An engine is a self-contained piece of mathematics - named, tested, and deterministic. Give it the same inputs and it returns the same figures, every run. Every number on a Nexus screen was computed by an engine - its name printed on the receipt.
The model you talk to is good at two things: understanding your question and narrating the answer. It is never allowed near the arithmetic. When Nexus needs a number, it calls an engine - physics, statistics, economics, or engineering, built for your vertical and tested before it ships. That separation is the architecture, not a policy or a prompt.
Engines are organised by the kind of mathematics they run. Each family is led by a named engine in production use, with further engines behind it. Open any family to see what it covers.
CASCADE traces one change through every dependency: what's affected, in what order, when it lands. Built on the doctoral cascade research behind the company.
DEFERMENT prices the cost of waiting. Production or service deferred against intervention cost, day by day: what waiting costs, what acting costs, and when the two lines cross.
LIFESPAN runs material-based deterioration curves: how an asset ages, its remaining life, and what repairing now versus year-N buys.
HORIZON runs climate and weather scenarios against your assets on real ground - flood, heat, and wind: what's exposed, when, and at what cost. Built on the doctoral research validated against real hurricane impacts.
CODEX checks a submission clause by clause against the written regulation - every pass and fail cited to its source, and a refusal when the data cannot support a verdict.
SEVERITY grades every defect against industry reference frameworks, profiles severity along an asset, and ranks a whole network for capital planning.
SENTINEL runs detection and monitoring analytics - incidents, violations, anomalies - and refuses honestly when the data cannot support an answer.
TRIAGE answers what to do first: attention lists ranked by severity and degradation, and work orders with disruption flags and working windows.
A family we haven't named yet, built for your vertical - the spec written as a test suite, in service in days, governed from the first figure.
Every family computes on real ground. Geography is not a family here because it is in all of them - it is where Nexus lives.
The cascade research behind CASCADE and HORIZON is peer-reviewed: Gupta & Gevaers (2026), GeoAI, Ghent. doi:10.5281/zenodo.20306240
Every industry page names the engines for that vertical. Find yours → · See where the engines sit →
Engines are built for your vertical, and each one enters service the same way. The spec is written as a test suite: every number the engine must produce is pinned by an automated test before it ships. New engines are born quarantined - marked draft and unverified - and promoted into production by a person, never silently. And when a question needs analysis no engine covers yet, Nexus can draft one on the fly - restricted to whitelisted operations, marked draft and unverified on every figure it returns, and admitted to the working set only when a person promotes it. A tested engine appears in days, not quarters; our most recent proof of concept brought more than a dozen into service inside a single month.
Open any figure and the receipt shows the engine that computed it, the inputs it read, and the assumptions it made. Same inputs, same engine version - identical figures on the rerun. That is what makes an answer checkable rather than plausible.