Statistical evaluation of information source performance

The present work examines the problem of evaluating the performance of statistically-characterized information sources when ground truth is unavailable. Although exact verification may be infeasible, inter-source statistical dependencies may be used to test for information consistency. Through application of a Rosenblatt transformation on an input sample and subsequent Kolmogorov-Smirnov test against the uniform distribution, a given information source can be statistically evaluated for goodness of fit. An algorithm is derived for detecting the presence of suspect information and identifying the associated aberrant source(s). The paper concludes with an example that considers the detection of a malfunctioning radar system in the absence of ground truth.