Sophia — the Wisdom Gate
Provenance-aware, verifier-gated reasoning layer for LLMs with bilingual benchmarks and CI-gated measurement.
- Attribution hallucination reduced on a scoped local-model run (0% FP cost)
- 528 bilingual EN/ZH eval examples (HuggingFace)
- Zenodo DOI 10.5281/zenodo.20930874
- SimpleQA selective-prediction calibration (cross-model)