DataJig 0.8.0

DataJig 0.8.0

DataJig 0.8 introduces a reviewed, recoverable pipeline from local raw data to verified training delivery.

Highlights

• One YAML contract for CSV, Parquet, or JSONL inputs, bounded DuckDB SQL, dataset revisioning, training-bundle export, and consumer plans.
• Deterministic pipe_... authorization identities and independently verifiable piped_... success receipts.
• Crash-safe create/update/no-op execution with durable journals, explicit --resume, OS locks, atomic no-replace delivery, and compare-and-swap workspace HEAD updates.
• End-to-end lineage across source bytes, transform evidence, immutable revision, bundle, and consumption plans.
• Strict YAML, path, provider, SQL, input-drift, and receipt validation designed to fail closed for agent-driven workflows.

Install the core:

python -m pip install datajig==0.8.0

Install the pinned DuckDB transform provider:

python -m pip install 'datajig[duckdb]==0.8.0'

See CHANGELOG.md for the complete release notes.