In 1935, the statistician Sir Ronald Fisher introduced a class of statistical significance tests that are theoretically excellent but computationally very demanding. Until now, this has limited their practical usefulness, but we have discovered a radical new approach that allows the Fisher test to be performed at a cost that scales only linearly with the number of statistical data, thus rendering it practically feasible even for very large datasets. We hope to use a similar strategy to reduce the cost of electron correlation methods, such as Full CI.