Skip to content

geomlint

Monte Carlo for geometry columns. The checks every mainstream data-observability tool skips, because it treats a geometry column as an opaque blob.

Get started View on GitHub

$ geomlint check ./parcels/
SEVERITY CODE                 FILE                         FEATURE      MESSAGE
-------------------------------------------------------------------------------
ERROR    invalid-geometry     parcels/block_14.geojson      parcel-001   topologically invalid: Self-intersection[0.5 0.5]
ERROR    coordinates-out-of-wgs84-range parcels/poles.geojson -          12/40 feature(s) outside valid lon/lat bounds — likely UTM mislabeled as WGS84
WARNING  mismatched-crs-across-files (batch) -               -          multiple distinct CRS declarations found in this batch

2 issue(s) — 1 error, 1 warning
$ echo $?
1

What it catches

  • Geometry & topology

    Self-intersections, collapsed rings, zero-area shapes, duplicate vertices, RFC 7946 ring winding, antimeridian crossings, fake coordinate precision.

    Checks reference

  • CRS sanity

    Unrecognized or missing CRS, coordinates that fall outside WGS84 bounds, mismatched CRS declarations across a batch of files.

    Checks reference

  • GeoJSON, Shapefile, GPKG

    GeoJSON parsed natively, no system dependencies. Shapefile/GPKG behind an optional extra — still no GDAL to install.

    Getting started

  • Configurable, CI-native

    .geomlint.toml for thresholds and severities, a --fail-on gate for CI, a GitHub Action, and a pre-commit hook.

    CI integration

Why this exists

Bad geometry and CRS confusion don't throw exceptions — they produce quietly wrong maps, quietly wrong joins, and quietly wrong distance calculations that nobody notices until a downstream report is off and nobody can say why.

ST_IsValid and friends exist in every spatial database, but they're one-off functions you have to remember to run — not a check suite that runs in CI on every pull request or pipeline load, the way great_expectations or a dbt test does for tabular data. geomlint is that check suite for the geometry column itself.

Not yet implemented, on purpose

Positional/GPS-drift detection (multipath, sensor jumps) isn't a check here — see geomlint/checks/drift.py for why a naive distance threshold would just be noise, and the roadmap for doing it properly.

Where to go next