One backend holds the truth. Every workbench, AI component, and integration is a thin client over the same governed model — never a second copy of it.
The backend owns the model — the NFSI requirements, the three relationships, the library and its governance rules. Nothing else stores a rival copy. The workbench renders it, the AI proposes against it, and your existing tools read and write it through a single interface. Change it once, and everyone sees the same change.
Clients on top. One authoritative engine in the middle. The library beneath. Knowledge and AI feed in from the side — each part replaceable without touching the model.
Because the governed model is exposed as a REST API and a Model Context Protocol server, three very different clients meet on the same ground:
Engineers author requirements, trace links, and run reviews in a web app that is only ever a view of the backend.
Norn and your own agents reason over the methodology in its own terms — proposing against the live model, never a stale export.
DOORS, Jama, or Cameo read and write the same requirements. Integrate the engine behind the stack you already run.
The AI is a tier, not a dependency. Remove it and the methodology still stands — the model, the governance, and the workbench all keep working.
Proposes the structured requirement set from stakeholder needs, layer by layer. Every proposal passes a human review gate.
Model-free retrieval fed by your library. It works only from your data, so in practice it cannot hallucinate.
The generative half — drafts what the library is missing, always for a human to decide. Kept inside the airgap.
The same stack runs in three postures. You choose by data-sovereignty and risk — the methodology does not change between them.
Backends on Railway with Postgres alongside; workbenches on Vercel. Git-push-to-deploy, no infrastructure to run.
The whole stack as Docker containers on infrastructure you control — inside your own network and policies.
A self-contained box — database, retrieval, and UI — running disconnected. Nothing phones home.
Both frontends are deployed now. Open them and see the model rendered — the same backend, two views of it.
Requirements, traceability, architecture, design decisions, and the Norn derivation engine — the full engineering environment.
Open the workbench ↗ ᚠThe failure-propagation canvas over the same requirement set, scoped per solution node, with mitigation writeback.
Open the workbench ↗