The analyst’s loop, end to end.
Collect. Correlate. Fuse. Test. Package. Every step is traceable, and every claim in the final product carries the graded evidence beneath it.
01 · Collect
Set an area of interest — a place name, coordinates, or an MGRS grid — and pull in the feeds your team is granted: vessels, aircraft, earthquakes, weather, infrastructure, conflict events, and any feed your organisation adds. Contacts land on the map graded for reliability, and are shared with your whole team the moment they arrive.
An administrator can define a new feed from a form — no code. A dry run shows exactly what will be imported before anything is saved.
02 · Correlate
The correlation engine finds meaningful pairings across your entities — close in space, close in time, co-incident — plus density clusters and same-entity candidates. A vessel and an aircraft loitering at the same point within minutes of each other stops being two dots and becomes a relationship.
03 · Fuse — indications and warnings
Fusion runs pattern-of-life over movement and computes multi-source convergence: where several independent sources agree that something is happening. Indicators are severity-ranked and weighted for recency and source reliability, so genuinely significant activity rises instead of drowning in volume.
04 · Test your own judgement
Analysis of Competing Hypotheses lays every plausible explanation side by side and scores the evidence against all of them — the analytic opposite of cherry-picking to confirm a favourite. A timestamped notebook captures your reasoning as you go, with entities referenced inline. Chain-of-reasoning, beside chain-of-custody.
05 · Package
VALIS assembles your work into a structured intelligence product — an INTSUM and a briefing deck — that a decision-maker can act on. Every indicator traces to its contributing entities, every entity to its source and Admiralty grade, every assessment to the notebook entries and the ACH matrix behind it. Handling caveats travel with the data, so sharing can never quietly widen a marking.
Deliberately deterministic
Every one of those steps is deterministic: given the same data, it gives the same answer, and it can always show its working. The document analysis runs on rules, not a model — nothing to download, no network call, and it behaves identically inside an air-gapped enclave. It is precision-biased: it would rather miss a weak candidate than clutter your graph with a false one.
That is what makes the output defensible. When someone asks “how do you know?”, the answer is on the screen: this evidence, graded this way, correlated by this rule, at this time.