Pattern Breaker is a tiny 15-round experiment disguised as rock-paper-scissors.

The opponent watches which moves you use most often and which move tends to follow your previous move. It then tries to counter its prediction. Repeated choices and simple cycles become exploitable; genuinely mixed play is much harder to read.

One thing worth knowing: the bot reads its confidence from the row for the move you just played, so the first time you play a move it has never had to follow, that row is still empty and its confidence drops straight back to 1 in 3. Staying unreadable is less about mixing evenly and more about visiting transitions it has no history for.

FEATURES

• One complete run takes about one minute.

• Works on desktop and mobile browsers.

• No account, install, tracking, ads, or uploaded data.

• An optional explanation shows how the prediction works.

Pattern Breaker is a compact companion experiment from Small Loop Studio. The end screen links to the One-Rule Playtest Kit for designers who want to test one mechanic at a time.

AI-ASSISTED PRODUCTION DISCLOSURE

Generative AI assisted with adapting an internal research prototype into this public-facing build, and with drafting the listing text. A human directed the scope and reviewed the interaction, claims, and code behavior. The cover image is not AI-generated: it is modelled and rendered by the Blender and Python scripts in the free cover-source download on this page, and the 15 bars in it are a real replay of the game's own prediction function. The adaptive opponent is local statistical JavaScript, not a hosted generative-AI model.