Case Study 17.1 — Celeste: The Introduce-Test-Twist-Master Masterclass

Game: Celeste (Maddy Makes Games / Extremely OK Games, 2018) Designer / Director: Maddy Thorson Co-creator: Noel Berry Platforms: PC, Switch, PS4, Xbox One Why it matters: Celeste is the canonical modern reference for 2D level design pedagogy. It is taught in design schools. Its mechanics tutorial is held up as a model of "show, don't tell." Its difficulty curve is often cited as the best in the genre. And it does all of this with one button (jump), one trigger (dash), one stick, and a tilemap.

The Core Idea

Celeste is the story of Madeline, a young woman climbing a mountain. The mountain is a metaphor — for anxiety, for self-doubt, for the work of becoming. The game is in Maddy Thorson's words "a game about overcoming."

Mechanically, it is a precision platformer with three core verbs: 1. Jump (variable height, with coyote time and input buffering) 2. Dash (one in any of eight directions, recharged by touching the ground or specific objects) 3. Climb (grab and shimmy on walls, until your stamina runs out)

That's it. Three verbs. From those three verbs, Maddy Thorson and team built eight chapters, hundreds of screens, and a mechanical depth that compares favorably to games with five times the verb count.

The lesson: mechanical depth comes from interactions, not additions. Adding new verbs is one way to add depth. Recombining existing verbs across new contexts is another — and it scales further.

🎯 Tradeoff Spotlight: A small verb set is harder to design with (you must constantly find new combinations) but easier for players to master. A large verb set is easier to design with (each new mechanic is a fresh idea) but harder to master and easier to forget. Celeste chose mastery. The result is a game many players replay annually.

The Introduce-Test-Twist-Master Pattern

Mark Brown of Game Maker's Toolkit did a video on Super Mario 3D World where he documented Nintendo's "four-step pattern." The pattern, abbreviated ITTM:

1. Introduce: Show the mechanic in a safe context. The player sees it and understands what it does.
2. Test: Place the mechanic on the critical path. The player must use it to progress.
3. Twist: Combine the mechanic with another mechanic, or change one parameter (faster, sharper, inverted).
4. Master: Demand the mechanic under pressure — combined with multiple other mechanics, or required for a difficult execution.

Celeste applies this pattern not just to its three core verbs but to every chapter-specific mechanic — and there are dozens. Wind currents in Chapter 4. Ice blocks in Chapter 5. Photo-reactive crystals in Chapter 6. Reverse-gravity zones in Chapter 7.

For each, Maddy Thorson runs the ITTM cycle, often across just 5-10 screens. The introduction is usually a single screen with the mechanic alone in safe context. The test is the next screen, requiring the mechanic. The twist appears within 3-4 screens, combining the mechanic with another from earlier chapters. The master appears at the chapter's climax — a screen that demands the mechanic in concert with everything you've learned.

💡 Intuition: The introduce-test-twist-master pattern is not a formula to apply mechanically. It's a mental model that prevents you from forgetting any of the four phases. Many games introduce and test, then never twist or master — leaving mechanics underexploited. Many games twist and master without proper introduction, leaving players confused.

Chapter 1: "Forsaken City" — A Microcosm

Let us walk through Chapter 1 in detail, because it is a microcosm of the whole game's design philosophy.

Screen 1. Madeline awakens at the foot of the mountain. The screen shows a flat path with a single character on a hillside. The player presses the d-pad and walks. The screen scrolls, revealing the mountain ahead. Lesson taught: d-pad = movement.

Screen 2. A small gap appears. The character cannot walk over it. The player will, almost certainly, press jump. The character jumps over the gap. Lesson taught: A button = jump. Jump distance is what it is — internalize the arc.

Screen 3-4. A few platform jumps. Some are easy, some require slightly tighter timing. The character meets the old woman at the trailhead, who delivers the first dialogue and the game's emotional setup. Lesson taught: the world is climbable; the mountain is the destination.

Screen 5. The first dash chevron appears: a glowing crystal floating in space. Madeline grabs it. Her sprite turns from blue to white. The player likely presses every button trying to figure out how to use the dash. They press the dash button (or trigger), and Madeline rockets in the direction of input. Lesson taught: the dash is a directional rocket, and it has a use-once charge that recharges on landing.

Screens 6-8. A series of dash-required jumps. None difficult. The player is simply practicing the verb. Phase: introduce + early test.

Screen 9. The first dash chevron is followed by a second dash chevron mid-air. The player must dash to one, then dash from it to the next, chaining dashes without touching the ground. Phase: twist — the dash-chain is born.

Screens 10-12. Wall jumps. Madeline can grab walls and jump from them. The player learns this through a screen that simply doesn't have any other way forward. The level itself teaches the mechanic by demanding it. Phase: introduce + test, in a single screen.

Screens 13-15. Wall jumps combined with dashes. Phase: twist.

Screen 16-onwards. The mountain begins. Spikes appear. Falling blocks appear. Each new chapter mechanic gets its own ITTM micro-cycle, all while the dash and wall jump from earlier are required throughout. Phase: master, plus continuing introduction of new ideas.

By the end of Chapter 1, Madeline has been climbing for about 30-45 minutes (for a careful player). She has died, depending on skill, between 0 and 200 times. She has learned three verbs and used them in dozens of combinations. Maddy Thorson has taught her a language without speaking a word of tutorial text.

💀 Design Autopsy: Compare this to games that gate the dash behind a 5-minute "explanation" cutscene with text boxes. Celeste's dash teaching takes 0 seconds of dedicated tutorial time and is impossible to forget because it was learned through use, not described through telling.

Maddy Thorson's Design Philosophy

In several public talks and the postmortem, Thorson and Berry have articulated the principles behind Celeste's design. A condensed version:

1. "Death is not punishment."

In most games, death sets you back. In Celeste, death takes 1.2 seconds — a quick black wipe and you're at the start of the screen. There is no penalty beyond the brief pause. This radically changes the player's relationship to risk.

When death is cheap, the player tries things. They attempt risky jumps. They experiment with mechanics. They explore the screen for secrets. In a high-cost-death game, the player plays conservatively — they hate dying and so they avoid risk. Celeste's cheap death turns mistakes into iterations.

✅ Best Practice: If you want players to experiment and learn, make failure cheap. Long death animations and large setbacks teach players to play it safe — often the opposite of what your design rewards.

2. "Forgiveness without lowering the ceiling."

Celeste implements coyote time, input buffering, and "buffered dash" (you can press dash a few frames before landing and it'll trigger when you land). It also has "Assist Mode" for players who want to slow time, give themselves invincibility, or skip to the next chapter.

But the game's ceiling is not lowered by these. The C-side challenges are some of the hardest platforming sequences ever shipped. Speed-running Celeste requires frame-perfect inputs. The forgiveness lifts the floor — it does not lower the ceiling.

This is the design balance: make it easy to play, hard to master. The two are not in tension if you treat them as orthogonal axes.

3. "Each room is a sentence."

Maddy Thorson described this in an interview. Each room in Celeste is a complete unit of meaning. It introduces an idea, develops it, and resolves it. The player completes it in 5-30 seconds. The transition to the next room is a punctuation mark.

This single-screen structure (despite Celeste's continuous chapters) is what enables the ITTM pattern to work so cleanly. Each room is one beat in the cycle. The chapter is the paragraph. The game is the essay.

🎮 Play This: Replay any Celeste room. Notice how the camera never scrolls horizontally during the room — only when transitioning. Each screen is fixed. This is intentional. It's the room-as-sentence philosophy in technical implementation.

4. "Hard is fun, frustrating is not."

There is a difference between difficult and frustrating. Difficult is when the obstacle is clear and the solution is just outside your current ability. Frustrating is when you don't know what went wrong, or when the game's systems seemed unfair.

Celeste avoids frustration through: - Clear hazards. Spikes are red. Bumpers are bouncy-looking. Hazardous tiles never look like safe tiles. - Death feedback. When you die, you know exactly where and why. - Restart locality. You restart at the start of the current screen, not the chapter. - No hidden mechanics. Every game system is teachable through play.

The result: Celeste players die hundreds or thousands of times in a playthrough but rarely rage-quit. The deaths feel earned. The next attempt feels possible.

⚠️ Common Pitfall: Designers often confuse difficulty with difficulty spikes — sudden, unfair-feeling escalations. Celeste's curve is smooth. Each screen is slightly harder than the last in a way the player can feel and learn from. Spikes feel like cheating; smooth curves feel like growth.

The Heart Side / B-Side / C-Side System

Beyond the main chapters, Celeste has hidden harder versions of each chapter: - B-Sides are unlocked by finding tape cassettes in main chapters. Each B-Side is a remixed version of the chapter — same mechanics, dramatically harder. - C-Sides are unlocked after the main game. Each is a 5-room ultra-difficulty test of the chapter's mechanics.

The B-Side and C-Side system is a level design lesson in itself. It demonstrates that the same mechanical vocabulary can be reused at vastly different difficulty levels by recombining elements more aggressively. The dash, jump, and climb verbs that taught you Chapter 1 are the same verbs Chapter 1 C-Side demands you execute with frame-perfect timing.

This shows that mechanical depth scales with combination, not addition. Celeste doesn't add new verbs for the C-Sides. It just stacks the existing verbs into denser, faster, more demanding sequences.

Lessons for Your Own 2D Level Design

What can you steal from Celeste? Five concrete takeaways:

1. Make death cheap. A 1-second restart animation enables aggressive experimentation. Long penalties enforce conservatism.

2. Apply ITTM rigorously. Every mechanic should be introduced, tested, twisted, and mastered. If any phase is missing, the mechanic is underused.

3. Treat each screen as a sentence. Even in continuous-scrolling games, design screens as discrete units of meaning with their own mini-arcs.

4. Implement forgiveness without lowering the ceiling. Coyote time, input buffering, and accessibility options help everyone. They do not require you to make the game easier overall.

5. Mechanical depth is a recombination problem. You don't need 30 mechanics. You need 3-5 mechanics that interact in many ways. Spend your design budget on combinations, not new verbs.

📐 Project Checkpoint: Open your Level 1 from this chapter's progressive project. For each mechanic in your level (jump, gap-cross, enemy-avoid), can you identify an introduce screen, a test screen, a twist screen, and a master screen? If not, which phase is missing? That's where your level design needs more work.

Closing Reflection

Celeste is not a great game because it has great mechanics. It has great mechanics — but so do many games. Celeste is great because every design decision was made with intent, every screen was tested by people who understood the difference between difficult and frustrating, and every mechanic was given the full ITTM cycle.

The lesson is not "copy Celeste." The lesson is "design with this much intention." When you design your own platformer, take 5x as long as you think you need. Iterate 10x as much as you think you should. Cut the levels that don't earn their place. Trust that the player can learn anything you teach them well.

Maddy Thorson once said that Celeste's development took five years and that most of those years were iteration. The game shipped in 2018 to universal acclaim. It has sold millions of copies. It is a genre-defining work.

It started as a one-week game jam.

The lesson there is also worth holding: great design is iteration over time, not inspiration in the moment. Start small, ship often, listen to playtesters, and iterate. Celeste's roots are humble. Yours can be too.