Last released Jun 23, 2026
A fail-closed verification layer you wrap around your LLM: it never lets an unsupported claim through as ACCEPT, re-derives what is re-derivable, and emits a deterministic verifiable reward.
Supported by
