Setting the CRS on a parcel boundary layer before it hits the registry

A parcel boundary layer can look perfect on screen and still fail a registry import because the CRS tag is wrong, missing, or inherited from whatever the digitiser had open last. Area comes out wrong, boundaries sit a few metres off the true line, and adjacent parcels that should share an edge suddenly overlap or gap. None of that shows up until someone runs the numbers against the cadastral index, usually after the batch has already been accepted.
Why the mismatch doesn't show up until area sums are wrong
Most registries validate geometry topology on import: no slivers, no self-intersections, parcels close properly. A CRS problem sails straight through that check, because the shapes are still valid polygons. They're just valid polygons sitting in the wrong place, or measured against the wrong ellipsoid, or scaled by a false easting that nobody carried over from the source file.
The tell is usually area. If a parcel layer was digitised or vectorised in a geographic CRS (lat/long, degrees) and loaded into a registry expecting a projected national grid, area calculations done on the fly will be off, sometimes by a lot, sometimes by a little, depending on latitude and how the tool handles the conversion. A 4.2 ha field that comes back as 4.6 ha or 3.9 ha in the registry isn't a digitising error. It's a projection that never got set.
ETRS89 or the national grid: what decides it
For most EU paying agencies, what settles the question is whatever EPSG code the registry's import spec names, a geodetic reference frame or a projected coordinate system, and those are two different things that get used interchangeably in conversation.
ETRS89 (EPSG:4258) is the datum, the reference frame fields and parcels are located against. Most national cadastral grids (Lambert 93 in France, ITM in Ireland, RD New in the Netherlands, and so on) are projections built on top of ETRS89 or a close national realisation of it. Store the data in ETRS89 lat/long for exchange, then reproject to the grid the registry expects for area and distance calculations, or do the reverse, depending on what the import spec says.
The place this goes wrong is a datum shift being handled as if it were a straight reprojection, or vice versa. A file tagged ETRS89 that's actually still in an older national datum with no shift applied will look fine in the attribute table and sit metres off in the field. GIS software won't catch that for you unless the source CRS tag was correct going in.
A short checklist before loading:
- Confirm the EPSG code the registry's import spec asks for, not the one your last project happened to use.
- Check whether the source imagery or vector was delivered in a geographic or projected CRS, and reproject explicitly rather than assuming the tool guessed right.
- Recalculate area after reprojection and compare it against the pre-reprojection figure. A difference beyond rounding means something in the chain used the wrong datum.
- Spot-check a handful of boundary parcels against known reference points, not just the interior of large fields where small shifts are harder to see.
- Keep the CRS metadata in the delivery file itself (a .prj or embedded EPSG code), not just in a covering email that won't travel with the data.
A layer that passes all of that is edit-ready: an officer can open it, nudge a boundary where a farmer disputes it, and load it without a second reprojection pass. That's the standard a vectorised boundary layer should be held to before it's handed to a registry team, and it's the standard we build to on Field Boundary Map, where each annual delivery comes tagged to the CRS your registry specifies rather than whatever the satellite processing chain defaults to.
If your registry's import keeps kicking back area mismatches on otherwise clean polygons, it's worth checking the CRS chain before you blame the digitising.