It applies constraint propagation to reduce search space.

It uses gradient methods to find solutions.

It only moves forward in the search space.

It evaluates nodes based on a minimax criterion.

It returns to earlier states and tries different choices.