Quantification of search-extension benefits

This paper considers several search-extension rules and one pruning rule that have been described in the literature. An experiment was performed to see how effective each rule was in isolation, and in various combinations. The experiment was performed on a fixed testset of positions, and results measured using node counts. The emphasis of the work was to make repeatable measurements on well-defined tasks, for future comparison with other search-extension rules. In the test domain chosen, some extension rules were strongly advantageous compared with fixed-depth search, but disadvantageous in combination with others. Notably, singular extensions were strongly beneficial if added to a fixed-depth search, but detrimental if added to a search already using check extensions, recaptures and null moves.