Release Notes

Featuresmith v0.4.0

Featuresmith v0.4.0 completes Phase 4 of the roadmap: Recommendation & Planning. It adds a centralized Recommendation Engine, the FeatureQualityReviewer, and the Plan primitive.

Highlights:

• Recommendation Engine: Merges findings from every review section into a single ranked, explainable list with deterministic confidence scores and full traceability to originating findings and reviewers.
• FeatureQualityReviewer: 10th built-in reviewer detecting near-constant columns, redundant column pairs, and low-signal high-cardinality columns.
• Plan Primitive: fs.plan(result, accept=[...]) and featuresmith plan compile accepted recommendations into an inspectable, deterministic Plan of transformation steps.
• Score Dimension Reconciliation: Consistency and Data Quality dimensions consolidated (cardinality double-count eliminated); Class Balance omitted until its detector ships — 7 effective scored dimensions.
• Diff-Aware Review Continues: The v0.3.0 DiffReviewer remains fully supported via fs.review(source, previous=...) and featuresmith review --previous.

Featuresmith v0.3.0

Featuresmith v0.3.0 adds the DiffReviewer, integrating snapshot comparison directly into the Review Engine pipeline.

Highlights:

• DiffReviewer: 9th built-in reviewer that activates when a previous snapshot is provided, comparing the current dataset against it and attaching DatasetDiffResult to result.diff.
• Diff-Aware SDK Review: fs.review(source, previous=...) profiles the previous snapshot once and includes diff findings in the review.
• Diff-Aware CLI Review: featuresmith review --previous enables snapshot comparison with exit-code CI gating (exit 3 on missing/unparseable previous, exit 2 on unknown target column in either snapshot).
• GOVERNANCE.md: Baseline governance document for release, versioning, and contribution processes.

Featuresmith v0.2.0

Featuresmith v0.2.0 introduces the Review Engine, Dataset Diff comparison framework, ML Readiness Scorecard, and Intelligent Leakage Detection.

Highlights:

• Review Engine: Orchestrates multiple parallel dataset reviewers in a 5-stage pipeline, outputting structured reports.
• ML Readiness Score: Calculates a deterministic 0–100 score across 7 effective health dimensions (Missing Values, Duplicates, Leakage, etc.) with actionable feedback.
• Dataset Diff Engine: Compares two snapshot profiles to identify schema changes, distribution shifts, and quality regressions.
• Intelligent Leakage Detection: Features 6 pattern-matching detectors to catch target leakage, duplicate targets, and future information leaks.
• CLI Expansion: Introduces featuresmith review and featuresmith diff subcommands for terminal validation and CI/CD gating.

Featuresmith v0.1.0 (First Public Release)

Featuresmith v0.1.0 focuses on deterministic statistical summaries, tabular ingestion connectors, and rule-based validation audits.

Highlights:

• Tabular loaders mapping CSV, Excel, Parquet, and in-memory pandas/Polars DataFrames to normalized Dataset layers.
• Vectorized Profiling Engine computing 23 continuous numerical metrics, categorical frequencies, datetime ranges, text shapes, and correlation caps.
• Rule Engine executing 8 deterministic quality, statistical, and leakage audits.
• Command line tool (featuresmith-cli) for running audits in scripts or CI with structured exit codes.

Distribution Packages Scope

The following packages are officially published on PyPI for v0.4.0:

• featuresmith-core: Core engine library.
• featuresmith-cli: CLI thin wrapper client.

Note: featuresmith-dashboard is deferred to a future roadmap phase and is not published in v0.4.0.