Install & first validate

pip install laterite

That one line gives you the library and the lat CLI. The base install is polars + duckdb — pyarrow-free, no pandas. Optional extras add drop-in surfaces only if you want them:

pip install laterite[compat]          # the python-ags4 drop-in shim (adds pandas) — pyarrow-free
pip install laterite[compat,pyarrow]  # + pyarrow accelerator (faster pandas hop + string dtype)
pip install laterite[pyarrow]         # the Arrow backend (adds pyarrow)

[compat] alone is already ~2× faster than python-ags4 (object-dtype pandas via DuckDB). pyarrow is an optional accelerator — see Dependency shape.

First validate, from the command line

Point lat at an AGS4 file. A clean file says so and exits 0:

lat validate examples/sample_site.ags

examples/sample_site.ags: clean (0 findings)

Break one value — say a 2DP easting that isn't a number — and the same command prints a findings table and exits 1:

lat validate sample_site_typo.ags

sample_site_typo.ags: 1 finding(s)
Rule | Line | Group | Description
-----+------+-------+--------------------------------------------------------------------------
8    | 51   | LOCA  | Value "not-a-number" in LOCA_NATE does not match its declared TYPE "2DP".

Each row is one numbered-rule violation: the Rule that fired, the Line and Group it landed in, and a one-line Description. Here Rule 8 caught a value that doesn't match its column's declared AGS TYPE.

Need machine-readable output? Add --json — findings is an empty object when the file is clean:

lat validate examples/sample_site.ags --json

{
  "file": "examples/sample_site.ags",
  "findings": {}
}

The exit-code contract

lat exits 0 when the file is clean and 1 when there are findings — nothing else. That makes it a drop-in gate for CI or a pre-commit hook: lat validate delivery.ags fails the step the moment a rule fires, no output parsing required.

Next → Read & explore a file