File formats
Every file Mink serves is pinned by a specification in the repository, with conformance fixtures that tests check byte for byte.
| Specification | Pins |
|---|---|
file.md | A data file's path, Parquet bytes and statistics |
records.md | How Kafka record batches become a file's rows |
decoding.md | How Avro, Protobuf and JSON Schema payloads become typed columns |
schema.md | Table schemas, type mappings, field ids and evolution |
layout.md | Time partitioning, assembled files and manifest grouping |
metadata.md | Snapshots, table metadata, manifest lists and manifests |
statistics.md | The Puffin statistics file with per-column distinct-value sketches |
Fixtures live in conformance/fixtures/. Independent readers check them: DuckDB, PyIceberg, Apache DataSketches and a real Kafka broker.