RaceFloor

When N AI agents operate in parallel on shared state, conflicts scale quadratically — N×(N−1)/2 collision pairs, none of which raise an exception. RaceFloor makes it visceral: animated agents compete for shared resources across four scenarios (support tickets, shared documents, database rows, and agent memory), and when two reach the same resource a collision overlay shows exactly what both wrote, confirms both returned HTTP 200, and reveals the last-write-wins damage. A Force Race Condition button triggers a collision on demand, and the agent slider scales to 20 to watch the conflict count explode. Single-file HTML and Canvas 2D, runs entirely in the browser with no backend.