What we are building, clearly.

The planned design keeps source scans in storage you control. Nitsor would store references and the records needed to explain changes and decisions. A released product still needs to prove temporary processing, caching, export, retention, and network behavior, so this is not yet a deployed guarantee.

In the planned version model, a commit records one meaningful dataset change and links it to the earlier state, its author, and the relevant evidence. Source references, labels, label definitions, transformations, and review decisions remain distinguishable. A commit is a proposal, not an approval or release.

This behavior is still planned. The goal is to detect meaningful overlap, show both proposals, and send a real disagreement to an authorized reviewer who records a reasoned decision. The exact comparison and merge rules are not available until they are proved by runnable product tests.

The intended model gives each automated tool its own identity and only the permissions it needs. Its work should stay identifiable, and a protected decision should require an authorized human who did not create the proposal. No public agent interface or production permission system is available today.

This remains open. Nitsor intends to publish useful parts of the product and support a self-hosted path, but the licence and package boundaries are not settled. No public SDK, command-line tool, manifest format, or supported self-hosting package is available today.

Not as a self-serve production product. Nitsor is pre-release and is looking for a small number of design partners with recurring industrial CT, CT, or MRI work. Any evaluation is supervised and must be agreed directly; the public site does not promise acceptance, production access, or a service level.