Pick the odds first: setting dice target numbers backward from the success rate you want

Here is the usual way a difficulty number gets set at the table: the GM thinks "this lock is pretty hard," picks 15 because it sounds hard, and moves on. The player rolls, and whether the scene works hinges on a number nobody actually reasoned about. Sometimes it lands. Often the "pretty hard" lock turns out to be a coin flip, or a brick wall, and the GM only finds out after the dice hit the table.

There is a better order of operations, and it is the reverse of the habit above. Don't start from the target number and hope the odds feel right. Start from the odds you want — "this should succeed about two times in three for a competent character" — and then find the target that produces them. The target number is an output, not an input. Once you think of it that way, difficulty stops being vibes and becomes a dial you can set on purpose.

Why the target number alone tells you nothing

A target of 15 means completely different things depending on what is being rolled. On a single d20 with no modifier, hitting 15 or higher happens 30% of the time. Add a +5 and the same target is suddenly a 55% proposition. Roll 3d6+5 against that same 15 and the odds are different again — and far steadier — because the bell curve of three dice clusters the results near the middle instead of spreading them flat.

So "DC 15" is not a difficulty. It is a difficulty only in combination with the dice and the modifier the character brings. The same nominal target is trivial for one build and near-impossible for another. This is exactly why setting the number first leads you astray: the number has no fixed meaning until you attach it to a specific roll.

Decide the feeling, then the frequency

Start by naming what the check should feel like, and translate that into a rough probability band. A useful ladder:

• Routine but not automatic — 80–90%. The character almost always succeeds; the roll exists to create the occasional surprise.
• Favored — 65–75%. Competence shows, but failure is a real outcome you should be ready to narrate.
• Even — roughly 50%. A genuine coin flip; use it when either result is dramatically interesting.
• Long shot — 20–35%. Failure is expected; success is a moment.
• Desperate — under 15%. Reserve for stakes where trying at all is the story.

Notice these are ranges, not points. You are not trying to engineer a 63.4% check — you are trying to keep the check out of the wrong band, so a "favored" action doesn't accidentally become a coin flip.

Read the target off the curve

Once you know the band, the mechanical question is simple: given the dice and modifier the character actually rolls, what target lands in it? Punch the character's real roll into the dice probability calculator — number of dice, sides, and their total modifier — and slide the target until the success percentage sits inside your band. The bar chart makes the trade-off visible: each step you push the target up chops off another slice of the curve, and on a bell-curve pool near the peak, a single point of target can swing the odds far more than a point out in the thin tails.

That last observation is the one worth internalizing. On a flat d20, every point of target is worth exactly 5% — the swing is uniform. On a 3d6-style pool, a point of target near the center of the curve might be worth 10% or more, while a point out at the extremes is worth almost nothing. If your system uses dice pools, small difficulty adjustments bite hard in the middle and barely register at the edges. Set targets accordingly: nudge gently near the center, and don't bother fine-tuning out in the tails.

Advantage is a difficulty adjustment in disguise

Granting advantage (roll twice, keep the higher) is not a flat bonus — it is a shift whose size depends on where you sit on the curve. Advantage helps most when your baseline odds are near 50% and helps far less when you were already very likely or very unlikely to succeed. If you toggle the calculator between standard and advantage while holding the target fixed, you can read off exactly how many effective points of difficulty you just handed the player. Often it is easier to grant advantage than to fiddle with the target — but know that you have, in effect, lowered the difficulty, and by a variable amount.

Calibrate once, reuse forever

The payoff of working backward is that you only have to do it a few times. Build a small table for your system: for a typical competent character's roll, what target gives 75%, 50%, 25%? Those three anchors cover almost every check you will ever call for. After that, "favored," "even," and "long shot" each map to a number you trust, and you can set difficulty at the speed of narration without guessing.

Difficulty is one of the few levers a GM controls with real precision, and most of that precision is thrown away by choosing the number first. Flip the order. Decide how often the check should succeed, then let the dice probability calculator tell you the target that delivers it — for a plain d20, an exploding pool, or a roll with advantage — and every difficulty you set will mean exactly what you intended it to mean.