Scout and Watchtower audit the AI. SitRep audits the data — the material underneath it. Every AI answer, every RAG deployment, every human decision made off a reference document inherits the errors in the source: a number that conflicts between two files, a policy that two documents state differently, a claim in a finished report that the underlying records don’t actually support. Those errors don’t announce themselves, and no model can grade around them — garbage in stays garbage. SitRep is how you find it first.
SitRep establishes what your authoritative source material actually says — deterministically retrieved, cross-vendor, cited — and then does one of three jobs against it. It can verify a finished document claim by claim, check your material against itself for internal contradictions, or find the questions your material can’t answer. Because it runs offline, it doesn’t sample and estimate the way a live monitor must — it performs a full audit of the whole body of material.
Two of those jobs rest on opposite premises, so SitRep keeps them cleanly separate. Verify trusts your source data and uses it to judge a document. Audit your source (consistency and coverage) distrusts the material and interrogates it directly. You pick which you’re doing; SitRep never quietly conflates “is this document right?” with “is my source any good?”
It is cross-vendor by construction: one vendor’s model pulls the discrete, checkable claims out of the prose and tables, a second vendor’s model derives what the source actually says, and a third judges whether the claim truly follows — so no single model both proposes and confirms. The judging is built to catch the errors that slip past naive checks: it’s magnitude- and unit-aware (millions isn’t billions), reasons about entailment (“three” supports “more than two”), stays attribution-aware (an analyst’s figure isn’t the company’s claim), and scopes to the right entity (metformin’s warning is not warfarin’s). Above all it is safe-failing: a claim it can’t establish is marked unverified, never turned into a false accusation.
When your source includes scanned or image documents, SitRep reads them with fail-safe OCR — several independent engines have to agree on a number before it’s trusted; where they disagree, it abstains rather than confidently misread. Aggressive about coverage, never about trust. A verify run ends not just in a report but in a redlined copy of your document — each contradiction struck through with a corrected line and the source beside it, in the document’s own format. And it all lands as one sealed, signed, independently verifiable package. Deploying into Europe? SitRep is built for the EU AI Act’s data-governance obligations (Article 10 / the Annex IV data provisions): the evidence a provider’s technical file relies on that the training and reference data is accurate and representative — evidence, never a claim of compliance.
