geographical accuracy

[Graham Seaman]

Hi all,

I'm trying to generate a set of latitude/longitude points for use with 
google maps (to avoid having to query google for the actual locations). 
What I have as source data is a set of northing/easting values with 6 
digits for northing, 6 digits for easting. I'm using 
Geography::NationalGrid::GB to convert these to lat/long pairs. The 
results are all approximately correct, but generally about 20 metres or 
so off in an apparently random way. The features I'm trying to locate 
are mostly quite large and clearly visible on google's satellite/hybrid 
view (but I have around 20,000 locations to deal with, so can't correct 
them all manually). I have seen the same features located accurately on 
another google map application, so I know the problem is not at the 
google end of things.

I'm trying to pin down where the errors are coming from, and what I can 
do about them.  I believe that the northing/easting values have enough 
precision to give an accuracy of a metre (from a back-of-the envelope 
calculation...). However, it's possible that they were originally 
derived from post office data (ie. corresponding to postcodes) in which 
case I've read that the accuracy is limited to 10 metres, and the last 
digit is effectively noise. Then there's the question of the accuracy of 
the conversion - I assume there's some inevitable distortion in 
converting from a projection of a sphere onto a flat surface, followed 
by some loss from rounding in the calculations, but don't have an idea 
of how large this is likely to be.

Can anyone with a GIS background tell me if my assumptions are wrong 
here, and more to the point if there's anything obvious I can do to 
improve my results?

Cheers
Graham