The bots from the Flip7 game on this site play against each other as fast as the CPU allows - each simulated game is real computation. You control the demand below. Kubernetes adds and removes simulation pods to keep up, on a live single-node cluster, not a recording. Try killing a pod mid-run and watch it come back.
Three bot personalities from the live game:
| Tactician | plays it safe (stops at three number cards) |
| Gambler | plays for maximum (goes for seven number cards) |
| Wildcard | plays randomly (flips a coin, makes a random decision every time) |
Every game seats a random mix of them, and the table size changes the odds - compare how each strategy holds up at a 4-bot, 8-bot and 18-bot table. The freeze card can be played by either rule: officially a frozen player keeps the points collected so far (Freeze Normal); by the house rule a frozen player scores zero for the round (Freeze No Points).
Each demand unit is a load-generator pod hammering the simulation service with requests. Pick a table, a freeze rule and a level, hit Start, and the autoscaler reacts when CPU crosses its target. Table size changes what every simulated game costs - an 18-bot game is several times the computation of a 4-bot game.
table: 4 bots · freeze normal · demand: 0
Idle for 5 minutes and demand resets to 0 on its own. The whole namespace is capped by ResourceQuota - this demo cannot eat the server, by design rather than by hope.
—
Your session counts games since you opened this page; historic is everything this cluster has ever played.
The same bots fill empty seats in the live multiplayer game - see how you do against them.