Cortessa - Follow the evidence

Before the review, two versions arrive on the desk. The title is familiar. The revision is different. Which document governs the work, and are the references behind it actually present? The reviewer needs more than a search result. They need a route back to the wording, its context and its history. This is the problem Cortessa is being built around: controlled documents, questions that can be traced, and people who remain accountable for the decision. We will follow that route through the actual product, using public NASA documents as a demonstration. The people and workplaces are illustrative.

Start with Audit. It cross-references the selected corpus and surfaces documentary questions for human review: a superseded revision, an uncertain reference, or a cited document that is not held. On the graph, a document on one side points to the references it cites. Follow one relationship into the candidate behind it. In this public example, a held document cites a reference that the corpus does not hold. The candidate identifies the citing source and passage, so the next step is inspectable. The important result is not the colour of a warning. It is the evidence route: which document made the reference, where it appears, and what the held corpus contains. A missing document becomes a visible review question. That gap remains part of the evidence the reviewer examines. The reviewer can now decide what to inspect next.

Open the cited passage. Then open the original document. The source matters because a sentence can change meaning when it is separated from its surrounding wording, revision or applicability. A matching phrase alone does not settle the engineering question. Here, the reviewer follows the reference back to the page and examines that context. They can compare the relevant revision and check whether the wording governs this particular review. Cortessa keeps that route inspectable. It surfaces a cited candidate; the responsible specialist examines the evidence and decides what follows. That is the working relationship: the product brings the documentary issue into view, and the human challenges it against the source.

The same discipline applies when the reviewer has a specific question. In Ask, open the citation and inspect the source behind the response. This recorded public example is labelled Draft. A citation is available to inspect, while the response remains explicitly Draft, not a verified conclusion. Another recorded example is marked Refused. The product can decline to give an answer that the evidence does not support. Verified, Draft and Refused have different meanings. Verified establishes provenance: wording is anchored to a source. It does not establish correctness, applicability or approval. The reviewer still decides whether the passage answers the question and whether it governs the work.

They have a source to inspect and a question to resolve, rather than an unexplained instruction to accept. At a wind farm, a nuclear facility or a carrier deck, the scenery changes; the evidence question remains. In aviation planning, it may concern the documentary basis for the next human approval step. Before a police briefing, it may be a question about superseded guidance. These are illustrative situations, not customer deployments. The documentary task is the link between them. The authorised professional records the judgement through their organisation's process. Cortessa surfaces evidence for that review. Human authority stays with the people responsible for the work.

That evidence work begins locally. Each Cortessa engineer box holds its documents, ingestion pipeline, index and resident model. The on-device model is the standard generated experience, without a public-cloud dependency. Analysis stays on-box. If the selected model is unavailable, that state is made explicit. Separate source extracts are not relabelled as a successful generated answer. This keeps the working context close to the reviewer. It also keeps the organisation's access rules and approval responsibilities in the story. Sovereignty is a working boundary around the evidence and its use. The local product supports that boundary; a label on an answer does not replace human judgement.

The wider design connects those local boxes through departmental nodes to an on-site hub, entirely inside the site boundary. What is intended to travel is an explicitly approved, signed revision, carrying its lineage and permitted scope. A receiving reviewer still examines it. Arrival does not make it the governing source or automatically add it to another searchable corpus. The self-sufficient local product and governed exchange foundations exist. Authenticated peer transport, provisioning and production rollout remain in development. That distinction stays visible here: solid local foundations, a developing site network, and human approval at the points where authority changes. The aim is a document thread people can inspect, with scope and history preserved across the organisation.

Begin with one bounded document workflow and a corpus cleared for the evaluation. Name the reviewers and agree what evidence they need before the test starts. Include the awkward cases: uncertain revisions, missing references, difficult source material and questions whose answers are absent. Watch how people find the source, challenge a candidate and record their judgement. Then decide what Cortessa has demonstrated in your own review process, and what needs further work. See Audit and Ask against evidence you can inspect. Get in touch to arrange a Cortessa demonstration and discuss a scoped evaluation. Cortessa. Sovereign document intelligence.

Arrange a demonstration: info@cortessa.co.uk | cortessa.co.uk