New worldbuilders design magic by listing powers: fire, flight, healing, speaking with the dead. Experienced ones design it by listing what magic cannot do, and what it costs when it can. The difference shows up around the midpoint of book one, when the plot needs tension and the magic keeps offering to dissolve it.

Here's the working principle: readers are never impressed by what magic can do for long. Power inflates; wonder decays. What holds a story together is the opposite — the limits, the price, the rule that will not bend even when the protagonist is bleeding. Magic gets interesting exactly where it stops.

Start from the cost, not the power

Ask the cost question first and most design problems solve themselves. What does using magic take — years of life, memory, heat, a debt to something that keeps accounts? A cost does three jobs at once:

It creates decisions. If healing a wound costs a memory, every healing is a scene with stakes instead of a reset button.
It creates texture. Costs imply economies, taboos, black markets, professions. A world where magic burns calories has fat wizards and famine wars. The setting builds itself outward from the price.
It caps escalation. The villain can't simply out-power the hero forever if power has a bill attached. Conflict shifts from "who has more magic" to "who can afford this" — a far better question.

Then add constraints that are not costs: conditions (only at night, only with consent, only what you've truly seen), materials, training. And decide what magic simply cannot touch. Death, love, time, and memory are the classic untouchables for a reason — whatever you exempt from magic is where your human drama lives safely.

Hard, soft, and the only rule that spans both

You don't need a fully "hard" system with published equations. Soft magic — vast, numinous, unexplained — works beautifully for atmosphere and dread. The craft rule that spans the whole spectrum is the one popularized by Brandon Sanderson: your ability to solve problems with magic is proportional to how well the reader understands it. Gandalf's magic can create mood and mystery, but it cannot get the ring to Mordor — the moment vague magic solves a concrete plot problem, the reader smells a cheat.

So decide, per system, what it's for. Atmosphere? Keep it soft and never let it rescue anyone. Problem-solving? Then the rules must be on the page before they're used — a limit revealed in the same scene it's conveniently overcome isn't a limit, it's an excuse.

A rule broken once is a promise broken

Here's the part that separates a magic system from a mood board: rules are promises to the reader. When chapter five establishes that the dead cannot be raised, that sentence is a contract. The reader builds every expectation, every fear for the characters, on top of it. Break it in chapter thirty — because the plot cornered you, because you forgot — and you don't just create a plot hole. You retroactively refund the tension of every scene that relied on the rule. Death stops mattering the moment resurrection becomes negotiable.

This is why magic systems are disproportionately where long books spring leaks. A magic rule is the highest-consequence kind of canon: it's load-bearing for the entire plot, and it's established in a single sentence that's easy to lose track of 80,000 words later. The plot-hole taxonomy calls these world-rule violations, and they're the screenshot-able kind.

If you want to break a rule — the forbidden thing happens, once, at terrible cost — that can be the best scene in the book. But it must be a paid exception: foreshadowed, costed, and treated by the world as the violation it is. An exception the world shrugs at is just an inconsistency with good lighting.

Write the rules down like the canon they are

A magic system that lives in your head is a system that will drift. Give it a section in your story bible and hold it to the same discipline as your characters and timeline:

Each rule, one line, with the chapter where it was established. "No resurrection — ch. 5." The chapter reference matters: it tells you what the reader has been promised and since when.
Costs and conditions as their own entries, not folded into prose descriptions where they can blur.
Exceptions logged with their justification. If you paid for a violation, record where the payment happened.
Update in the same session as the scene that touched the rule — a stale magic bible is how "only at night" quietly becomes "usually at night."

If you're starting from a blank world, structured prompts speed up the design pass considerably — our free worldbuilding prompt library has a category for exactly these cost/limit/consequence questions.

Pressure-test it like a rules lawyer

Before you trust the system, attack it. Take your protagonist's cleverest use of magic in the outline and check it against every rule you've written — not just the relevant-seeming ones. Then do the same for your villain's plan. The bugs you're hunting: a loophole that trivializes the ending, a cost that should have bankrupted someone three chapters ago, a limit that only applies when you remember it.

That last one is the killer, and it's mechanical enough to hand off. Checking "does this scene violate any established world rule?" across a whole manuscript is exactly the kind of cross-referencing humans do badly at length — which is why Creader keeps your world rules as structured canon and Guardian checks new chapters against them as you write, flagging the resurrection in chapter thirty against the promise you made in chapter five. (How it works)

Design from cost. Promise carefully. Keep the ledger. A magic system built this way doesn't limit your story — it is your story, generating conflict every time someone reaches for power and finds the price tag.