The project ships reproducible benchmarks including authored144.jsonl (144 labeled decision rows) and a larger frozen evaluation matrix, and documents direct-readout speed advantages over compact JSON generation on shared Qwen3.5-4B weights. LangChain hosts a gateway model semif-qwen3.5-4b aligned with the System One API; it is independent of TypeSafe AI's commercial Jev service.