WGS84 vs OSGB36: Why Your GPS and Your OS Map Disagree
GPS coordinates and the British National Grid are built on two different models of the Earth. WGS84 is a global best fit, computed from satellites; OSGB36 is a 1936 triangulation on the Airy 1830 ellipsoid, shaped to hug Great Britain. The same hilltop's latitude and longitude differ between them by dozens of metres — up to around 120 m — which is why pasting raw GPS numbers onto an OS map puts you in the wrong field. Transformations bridge the two: a single Helmert transform gets within a few metres (plenty for walking), and Ordnance Survey's OSTN15 grid gets within centimetres (what surveyors use).
What a datum actually is
The Earth isn't a neat shape, so every coordinate system starts by choosing an ellipsoid — a smooth stand-in — and deciding exactly where to park it. That parked ellipsoid is a datum. WGS84 uses an ellipsoid sized and centred to fit the whole planet, because satellites orbit the whole planet. OSGB36 uses Airy 1830, sized and tilted to fit Great Britain as snugly as possible, because in 1936 the job was mapping Britain, not the world. Neither is wrong; they're answers to different questions — and the same point on the ground gets different numbers in each.
How big is the disagreement?
Treat a WGS84 latitude/longitude as if it were OSGB36 and you'll plot a point that's shifted by roughly 50 to 120 metres, varying smoothly across the country. That's not a rounding error — it's a couple of fields, the wrong side of a river, the wrong crag. It's also why the National Grid numbers your GPS app shows are never the raw satellite output: something, somewhere, ran a datum transformation first. The only question is which one, and how accurate it was.
The Helmert transform — seven numbers that move a world
The workhorse conversion is the Helmert transformation: shift the whole coordinate frame along three axes, rotate it about three axes, and scale it — seven published numbers. Applied to Britain, Ordnance Survey's national Helmert fit is accurate to about 3 metres. The limit isn't sloppiness in the seven numbers; it's that OSGB36 itself is subtly warped. It was built by triangulating hilltop to hilltop with 1930s instruments, and those accumulated stretches can't be undone by any single rigid shift-rotate-scale. Our grid reference tool uses exactly this transform and states its accuracy honestly — comfortably inside a six-figure square.
OSTN15 — the survey-grade answer, and its trade-off
To beat the warp you stop treating Britain as rigid: OSTN15 is Ordnance Survey's official transformation, a 1-km grid of easting and northing shifts covering the country — a rubber sheet that absorbs the triangulation's local distortions. It reproduces definitive coordinates to about 0.1 m, and it's what professional GNSS gear and GIS software apply. The trade-off is the grid file itself: megabytes of shift data that must ship with the software. For property boundaries and engineering, that's non-negotiable. For reading out a 100 m square on a hill, the 3-metre Helmert answer is the right tool — instant, offline, and more precise than the GPS fix feeding it.
What this means in practice
Walking, geocaching, rescue call-outs: any modern tool that says it outputs National Grid has done a transformation for you — the practical question is only whether it's Helmert-class (metres) or OSTN15-class (centimetres), and for a person with a phone the difference is invisible next to the GPS fix itself (test yours on the GPS accuracy page). Surveying, mapping data pipelines, anything legal: insist on OSTN15. And when a coordinate looks about a hundred metres wrong, you now know the first question to ask: which datum?