v1.4.0 — Scoring Quality + Smart Auto-Config

What's New

Scoring & Survivorship Quality (#30)

• MST-based cluster auto-splitting — oversized clusters split at weakest edge automatically
• Cluster quality labels — strong, weak (confidence downgraded), split (auto-split)
• Quality-weighted survivorship — merge strategies use GoldenCheck quality scores
• Field-level provenance — tracks source row, strategy, confidence per golden record field

Data-Driven Strategy Selection (#32)

• Learned blocking auto-selected for datasets >= 5000 rows (96.9% F1 matching hand-tuned)
• Cross-encoder reranking enabled for weighted matchkeys with 3+ fields
• Adaptive thresholds from data quality: -0.05 for high null rate, +0.05 for short strings

LLM + Memory Auto-Enablement (#36)

• llm_auto flag — dedupe_df(df, llm_auto=True) auto-enables LLM scorer ($0.05 budget cap) and memory store when API key detected
• Applied uniformly across all config paths (zero-config, explicit kwargs, YAML config)
• Memory store (SQLite, persistent corrections) enabled alongside LLM scorer

Install

pip install --upgrade goldenmatch

Full Changelog: v1.3.2...v1.4.0