Metrics for Better Puzzles Cameron Browne Computational Creativity Group Imperial College London
[email protected] Abstract. While most chapters of this book deal with run-time telemetry observed from user data, we turn now to a different type of metric, namely design-time metrics for level generation. We describe Hour Maze, a new type of pure deduction puzzle, and outline methods for the automated generation and solution of lev- els. Solution involves a deductive search that estimates a level’s iterative and strategic depth. This information, together with sym- metry analysis of wall and hint distributions, provides useful met- rics with which levels may be classified and described. Results from a user survey indicate that players’ enjoyment of computer- designed levels may be affected by their perception of whether those levels are indeed computer-designed or handcrafted by hu- mans. We suggest ways in which puzzle metrics may be used to increase the perception of intelligence and personality behind level designs, to make them more interesting for players. Keywords. Logic puzzle; Hour Maze; Nikoli; deductive search; quality metrics; procedural content generation. 1 Introduction The current crop of smart phones and handheld game devices are the ideal plat- form for logic puzzles, which are enjoying a surge in popularity due to the main- stream success of titles such as Sudoku and Kakuro. Such puzzles can be played easily on small screens without losing any of their appeal, and can provide a deep, engaging playing experience while being conveniently short and self-contained. Fig. 1 shows a new logic puzzle called Hour Maze currently under develop- ment for iOS devices.