feat: wire multilingual-bias-classifier-v1 into HF Space

fix(ui): add HAUSA and ZULU to Language enum — fixes AttributeError on HF Space

feat(ui): add Hausa and Zulu to Gradio UI — LANGS, metrics, examples

feat(zu): Zulu lexicon v1 + GT + precision gating — ZU F1=0.732 P=1.000 R=0.577

feat(ha): Hausa/Zulu lexicons, caller=studylabs gate skip, budurwa precision fix — HA P=1.000

feat(api): Hausa and Zulu rewrite rules, lexicons, and aibridge gate wiring

docs: update metrics to SW F1=0.851 P=0.822 R=0.881

feat(sw): add 69 Sheng lexicon terms — SW F1=0.851 P=0.822 R=0.881

chore(hf-deploy): sync v0.9 registry entry

feat(registry): add v0.9-sw-recall-boost — SW F1=0.851, R=0.881

fix(hf-deploy): restore pinned starlette/jinja2/anyio — fixes Jinja2 unhashable dict crash

chore(hf-deploy): force rebuild from latest SHA

chore(hf-deploy): trigger Space rebuild

chore(hf-deploy): trigger Space restart

chore(hf-deploy): sync main — SW recall boost +39pp, 28 new derogation patterns

fix(ui): ML metrics read from registry dynamically by active model version

fix(ui): dynamic ML model label + sys.path robustness for HF Space

fix: sync _preserve_case + statistical bare-count fix from main

fix: add COPY core/ to Dockerfile — core/ was missing from HF Space container

fix: ensure /app on sys.path + os.chdir in gradio_app.py — resolves core ModuleNotFoundError on HF Space

fix: use relative imports in core/__init__.py — fixes ModuleNotFoundError on HF Space

fix: make eval/context_checker.py self-contained — no core/ import for HF Space

fix: add repo root to sys.path in eval/context_checker.py — resolves HF Space ModuleNotFoundError for core

feat: sync main → hf-deploy — SW F1=0.840, EN=1.000, FR=0.970, KI=0.667

docs: update README with current metrics, architecture, API, dataset info

fix: restore EN + FR lexicons to correct state — EN F1=0.885, FR F1=0.793

fix: remove duplicate EN lexicon entries (539→69 rows) — restores EN F1=0.885

feat: auto-write eval/metrics.json on every eval run — gradio_app loads live metrics

chore: update SW metrics — F1=0.821, P=0.741, R=0.919

fix(sw): tester feedback — malaya detection, proverb partial match, wake/yake biographical suppression

fix: restore self-contained eval/context_checker.py — core import breaks HF Space

fix: add core/__init__.py so core is importable as a package

chore: sync core eval + detector files from main

chore: sync evaluator from main (ml_threshold=0.56 tuning)

chore: gitignore eval/results, eval/splits, CLAUDE.md

feat(sw): add missing lexicon entries flagged by tester + update metrics

chore: sync lexicon from main (missing terms fix)

feat(sw): expand derogation patterns + counter-stereotype suppressors — SW F1 0.816→0.819

fix: simplify to bare demo.launch() — drop middleware/rewrite on Space

fix: add middleware before launch() to avoid RuntimeError

fix: disable gradio analytics + show_error to suppress 401 on startup

fix: show N/A for EN/FR/KI in ML classifier registry rows

fix: call demo.launch() at module level so demo.app exists before Docker starts uvicorn

chore: restart after HF_TOKEN secret added

fix: add app_port: 7860 to README — required for Docker SDK health check

chore: force rebuild

fix: fall back to rules for EN/FR/KI when ML classifier selected, show warning

fix: restore model_registry.json + unignore eval/results

fix: lazy-load ML detector so Gradio starts fast on HF Space

fix: add requests dep missing from gradio 4.44.1 cli commands