in a system that tells you to rule success, failure, or call for a check?
You're also saying it like you can't still do this in 5e...
You can get a +11 to a check, a +17 with Expertise. That doesn't overwhelm the d20. The 20th level expert can fail a check that an ordinary warm body might succeed at. (specifically, while rolling a 1, while the warm body rolls a 19, but it's still not quite overwhelmed).
So the system still has the same problems - the DM always needed to decide that things at the extreme end of the spectrum were impossible
In 5e, he's required to make that determination every time.
Like I said - I'd rather not pretend that I'm infallible.
It makes for better 5e games if you can convincingly feign infallibility from behind the screen.
