Wrangle 1.0 prepares scientific data with native Polars. Researchers and agents use the same inspect and prepare engine, compact YAML protocols and readable CLI.
This is a breaking migration from the legacy pandas/NumPy and neural-network preparation interfaces. Use Python 3.10 or newer and follow the migration guide before upgrading.
- Guided
startrecords observation identity, units, missingness, exclusions and metadata matching; unresolved scientific decisions block preparation. - Preparation enforces declared join cardinality, ordering, numeric domains and observation transitions. Results retain checked data, the YAML protocol, reports and input/output evidence.
- Disk preparation keeps complete source/intermediate snapshots and file-backed evidence. Global operations still require memory; inspection and guided start are eager.
- The package ships a short agent entrypoint, 123 generated operation contracts and eight executable workflows. Unicode scientific labels, timezone handling and saved evidence work across Linux, macOS and Windows.
- Release controls test the actual package, scan dependencies and code, verify repeat builds and produce signed provenance for the wheel, source archive and checksums.
The artifact workflow attaches the distributions, SHA256SUMS and attestation.json after successful validation and signing. Require all four assets and follow the verification instructions before installing. Receipts establish consistency; the researcher remains responsible for scientific meaning.