Six environments. We add one every couple of months and retire whatever nobody trains on.
Five reels, three rows. Matches read left to right. The state you get is the grid plus your credit count.
Match three in a mine shaft. Clearing four in a row carries the multiplier into the next drop.
Two minute duels against a fixed house deck. The whole ground is a pattern recognition problem.
Time each spin to land the next floor. Miss twice and the stack goes. Punishing for greedy strategies.
Drop gems, clear rows, no timer. Runs until you jam the grid, so scores get long.
Nine lanes, one ball, the middle pocket pays triple. Short rounds, high variance, hard to tune.
Every ground shares one credit balance. Credits buy longer runs and extra parallel trials. You earn them by training, which means a bot that improves pays for its own next run. They are virtual. They cannot be bought, cannot be withdrawn, hold no cash value, and are never exchanged for money or prizes.
Training runs need somewhere to save. Give us an email, we send a link, and your strategies and logs are waiting next time. Nothing to pay, now or later.
Training credits and items are virtual. They have no cash value, cannot be bought, cannot be cashed out, and are not exchangeable for money or prizes of any kind. The grounds are practice environments for writing strategy code.