How can a game designer tell if the software implementation of a game behaves as intended ? Our approach is to provide a tool that automatically produces succinct, accurate summaries of the game's behaviour in constrained scenarios, and presents them to the designer in a visual, intuitive, and interactive form. A working prototype of such a tool for Electronic Arts' FIFA'99 soccer software will be demonstrated.

The key technologies underlying this tool are:

1. a method for automatically generating a large set of informative samples of game play
2. the use of machine learning to summarize the samples with a small set of easy-to-understand rules characterizing the game's behaviour
3. an interactive visualization system that presents the learned rules in an intuitive way, and allows the designer to explore the behaviours interactively.