Case Study 40.1 — Celeste: From TowerFall Side-Project to Indie Landmark

This is a book-closing case study. The arc it traces — from a four-day game-jam prototype to a genre-defining release to a decade-long career — is the pattern this chapter has been arguing exists. Celeste is the pattern in concentrated form.

The 2015 Prototype

In July 2015, Matt Thorson and Noel Berry were working on a small platformer prototype during a four-day break between other projects. Thorson had just shipped TowerFall Ascension and its Dark World DLC — a couch-competitive archery game that had put him on the indie map. Berry was his frequent collaborator. The prototype was a constrained experiment: eight-direction movement, a climb mechanic, a dash with fixed cooldown, and a simple screen-by-screen challenge structure.

They built it in PICO-8, the virtual fantasy console by Lexaloffle that forces games into severe constraints: 128x128-pixel resolution, 16-color palette, tiny code and memory budgets. PICO-8's severity was the point. The constraints meant that a four-day project could actually ship. No time to overscope. No time to obsess over art. Just the mechanics, the levels, and the feel.

The result was a 30-screen platformer where you climb a mountain, dying constantly, with dash-and-wall-jump movement that felt, even in the prototype, unusually tight and readable. They called it Celeste Classic. They released it free on the PICO-8 community BBS.

It was a small success. PICO-8 enthusiasts noticed it. It showed up on a few "PICO-8 games worth playing" lists. That was the scope of the impact in 2015.

Thorson and Berry went back to other projects. TowerFall's console ports. New prototypes. The normal drift of indie development.

The 2016 Decision

In 2016, Thorson began to reconsider Celeste Classic. The PICO-8 version had stuck with players in a way his other prototypes had not. The movement feel — the dash, the climb, the specific arc of the jump, the grace of the forgiveness frames — had a quality that few indie platformers were achieving. He started to think about expanding it.

The expansion decision was non-obvious. Indie developers do not typically scale up their game-jam prototypes into full games. The usual pattern is: the prototype sparks something, you learn a lesson, you apply the lesson to your next project. Scaling a prototype up is hard because the prototype's charm often depends on the constraints — eliminate the 128x128 resolution and the 16-color palette, and the game becomes a generic platformer instead of a specific one.

Thorson made the decision anyway. He and Berry began work on what would become Celeste (the full version) in early 2016. They brought in composer Lena Raine, artist Pedro Medeiros, and several other collaborators across the project's life. Berry handled much of the engine and system work; Thorson drove the level design and story; Raine composed one of the most-praised indie soundtracks of the decade.

The scope of the expansion was significant: from 30 screens in PICO-8 to roughly 700 screens across nine chapters in the full game. From a 4-day build to a 2.5-year build. From a nameless protagonist to Madeline — the named, characterized, mental-health-focused player-character whose journey up the mountain becomes an allegory for the developer's own anxiety and depression.

The January 2018 Release

Celeste released on January 25, 2018, on Windows, macOS, Linux, PS4, Xbox One, and Nintendo Switch. The Switch launch in particular was a turning point — Celeste became one of the platform's most-celebrated indie launches and rode a wave of word-of-mouth through that spring.

The critical response was immediate and unanimous. Scores in the 90s across major outlets. The Game Awards 2018: Games for Impact (win), Best Independent Game (win), Best Score/Music (nomination), Game of the Year (nomination — unusual for a 2D indie platformer). The Independent Games Festival 2019: Seumas McNally Grand Prize (win). BAFTA: Game Beyond Entertainment (win).

Three things drove the reception.

The movement feel. Celeste's dash, climb, and jump are cited by designers and speedrunners alike as among the most carefully-tuned 2D movement in the medium. Every frame of the character controller is deliberate — the pre-dash freeze frame that communicates the direction the dash will go, the coyote-time that forgives a late jump, the input buffering that forgives an early press, the tiny upward nudge at the end of a successful jump that makes the arc feel generous. Chapter 8 of this book on feedback systems, Chapter 11 on flow, and Chapter 26 on combat design all pull from Celeste's controller as a reference. Every line of feel was engineered.

The difficulty curve paired with the assist mode. Celeste is hard. The baseline game is hard, and the optional B-sides and C-sides are brutal. But early in development, Thorson made a decision that became genre-defining: the assist mode. A settings menu that lets players slow the game down, give themselves infinite dashes, make themselves invincible, or any combination. The assist mode was not hidden. It was front-and-center, introduced via an in-game tutorial, and — crucially — framed without shame. The game literally tells players that using the assist mode is legitimate, that the game is for everyone, and that the difficulty setting is a tool rather than a confession.

The assist mode changed the industry. Accessibility discussions in games had been ongoing for years, but Celeste's matter-of-fact framing — "everyone should be able to play this game, so here are the knobs" — shifted the conversation. Within two years, major studios were citing Celeste as a reference for their own accessibility systems. The Last of Us Part II (2020), God of War Ragnarök (2022), and dozens of others adopted similarly comprehensive accessibility settings. Celeste did not start the accessibility conversation in games, but it moved it forward in a way that the industry has not retreated from.

The narrative — specifically, the mental-health framing. Madeline's climb is an allegory for struggles with anxiety and depression. The antagonist for much of the game is her own shadow-self — a literalization of the depressive inner voice. Chapter 6 ("Reflection") takes place inside her consciousness. The resolution does not come through defeating the shadow but through integrating her. The final chapter, released free as a post-launch update, extends this into an extended meditation on self-acceptance.

This framing landed because it was honest. Thorson has spoken in post-launch interviews about his own struggles with anxiety during Celeste's development. The game's narrative was not a theme bolted on to mechanics — it was the developer's life threaded through the mechanics. The climb was hard. The dash was hard. The B-sides were brutal. Madeline's climb up the mountain and Madeline's struggle with herself were the same climb, and the mechanical difficulty of the game was the emotional difficulty of the narrative. Form and content, inseparable.

The Post-Mortem and the Specifics

Thorson and Berry wrote a widely-read post-mortem published across several channels in 2018. Like most good post-mortems, its lessons are specific enough to apply.

What went right, specifically. The decision to work in GameMaker (ultimately FNA for the final build) let them iterate fast. The choice to keep levels small (single-screen units) let them tune each screen individually. The strawberry system — the optional collectibles — was added late in development and became one of the game's most effective pacing tools. Lena Raine's score, delivered in close collaboration with the level designers, shaped each chapter's emotional register in ways static music could not have.

What went wrong, specifically. The B-sides expanded from a small secret to a significant portion of the project's final scope, contributing to the extended development timeline. Some of the middle chapters were iterated on for too long — Thorson has said in retrospect that Chapter 5 ("Mirror Temple") was the place where scope and design coherence were hardest to keep together. Optimization for Switch consumed more time than anticipated.

What they'd do differently. Greater willingness to cut early when a chapter was not working. Tighter scope discipline on the optional content. Earlier involvement of playtesters from outside the developer's immediate circle.

These are the lessons of a post-mortem. They are specific. They are honest. They are actionable for other designers.

The Career Arc After Celeste

The post-Celeste trajectory illustrates the broader pattern of "what comes next" after a breakthrough first real game.

Thorson and Berry founded Extremely OK Games in 2019 to develop future projects with creative independence. The studio has worked on several projects, including Celeste 64: Fragments of the Mountain — a free 3D reinterpretation released in January 2024 to mark the sixth anniversary of the original Celeste. Celeste 64 was a 7-person, two-week game-jam-style project within the studio, and was notable for the creative contribution of Bennett Foddy (QWOP, Getting Over It) as a guest collaborator — an example of the industry's small-world nature once a designer has shipped.

The studio announced Earthblade — a successor game, larger in scope — in 2023, with a planned release later in the decade. Development has moved at a deliberate pace, a pattern consistent with Thorson's preference for quality over speed.

Beyond Thorson's studio, Celeste's influence has been sustained. Speedrunning communities built elaborate subcultures around its mechanics. Fan-made DLC chapters (including the community-produced Strawberry Jam collab) extended the game's life into 2023 and beyond. The soundtrack has been performed live at multiple classical-ensemble events. Academic design programs cite Celeste as a touchstone. The assist-mode framing has permeated the industry's approach to accessibility.

What This Case Study Argues

Celeste's arc demonstrates several things this chapter has claimed.

The first shipped thing matters more than its scope. Celeste Classic — the four-day PICO-8 prototype — was a small shipped thing. It was not the masterpiece. The masterpiece came two and a half years later. But the masterpiece did not exist without the small shipped thing. The prototype's shipped status made the expansion decision real. The unshipped prototype — the one Thorson might have left on his hard drive — could not have become Celeste.

Craft consolidates between the first and second versions of an idea. Celeste Classic had good movement. Celeste had extraordinary movement — same core, refined across 700 screens of design pressure. The second version is where craft consolidates. Your second game will be to your first game what Celeste is to Celeste Classic: the same vocabulary, now turned into a language.

Constraint is generative. Celeste Classic was built inside PICO-8's severe constraints, and those constraints are what made the prototype possible in four days. The full Celeste maintained design constraints at every scale — single-screen rooms, a small core mechanical vocabulary (run, jump, dash, climb), a focused emotional register. The craft of Celeste is inseparable from its willingness to be a small game well, rather than a big game averagely.

Accessibility is design. The assist mode is not a concession. It is a design choice that expanded the game's audience without diminishing its challenge. Thorson and Berry could have made the game harder, or easier, or gated the hardness behind progression. They made the hardness a player-chosen parameter. The design lesson applies beyond Celeste: the question "how hard is my game?" is often less useful than the question "who is allowed to play my game?"

The craft community is a gift economy. Thorson and Berry wrote a post-mortem that is, itself, a contribution to the craft community — other designers have built their movement controllers by studying Celeste's techniques. The post-mortem was published freely. The patches that added content and the release of Celeste Classic 2 in 2024 were likewise gifts. The pattern is not charity; it is how the field renews itself across generations of designers.

You have a game. You built it. You are about to write your post-mortem. You are about to consider what comes next.

If the next thing is a Celeste-scale project, so be it — but you do not need to plan for it. You need to plan for your next small shipped thing. The one whose mechanics stick. The one that surprises you a year later when you realize it could be the prototype of something larger.

Matt Thorson shipped a PICO-8 prototype in July 2015 and did not know it would become Celeste. That is the point. You ship the small thing, and years later the small thing becomes either the story you tell in your post-mortem or the seed of the next project. Either way, the shipping is what makes the rest available.

Takeaways

Celeste teaches the capstone's lessons in concentrated form. The four-day prototype is the ship-small-and-often principle in action. The two-and-a-half-year expansion is the craft-consolidates-on-the-second-try principle. The assist mode is the accessibility-is-design principle. The post-mortem is the give-back-to-the-community principle. The ongoing studio career is the this-is-a-long-career principle.

You will not be Matt Thorson. You do not need to be. The arc he traces is one version of the career you are now starting, and there are thousands of others — some bigger, some smaller, some different in every particular. What matters is that he started by shipping a small game, and that he kept shipping. The rest followed from that discipline.

You have just shipped your small game.

The rest is up to you.