> I gave up looking for an accuracy figure on wikipedia, but knew I had seen 1.5 metres in the past, and that 10 metres is too inaccurate.

It all has to do with the statistical error bound. For a specified accuracy, one must also state the deviation for which that error is true. My point is that GPS doesn't have a specific accuracy, always true, with a probability cliff on each side. For a given accuracy specification, one must always include the probability for that accuracy.