🛡️ ModelRefiner Pro v3.5 — Multilingual QA Framework

---

🎯 Intended Audience

Translation QA specialists, LLM data quality managers, researchers

This repository provides the design documentation for a multilingual translation quality evaluation framework, integrating:

• EUQS (7-dimension scoring)
• RoPE (SVO retention / tone / rhythm + emotion / aesthetics / causality)

---

📖 Documentation

• docs/system_instructions.md → Full specification (8000 chars, transparency & reproducibility)
• docs/translation_matrix.md → Language pair thresholds (BLEU/ROUGE, RoPE priorities, cultural weights)
• docs/rope_metrics.md → Definitions of SVO retention, tone consistency, rhythm, aesthetics, causality
• docs/examples.md → Practical QA case guide

---

🧪 Examples

• examples/en-ja_sample.md → Japanese high-context QA example
• examples/en-de_sample.md → German technical translation QA
• examples/creative_eval.md → Creative marketing copy evaluation (JA / ES / DE)

---

⚠️ Notes

• This repository is intended as a reference for translation QA and LLM data quality improvement.
• Exact reproduction of system behavior is not guaranteed, as results may vary depending on model and environment.
• The focus is on framework transparency and applicability, not strict reproducibility.

---

📜 License

This project is licensed under the MIT License – see the LICENSE file for details.

---

🚀 Usage

Browse the examples/ folder to see how ModelRefiner Pro v3.5 can be applied to multilingual translation QA:

• en-ja_sample.md → Japanese QA example
• en-de_sample.md → German QA example
• creative_eval.md → Creative text evaluation

For detailed scoring dimensions and evaluation metrics, check the docs/ folder.