BOT Hard
- almost always plays the best move
- All four play with the same engine: a neural net (wildbg, 2-ply) in standard backgammon, an open heuristic evaluation in the regional variants. No level thinks shallower — a weaker level just deliberately plays the second- or third-best move from the same list every so often.
- deliberately passes on the best move about one time in four
- Computer games are never rated (that is why there is no number here).
- Running games · Computer opponents
Statistics
- Backgammon 1648 matches (391 wins) 6911 games (2481 wins)
- Long Narde 23 matches (14 wins) 172 games (96 wins)
- Gul Bara 15 matches (10 wins) 96 games (57 wins)
Faces rolled
271870 dice evaluated
- 1
- 2
- 3
- 4
- 5
- 6
Chi-squared: 3.53 (5 degrees of freedom)
With honest dice this value stays below 11.07 in 95 % of samples and below 15.09 in 99 %.
The value is in the ordinary range.
The opening roll is not counted here: it is one die per side, recorded under whoever is on turn afterwards — which is always the one who threw higher. Every single roll can be recomputed on that game's dice page.
Mistake rate
56 decisions · 1680.5 per decision (×500 scale: 840.2)
The value is the mean equity lost per decision. Lower is better; zero would mean playing every turn the way the engine does. Turns with no choice are left out — they would flatter every rate.
The reference is our own engine, not a tournament analysis: the number compares two players here to each other and to nothing outside.