It would be interesting to run something like <a href="http://wiki.openstreetmap.org/wiki/That_Shouldnt_Be_Possible" rel="nofollow">http://wiki.openstreetmap.org/wiki/That_Shouldnt_Be_Possible</a> on the data, and aggregate the results to detect segments where OpenStreetMap data is likely to be wrong.