How the computer plays

The computer opponents are not scripted – they read the board the same way you do: by asking the game engine which moves are legal and what each one would change. They differ in how far they look.

Level	How it chooses
Easy	Half of its moves are what Medium would play, the other half are random legal moves.
Medium	The move with the best value right now. Alone that is the most gold. In a duel it judges the position after the move: 12 points per ruled land, 1 per held cell, a quarter per gold – its own minus yours.
Hard	Takes the 4 best candidates by that value and plays each one 10 moves into the future, with Medium-style moves for both sides. It picks the candidate that leads to the best position.

How strong are they?

Measured in duels with the real rules (lands, bandits, fog), each pairing on 150 maps with both seats swapped:

Match-up	Stronger wins
Easy vs. pure random play	72 %
Medium vs. Easy	77 %
Hard vs. Medium	84 %

The Hard computer needs about a tenth of a second per move on a 9 × 9 map, about a third of a second on 15 × 15 – on a phone-speed processor. That fits in its short "thinking" pause.

Computers you don't see

The same players work behind the scenes:

They set the star goals of every map (see Stars & goals).
They picked the campaign maps: for each station, dozens of candidate maps were played and sorted by how many stars the computer reached – early stations easy, the last of each chapter the hardest.
In online duels, when your time runs out, a Medium move is played for you.