ksanyok/TextHumanize v0.30.0 on GitHub

This release closes the public roadmap to 100% and ships it as tested, offline library capabilities.

Promopilot product layer (`texthumanize.product`)

audit_widget_html() — self-contained paste-text AI & watermark audit widget (score, risk map, safe improvements).
audit_batch() — bulk detector/watermark/readability/quality audit for sitemap/folder workflows.
compare_versions() — original / AI draft / humanized / editor-final scoring with pairwise diffs.
content_plan_risk() — pre-publication publish/review/block gate.
make_brand_voice() + brand_voice_lock() — lock brand terms while humanizing.
client_report_html() — neutral, print-ready client report (no detector-bypass claims).

benchmark_leaderboard(), release_snapshot() (before/after + latency p50/p95), acceptance_rate(), semantic_drift_rate(), watermark_eval(), count_regression_examples(), funnel_metrics().

load_bad_output_bank() turns every known-bad case into a permanent regression test.
humanize(..., with_quality_score=True), attach_quality_score(), explain_html(..., include_quality_score=True), and the new texthumanize widget / texthumanize leaderboard CLI commands.

check_version_sync.py now blocks stale hardcoded version assertions; scripts/dev_check.py is a fast offline pre-release guard.
Packaged watermark fixtures were double-escaped and not real watermarks — now genuine zero-width/homoglyph/invisible characters.

Full details in CHANGELOG.md.

🤖 Generated with Claude Code