The foundation of my new understanding of and appreciation for Dungeons and Dragons (especially at lower levels) is the Calibrating Your Expectations article from The Alexandrian. The main focus of that article is showing the the D&D system is fairly robust in terms of modelling realism, and then dismantling the arguments that D&D can’t model someone like Einstein, or Conan, or Robin Hood, or [insert your hero here]. Justin (who writes The Alexandrian) noted later that most people walked away from that post with a new desire for low level play (not his intended outcome), and I count myself in that crowd.
Part of how Justin went about his argument for D&D’s system was to establish what a regular person under the system would be capable of. He fished around in the DM Guide and found that most of the world — regular people — would have a standard attribute array of 13, 12, 11, 10, 9, 8 — only the top 5% would have an “Elite” array of 15, 14, 13, 12, 10, 8. So in general, regular people are lucky to have a +1 bonus in an attribute. Further down in the post he demonstrates how a 1st level artisan would conceivably have a +10 bonus on checks (+1 from attributes, +1 from skill ranks, +3 from class-skill bonus, +3 from skill focus feat, and +2 from an apprentice). With a +10 bonus, a character has about a 55% chance to meet DC 20, or if they’re unhurried they can Take 10 and meet DC 20 every time. Since this lets the artisan Take 10 and create masterwork products, Justin declares it to be master of the art.
What we’re coming to at this point is a notion of “how hard is hard.” early on in my DM career, deciding on DCs is something I really struggled with, and lacking and guidelines for what a DC 13 means compared to a DC 17 or (relatedly) how much of a penalty -4 on a check is, I found myself setting DCs based on whether I wanted my players to succeed or not (and rarely or never telling my players what the DC was, which I now think is a gross mistake).
To go back to the hard numbers, we can say that a talented, untrained person has a +1 on a check; a trained person would have a +4 or +5; and someone dedicated to the craft will have a +7 or +8. Rolling a 10 or better on d20 is a 55% chance, while a 5 or better is roughly 80% and a 15 or better is 30%. So a DC 18 check is something that has an even chance of success for someone dedicated to the craft, and is expected to fail for even trained practitioners. That is to say, for most people a DC 18 is a hard task. Conversely, a DC 11 check has a fair chance of success for anyone with a bit of talent, and basic training makes success likely (75% with a +5 bonus). A DC 11 is an easy task. A -4 penalty, though, is enough to make something that’s normally a sure thing a dicey proposal. A -8 is enough to shut down even masters of the art.
One of the things I was glad to see in the D&D Next playtest materials was a section in the DM Guidelines about DCs. They listed DC 10 or lower as Trivial (usually not worth a check), DC 11-14 as Moderate (requires minimal competence), DC 15-18 as Advanced (requires expertise or assistance), DC 19-22 as Extreme (beyond the capabilities of most people without aid or exceptional ability), DC 23-26 as Master (only the most skilled even have a chance of success), and 27+ as Immortal (the realm of demigods). I think the tiers work well with the 3rd Edition skills system (though I might dispute that DC 10 checks usually aren’t work it, unless “usually” is meant to stand for “any time you can Take 10).
(As an aside, Roles, Rules, and Rolls has a post from a week ago about how Disadvantage in 5E is roughly comparable to a -3 penalty, and thus serves a similar purpose as the -4 penalty; namely, moving a task one tier up in difficulty.)