Felixe mentioned:

"I would like to know, how can you be sure that a program can find all the best moves for the defensive king?"

The database starts by placing the pieces on every square and asking the simple question: Is the side to move in checkmate? Yes, identify this as a loss in 0. No? Is the side to move in stalemate? Yes? Identify this position as a draw. No? Can the side to move win one piece? Yes? identify this position as a draw. No? There is not enough informaiton to resolve the position yet, keep looping.

Eventually, it finishes "pass #1" and has identified all of the losses and draws. Then, by definition, every position on subsequent passes must be either play into: a pre-resolved win, a pre-resolved loss, a pre-resolved draw, or remain undetermined.

In order to resolve a win, only one move for the winning side need to lead to a win. In order to resolve a loss, EVERY move for the losing side must lead to a win for the other side.

As you loop around, doing pass after pass, you have a variable that indicated whether or not you resolved something on a particular pass. After so many loops, there may be nothing left to resolve. By definition, any of the UNKNOWN positions must be draws, so you mark them as draws then you are done.

Why you consult the database in a position, here is what you do:

1. Determine if you are in a win, loss, or draw to start. This is stored in the db.

2. Generate all legal moves from the position.

3. If in a win, look up the # of moves to lose for the other side to move. Pick the smallest number you find, and make the move leading to that position.

4. If in a loss, look up the # of moves to win for the other side to move. Pick the largest number you find, and make the move leading to that position.

5. If in a draw, make sure you do not move into a loss! (It is impossible to select a move leading to a win, or it would not be a draw.)

In this fashion, the program can always play the best defensive move for the weak side, and fastest winning move for the strong side.