Tabular datasets change shape over time. This corpus has 44 small, versioned Parquet scenarios that record what changed: schemas, structural mutations, rows, and values. It leaves compatibility decisions to the system you test.
Each scenario contains versioned Parquet files and a scenario.json manifest.
The validator re-derives every declared fact from the files. A Parquet catalog
at data/scenarios.parquet makes the scenarios easy
to browse. See int64_to_float64
for a concrete example.
uv sync
uv run tec list
uv run tec inspect int64_to_float64
uv run tec validateYou can also use the fixtures directly without installing the Python package. See using the corpus for examples.
- Scenario inventory — all 44 changes, with a manifest example.
- Using the corpus — catalog queries and fixture consumption.
- Engine observations: PyArrow, DuckDB, Polars, Delta Lake, dlt, and dataprof.
- Development and releases — regeneration, contribution, and versioning.
- Design — the fact model and validation rules.
The dataset is also available on Hugging Face. Licensed under Apache-2.0; citation details are in CITATION.cff.