GoldenMatch v1.5.0 — Auto-config verification layer

[Ben Severn (benzsevern)]

GoldenMatch v1.5.0 is out 🟡

What's new

Auto-config now has a preflight + postflight verification layer. dedupe_df(df) with no arguments is now runtime-safe on datasets that used to crash (DBLP-ACM, biblio, product descriptions).

• Preflight (6 checks) runs at the end of auto_configure_df — column resolution with domain auto-repair, cardinality bounds, block-size sanity, offline-safe remote-asset demotion, confidence-gated weight capping. Raises ConfigValidationError on unrepairable issues; attaches PreflightReport to the config.
• Postflight (4 signals) runs inside the pipeline after scoring — score-histogram bimodality, blocking recall, preliminary cluster sizes + bottleneck pair, threshold-band overlap. Attaches PostflightReport to DedupeResult / MatchResult.
• Offline-safe by default — embedding / record_embedding / cross-encoder rerank are demoted unless allow_remote_assets=True. CI won't silently download models.
• Strict mode (strict=True) keeps signals + advisories but suppresses adjustments for deterministic parity runs.

Classifier fixes

• Cardinality guard prevents unique-ID columns from being misclassified as phone / zip.
• New year and multi_name col_types.
• Low-confidence fields cap at weight 0.3.

Install

```bash
pip install --upgrade goldenmatch
```

Examples

• verification_inspection.py
• strict_mode_parity.py

Read more

• Release notes
• Wiki: Verification
• PRs #44 (feature) and #45 (release prep)

Feedback and bug reports welcome here or in Issues.