Designing Better Monster Arcade Levels

A level is a sentence about a mechanic

The most useful reframing we know: a level is not a space, it is a statement. Every level should be saying one thing about one mechanic, and you should be able to write that thing down in a sentence before you place a single platform.

'Crumbling stone holds for less time than you think' is a level. 'A corridor with some stuff in it' is not, and it will not be interesting no matter how long you spend arranging the stuff.

This is also the fastest way to find a bad level. If you cannot write the sentence, the level is decoration.

Introduce alone, then combine

The rule that governs every hand-built level set in the library: a mechanic appears on its own before it appears in combination, and never in a hard arrangement first.

Here is the full structure of Monster Castle's eight levels, which exists entirely to demonstrate this:

Note that levels 1 to 4 introduce one thing each and combine nothing. That is half the set spent teaching, which feels indulgent while you are building it and is the reason level 8 is satisfying rather than frustrating.

The first level should be almost free

Monster Castle's first level has no hazard at all — just a gap and three gems. Players routinely finish it in under fifteen seconds and some tell us it is pointless.

It is not. It does three things that nothing else can do as cheaply:

1. It calibrates the jump. Every player learns how far a tap goes versus a hold, in a place where being wrong costs nothing.
2. It establishes the vocabulary. Platforms look like this, gems look like this, the exit looks like this.
3. It produces one success. A player who has finished something is in a different state from one who has not, and that difference persists for several levels.

The same logic applies to the first thirty seconds of an endless game. It is not wasted time, it is the only opportunity you get to teach without pressure.

Irreversibility is the one thing to be careful with

Pixel Monster Adventure contains the only genuinely punishing design decision in the library: a block puzzle you can make unsolvable by pushing a block the wrong way.

We kept it because irreversibility creates a kind of tension nothing else does — you have to think before you act, which is otherwise rare in an arcade game. But it comes with two mandatory mitigations:

• It is signposted clearly before you touch a block.
• It is never permanent. Leaving the wing and re-entering resets the blocks, so the worst case is a walk rather than a lost run.

Without the second one it would simply be a trap, and traps are the thing this whole guide is arguing against. The distinction is whether a mistake costs you a decision or costs you the run.

Pacing within a level

A good level has a shape, and the shape is roughly: easy, hard, easy, hardest, release.

The second easy stretch matters more than it looks. After a difficult section, a player needs somewhere to re-establish rhythm before the hardest part — otherwise the hardest part is attempted in a rattled state and reads as unfair.

In practice, in a single-screen level this means: an obvious opening, a real challenge, a wide platform with nothing on it, the actual test, then the exit immediately afterwards. The wide platform with nothing on it is the piece that gets cut first and should be cut last.

Making a generator do the same job

A procedural generator cannot write sentences, so the teaching has to be encoded as constraints. Monster Tower does this with tagged chunk templates and a minimum floor for each tag.

The generator also weights templates by how far past their minimum floor you are, so a mechanic is rare when it first becomes possible and common later. That produces the same introduce-then-combine curve as a hand-built set, without a designer placing anything.

The final constraint is the important one: every generated floor includes at least one plain landing platform, so there is always a route through that needs no frame-perfect timing.

Testing a level properly

Three questions, asked in this order, catch almost everything:

1. Can a first-time player finish it? If not, it is too early in the set, not too hard.
2. Can they say what killed them? If they cannot, the hazard is not readable and the fix is visual rather than structural.
3. Do they want to try again immediately? This is the real test. A level that produces a shrug has failed even if it is perfectly balanced.

The second question is where most problems actually live. A surprising proportion of 'this level is too hard' feedback turns out to be 'I could not tell what that thing was', which is a colour and silhouette problem rather than a difficulty one.

Cut the level you are proudest of

A piece of advice that has been right every time we have followed it and painful every time.

The level a designer is proudest of is usually the one with the most going on, and the most going on is usually the least readable. It is clever, it combines four mechanics in a way nobody expected, and playtesters cannot tell what killed them.

The test is not whether it is impressive. It is whether a player can name their mistake. If they cannot, the level is decoration no matter how elegant its construction — and replacing it with something simpler almost always improves the set.

Two of Monster Castle's eight levels are simpler than their first drafts for exactly this reason, and the set is better for it.

Teach, test, combine

A designed level sequence works best in groups of three, and each group introduces exactly one idea.

1. Teach. The new element appears alone, in a safe context, where failing costs almost nothing. The player is allowed to experiment with it deliberately.
2. Test. The same element in a context where getting it wrong has a consequence. No other new ideas present, so a failure here is unambiguous.
3. Combine. The element interacts with something taught earlier. This is where the interesting level lives, and it is only interesting because the two previous levels did their jobs.

Skipping the first step is the most common level-design mistake, and it is tempting because a teaching level looks trivial when you already understand the mechanic. It is not trivial to the player meeting it, and without it the test level becomes a guessing game and the combine level becomes noise.

A useful test for whether your teaching level is doing its job: can a new player explain, in one sentence, what the new thing does, immediately after finishing it? If not, it taught nothing and the next two levels will not work.

Readability beats surprise

The most common temptation in level design is the unexpected — a hazard that appears from an unseen direction, a floor that gives way without warning, a hidden trap.

These work exactly once per player, and they cost more than they return. A death the player could not have anticipated teaches nothing and reads as unfairness, which converts directly into closed tabs.

The alternative is to make everything visible and make the challenge about response rather than knowledge. A hazard you can see thirty metres ahead is not easier — you still have to deal with it — but failing it is your fault, and a failure that is your fault produces another attempt.

Where surprise genuinely belongs is in combination rather than in concealment. Two familiar elements arriving together in an arrangement you have not seen is surprising and entirely fair, because everything needed to solve it was on screen.

Playtesting without a playtest team

Most people building a browser game do not have testers, and there are techniques that work alone.

Play it tired. A level that works when you are sharp and falls apart when you are not is relying on concentration you cannot assume. This finds readability problems better than any deliberate review.

Play it on the worst device you own. Frame rate changes timing, and a level tuned on a fast machine can be quietly impossible on a slow one.

Leave it for a week. The single most effective technique available. You will have forgotten your own solution, and the level will show you whether it communicates or whether you were remembering.

Watch one person play it once. If you can get a single stranger for five minutes, do not explain anything and do not help. Where they hesitate is where the level is unclear, and you will learn more from that than from fifty of your own runs.

Alex Morgan

Browser Games Editor & Interactive Gaming Content Specialist

Alex Morgan is the editorial byline for everything published on Monster Arcade — the design notes behind all 30 original games, the strategy guides, and the tuning decisions that set every difficulty curve on the site. More about our editorial byline.

Published August 27, 2026 · Last updated September 12, 2026

Games mentioned in this guide

Platformer

Monster Castle

Climb a trap-filled castle as a small round monster. Eight hand-built levels with a real ending.

Easy Arrows / WASD + Space

Adventure

Gravity Monster

Flip gravity to walk on the ceiling. Twelve chambers where the safest floor is usually above you.

Hard Left / right + flip

Platformer

Monster Tower

An endless procedural tower. Every floor is generated, so memorising it is worthless.

Medium Arrows / WASD + Space

Battle

Monster Cannon Challenge

Launch a small fire imp at towers of blocks. Angle and power, fifteen stages, no time pressure.

Easy Aim + power, one shot