I'm reprojecting a set of measured GPS co-ordinates from WGS84 to OSGB36, using two methods: the pyproj.transform method, which uses PROJ.4, and a "homemade" method from here. The latter method is accurate enough, but I'd like to know how best to measure its accuracy compared to the PROJ.4-based transform (both with and without the OSTN02 datum shift).
There is generic transverse Mercator test data available as part of the US NGA G & G Gold Data.
The oil and gas industry also has test data available through the Geospatial Integrity of Geoscience Software (GIGS) initiative.
You can also generate sample test data for OSTN02 and BNG via the Ordnance Survey's online converter.
Control points could be used. Get a differential GPS base-station collect a number of points from around the UK in both BNG and WGS84. Convert the results and look at the offset to the ground truth differential GPS coordinates.
If you are in a University setting, I would imagine lots of University geography departments around the UK would have base-stations set up. In addition the OS might help.