Most reasoning produced by AI systems today is consumed on faith. A retrieval summary arrives, a legal brief is synthesized, a compliance finding is issued — and the recipient has no means of auditing whether the underlying sources were actually consulted, accurately represented, or even real. The gap between output andevidence is papered over by brand trust, service-level agreements, and the quiet assumption that the system probably got it right.

NORA Foundation exists to close that gap in the open. We are building public-interest infrastructure for verifiable reasoning: an open standard called Canon, a reference implementation called Meridian, and an ecosystem of independent verifiers that anyone can write, anyone can run, and anyone can audit. The standard is published under CC0. The implementation is built in the open. The verifier ecosystem is something we actively seed but do not control.

Our work was forged in a demanding real-world evidentiary context — a setting where retrieval and interpretation of a large, heterogeneous archive had to withstand adversarial review by parties with no reason to cooperate with the system producing the outputs. That context was the proving ground. What emerged was a general problem, not a single-case tool: how do you give a recipient enough structure to independently falsify the claims an AI system makes about its own reasoning?

The answer, we believe, is a specification. Not a product, not a platform, not a trust mark issued by a single authority — but a machine-readable contract between issuer and recipient that makes every claim checkable without cooperation. That is Canon. Everything else NORA builds is in service of making Canon real, robust, and widely adopted.

We invite adoption across every field where independently-checkable reasoning matters: law, journalism, research, audit, archives, public records. In each of these domains, the people who rely on AI-produced artifacts are not the people who built the system — and they have every reason to demand proof.

Canon does not ask anyone to trust NORA. It asks systems to produce artifacts that are checkable without NORA. That distinction is the entire point. If we disappeared tomorrow, every Canon attestation already sealed would remain independently verifiable by anyone with the artifact and ordinary network access. That is the bar.