Doors
Four breakthroughs that were never on the roadmap.
I’ve seen enough products die under their own weight. So when I started building my word puzzle, SpellBurst, I scoped it tight. Everything ran the same loop: hypothesis, prototype, evaluate, abandon, or refactor. Along the way, four doors I’d once dismissed swung open, simply because following a hunch got cheap.
The first was the interface. SpellBurst is built on WebGL, not HTML. This isn’t a website. The board, the HUD, the overlays, and the cascade VFX all composite on the GPU through a single Phaser draw tree. I wrote a real React reconciler against Phaser 4. <Container> creates a Phaser container. Flexbox lays out game objects the same way it lays out a web page. Every draw call is still WebGL. One layout system, for every mobile screen. That’s a door.
The second was synchronization. I needed the client to be server-authoritative and run deterministic game logic without paying for a round trip. Keeping two independent systems in lockstep is the kind of problem that usually calls for a big studio with deep pockets. I built an isomorphic package that shares the deterministic logic between them. The client wraps it in a Redux slice. The server wraps it in pure functions backed by Firestore. None of it works if the two sides speak different languages. The same TypeScript runs on the client and the server, so the rules are written once and trusted everywhere. That is the quiet gift of the web stack. When the abstraction is right, the complexity disappears. That’s a door.
The third was generation. Random letters produce hostile boards. Sequences that don’t appear in English. This isn’t about letter frequency. NLP figured this out decades before LLMs made it famous. The same idea LLMs use to predict the next token, running on letter tiles. The boards stopped fighting the player and filled up with words, courtesy of a 19th-century Russian mathematician. That’s a door.
The fourth was dynamic audio mixing. Wwise is the industry standard, but its web tooling is an afterthought rather than a first-class target. So I built my own. It blends biome-aware ambient soundscapes tuned to the Monument Valley aesthetic: contemplative and restrained. Ambient SFX lands at 75% silence. Starting from nothing forced a question every other tool answers for you: how much sound is too much? That restraint became the philosophy. That’s a door.
Four doors. Four problems that turned on recognition. You usually know within a day. Trying is cheap now, so if you’re not onto something, you walk away. Walking away is the easy part. Walking through is the part that takes freedom, and that freedom is the first thing most careers quietly trade away. A door also needs someone willing to believe it’s there. The work ships either way at the places I’ve been. What could’ve shipped, given trust and imagination, is the difference. None of my four doors was on a roadmap. That’s the whole point. The best ideas aren’t always designed. They’re discovered.
