Bake direction (kn → en / en → kn) into dataset.name so HF sidebar clearly distinguishes them — fix apparent-duplicate UX

Fix model-index: add dataset.config per direction to prevent HF deduplication of kn↔en entries

Expand model-index: split FLORES kn↔en, add IN-22 conv (4 metrics × 2 directions) + Entity Preservation NER. 5 task entries total, ranked metrics: bleu/chrf/comet/cometkiwi/accuracy

Model card: link docs/ folder (journey, lessons, recipes)

v2.4 priorities locked: multi-token code-mix data slice (#1) + proverbs corpus

SDK 0.1.3 + CPU torch install hint in model card + Documentation URL on PyPI

Replace stray § symbol with "Section " (model card, DEPLOYMENT, SDK 0.1.1)

Model card: Available releases section linking int8 repo + SDK + demo

Model card §8: verified deployment matrix + pinned versions + honest not-supported list

Add Break-the-Model Challenge invite to model card + PRIVACY.md with full disclosure of opt-in research dataset

Add IN22-Conv scores to model card §4.4 — all floors cleared

Move full deployment guide to DEPLOYMENT.md; model card carries only a compact reference

Tone down NLLB claim + soften 100% claim + split detail into TRAINING_GUIDE.md + add citation-grounded comparison context

Add Section 8 Deployment Guide: matrix + recipes for local, HF Endpoints, Spaces, FastAPI, Docker, ONNX, quantization, batched inference

Expand training data section: public corpora citations + internal-generated specialized streams + scoring/eval set references

Reframe form-data section: lead with 100%-preservation capability, EN→KN PAN as minor edge case

v2.3 release — single-register retrain, FLORES BLEU 27.20/18.50, COMET 0.8459/0.8443; style endpoints hidden from API

Scrub internal CONTROLMT.md cross-references from README.md

Fix HF auto_map (add tokenizer_config.json + list form); honest style framing + add curated eval scores

Replace section sign (§) with spelled-out "Section" in README + CHANGELOG

ControlMT v2.2 initial release