v1.1.0

v1.1.0

New Features

• Fast-path exit for clean sentences — skips contextual strategies when no structural errors found, dramatically reducing FPR on clean text
• Polite register tier — 3-way register detection (formal/informal/polite)
• ParticleChecker and TenseAgreementChecker wired into grammar engine
• G2P phonetic mappings, named entity gazetteer, and loan word rules
• Benchmark expansion from 489 → 719 sentences
• Clean-text FPR corpus for false positive regression testing

Improvements

• Semantic MLM suggestions sorted by edit distance before logit score
• Shared false_compound suppression with text-level compound detector
• Tuned dangling_word detector and expanded false_compounds list
• Reduced FPR: disabled G04, fixed confusable logit_diff, added terminal SFP filter
• Surface semantic model load failures when paths explicitly configured
• NER data externalized to YAML + SQLite schema

Infrastructure

• GitHub Actions CI with path filters (skip CI for non-code changes)
• Bumped actions/checkout v6, actions/setup-python v6
• Coverage threshold adjusted 75% → 70% for new module additions
• 4,549 tests across 213 files