Each player moves 15 checkers around the board toward their home board, then bears them off.
Roll two dice. Move one checker by each die, or the same checker in two legal steps. Doubles give four moves.
A point with two or more opposing checkers is blocked. A single opposing checker is a blot — it can be hit.
Hit checkers go to the bar and must re-enter before any other checker moves.
Once all your checkers are in your home board, bear them off. First to bear off all 15 wins.

The doubling cube raises the game's value. Before rolling on your turn, you may offer a double.
Your opponent can take (continue at the new value, they now own the cube) or drop (concede the current value).
In a match, score and match length affect cube decisions. A correct take or drop is as important as a good move.
Crawford rule: the cube is disabled for one game after a player reaches match point (one point away from winning).

Local Play
Full match vs the AI. Choose match length (1–11 pts) and use the doubling cube. After each move you see the JUBLI engine's preferred play and your equity loss.
Training
A board position is generated and your dice are set. Pick what you think is the best move (or cube decision), click Submit, and the engine reveals the top play. You see your exact equity loss.
Daily Puzzle
One position, refreshed every day, the same for everyone worldwide. Six dice options are shown — pick the roll that gives the strongest position. Streak and results revealed after.
Blitz
Quick-fire positions under a 10-second timer. Find the best move before the clock hits zero. You start with 3 lives — a blunder costs one. Personal bests are saved in this browser.
Sandbox
Full analysis tool. Drag checkers to build any position, click dice to set any roll, then evaluate all candidate moves or run a rollout (hundreds of simulated games) for deep comparison.

Equity measures your winning chances. +1.0 is a certain win, −1.0 is a certain loss, 0.0 is a perfectly even game.
After each Training move, your equity loss is shown. 0.000 is perfect. Under 0.050 is good. Over 0.100 is a significant mistake.
In Sandbox, a rollout simulates hundreds or thousands of games from a position to accurately estimate each move's true equity — more reliable than a single neural-net evaluation.
The JUBLI engine uses the same neural network approach as world-class backgammon software. Its evaluations at depth 2+ are strong enough to improve any player's game.

The JUBLI engine runs entirely in your browser as WebAssembly — analysis and game logic need no server.
Once the app has loaded once, Local Play, Training, Blitz, and Sandbox all work with no internet connection.
Daily Puzzle loads precomputed puzzle files, so gameplay does not require an app server.
No account, no login, no personal data collected. Your games, settings, and personal bests stay in your browser only.

Turn on Help mode, then click the board, dice, cube, status area, move list, or log to get an explanation of that part of the interface.

Choose a mode

Gammon Guru