Strategy, updates, and deep dives.
In self-mirror matches between identical Elite copies, Player 1 wins 96.8% of 7x7 games. One extra move on odd boards turns tempo into a near-lock — and sometimes flips match outcomes 100-to-0.
Read →Labeling a tier "Hard" is free. Proving it beats "Medium" is not. Every time we change an AI, we run 9,000 games across every tier and board size — the two heatmaps that come out are how we know the difficulty ladder actually holds together.
Read →Elite isn't a cloud call, a scripted sequence, or a random chance. It's a tiny neural network trained to imitate a much stronger AlphaZero-style teacher — and it runs entirely on your phone.
Read →We spent three weeks building a full AlphaZero system for Cell Division — then didn't ship it. Here's why tree search was wrong for a phone, and why the trained network still taught the Elite AI you play today.
Read →How a decade of Checkers, Chess, Go, and Connect Four turned into Cell Division — via a knob-turning experiment that ended at connect-1 and a new scoring rule.
Read →The in-game hint used to have its own trained models, one per opponent. We deleted all of it. Here's why the hint now just asks Elite what to play — and why we deliberately chose not to ship a hint that could beat Elite for you.
Read →Why corners cap at 6, why the 1-away pivot captures territory, and how to think about Cell Division as a zero-sum margin game.
Read →Place cells, build connections, outscore your opponent — a visual walkthrough of the rules, scoring, and AI tiers.
Read →Cell Division has four AI opponents — Easy, Medium, Hard, and Elite. They're genuinely different from each other, not the same bot with more handicap. Here's what each one is doing and which tier to play.
Read →