Cell Division started as a thought experiment about ten years ago, lived as a scribble on graph paper, became a half-finished web app, sat on a hard drive for years, and finally turned into the game you can download today. This is the story of how it got here.
A childhood of two-player games
Before any of this was code, it was kitchen-table games. I grew up playing Checkers first — the friendly on-ramp every kid gets — then graduated to Chess, and eventually became fascinated by Go. Somewhere in there was a long stretch of Connect Four, which is the game I probably played the most before I knew what “opening theory” meant.
The thing I loved about all of them was the same thing: two players, full information, no dice, no cards, no luck. Every loss was my fault and every win was mine to take credit for. That’s a feeling the good abstract games share, and it’s the feeling I wanted to keep.
Chess and Go have a teaching problem
The trouble with Chess and Go is that they ask a lot of you before they give anything back. Chess has six piece types, each with its own movement rules, plus castling, en passant, promotion, and a mountain of opening theory you’re supposed to absorb by osmosis. Go is almost the opposite — two rules and a board — but the emergent strategy is so subtle that beginners bounce off hard. You can play a hundred games before you feel like you understand what you’re looking at.
I wanted something with the same kind of depth, but where a brand new player could sit down, read a paragraph of rules, and start playing a real game. Connect Four gets that part right. It just has a ceiling — you solve it once and it stops surprising you.
Connect-N, and then connect-1
The obvious thing to try was turning the knob on Connect Four. What if it were connect-5 instead of connect-4? Connect-6? What happens if you let players place anywhere, not just at the bottom of a column? Each variant taught me a little and then stopped being interesting for roughly the same reason: the winning condition is a binary flip. You either have N in a row or you don’t, and most of the board is dead space between those moments.
I kept turning the knob down. Connect-3. Connect-2. And then, on a whim, connect-1, which is obviously nonsense — every cell wins, the game is over before it starts. Unless you also change the scoring rule. What if every cell was worth something on its own, and connecting just made it worth more?
That’s when the game clicked. Once scoring is continuous instead of binary, every move matters, corners and edges start to feel different from the interior, and the board turns into a running tally instead of a hunt for a single winning pattern. Cell Division’s whole scoring system — one point for a lone cell, two points per axis for each connected neighbor, eight points when all four axes light up — comes directly out of that “connect-1” moment.
Pen, paper, and a lot of erasing
The nice thing about a game built on a grid and a simple scoring rule is that you can play it with nothing but a notebook and a pencil. The not nice thing is that you have to recount scores every single turn. I played a few games out by hand, on actual paper, squinting at each cell to figure out which of the four axes had just changed. It worked. It was also tedious enough that I knew I’d never convince anyone else to do it.
So I coded it up. Just a scratch version — a grid, two colors, click to place, numbers ticking up in the corners — enough to replace the paper-and-pencil loop with something I could actually play. The rules felt right. The game felt right. And then life happened.
A web app that sat on the shelf
The next version was a small web app. Nothing fancy: static page, some JavaScript, a simple rule-based opponent. I put it online, showed it to a few friends, and then moved on. The game stayed playable, but it never got past the “toy project” stage. Years went by. I’d occasionally open the page, confirm the game was still there, and close the tab again. That shelf-time was longer than I care to admit.
The gap ended up being useful, in a way. The idea had to survive a decade of me not touching it, and it did — I still enjoyed the game every time I came back to it. That’s not nothing. Most of my abandoned projects I open years later and immediately understand why I stopped.
Finally, an actual app
What finally got me to ship was deciding to build it as a real mobile app instead of a web toy. Touch input makes the board feel like a physical thing, the score counters animating on every move makes the rules teachable without a tutorial, and having it live on a phone means I can actually play a game of it while waiting for coffee. The Elite AI — the whole AlphaZero-style distillation stack, the 14 hand-crafted features, the hint system — all of that came later. None of it would exist if the core game hadn’t survived a decade on a shelf first.
The lesson I keep coming back to is that the right design for a game is whatever still feels fun after you’ve stopped thinking about it. Cell Division passed that test by accident, because I literally forgot about it for years at a time, and it kept working every time I came back. That’s the game I finally decided to ship.
Connect Four asks “can I line up four?” Chess asks “can I keep up with theory?” Cell Division asks the simplest possible question — “is this cell worth more than the one my opponent could play here?” — and lets the depth come out of the answer.
If you want to see what I ended up with, the How to Play post walks through the full scoring rule in a few boards. The Building the AI post covers the engineering that came after this origin story ended.