If White plays at H, G and H are connected. Playing at H also threatens to connect H and L. However the connection between G and H is not sente on the connection between H and L: if White plays at H, Black answers at L, and White has to play at J to keep G and H connected. We say the connection between G and H is not sente on the connection between H and L.

On the other hand, for Black, the connection between I and J is sente on the connection between J and K. If Black plays at J, I and J are connected. Playing at J also threatens to connect J and K. The connection (I,J) is sente on the connection (J,K) because whatever White plays after Black J, either it does not threaten the connection between I and J, or it threatens it, but Black has an answer that both connects I and J, and keeps the threat of connecting J and K.

A common approach to Go programming is to compute the status of tactical goals. Examples of tactical goals are connecting two strings, capturing a string or making a group live. The status of a tactical goals is assessed using heuristic search. Once unsettled goals are found, they are evaluated and the one with the highest evaluation is played. Recently, I have shown how to evaluate unsettled goals using a Monte-Carlo approach [Cazenave and Helmstetter 2005a]. It consists in evaluating an unsettled goal with the average of the random games where it has been reached. I build on this approach in this paper.