Abstract: The ability to automatically hoard data on a computer's local store would go a long way towards freeing the mobile user from dependence on the network and potentially unbounded latencies. An important step in developing a fully automated file hoarding algorithm is the ability to automatically identify strong relationships between files. We present a mechanism for visualizing the degree of long-term relationships inherent in a file access stream. We do this by comparing the performance of static and dynamic relationship predictors. We demonstrate that even the simplest associations (from a static/first-successor predictor) maintain relatively high accuracy over extended periods of time, closely tracking the performance of an equivalent dynamic (last-successor) predictor. We then introduce rank-difference plots, a visualization technique which allows us to demonstrate how this behavior is caused by stable static pairings of files that are lost by the adaptation of the dynamic predictor for a substantial subset of frequently accessed files. We conclude by demonstrating how a third pairing mechanism can make use of these observations to outperform both the dynamic and static predictors.