Ontology
Not five thousand tables. A model of reality.
The ontology is the heart of the platform: understandable objects with typed properties, links that carry validity periods, evidence attached to every value, and security expressed on the object itself.
01
Objects
A country-scale ontology starts with the entities institutions already reason about.
Object types
Each object carries typed properties, for Company: identifier, name, industry, registration, employees, revenue, location, status, risk score, and every one of those values is evidence-backed.
02
Links
Relationships are first-class, typed and directional. This is where the graph of reality forms.
Person | +-- WORKS_AT ------> Company +-- OWNS ----------> Company +-- DIRECTOR_OF ---> Company Company | +-- OWNS ----------> Property +-- BORROWS_FROM --> Bank +-- IMPORTS_FROM --> Supplier +-- CONTRACTS_WITH -> GovernmentAgency
03
Time is first class
Asking who owns a company today is easy. Asking who owned it on a given date is the question that matters in an investigation or an audit.
Person A | | OWNER_OF | start: 2021-04-13 | end: 2025-11-02 v Company X Query: who owned Company X on 2024-02-12? -> the link is evaluated at that instant, not at today's state.
04
Evidence on every value
A generated claim must never become truth by repetition. Values carry their origin and their confidence.
VALUE
|
+-- source
+-- timestamp
+-- confidence
+-- transformation
+-- author
+-- classification
+-- evidence
Company.revenue = <value>
why?
-> ministry filing
-> document <id>, page <n>
-> extracted <date>
-> confidence <pct>
Company.risk_score = <value>
why?
-> model Risk-v<n>
-> supplier concentration +<n>
-> debt ratio +<n>
-> late payments +<n>
-> external exposure +<n>05
Security lives in the object
Classification, purpose limitation and masking are properties of the object, not a separate application feature that a new client can forget to implement.
OBJECT
+-- data
+-- relationships
+-- provenance
+-- classification
+-- policy
Person.phone
classification: PII
purpose:
fraud-investigation
customer-service
allowed:
fraud team
authorized support
AI access:
masked unless purpose approved06
Interoperable domains
One gigantic model collapses. Separate domain ontologies that link through CORE do not.
CORE
Person, Organization, Location, Event, Document
Shared spine every other domain links into. Nothing is duplicated across domains.
ECONOMY
Company, Industry, Transaction, Price, Import, Export, Employment
Trade, production, prices and the firms behind them.
HEALTH
Hospital, Capacity, Supply, MedicalEvent
Capacity and supply, at facility resolution, without individual patient records in the shared layer.
TRANSPORT
Road, Vehicle, Route, TrafficEvent, Infrastructure
Network state and the events that degrade it.
ENERGY
Plant, Grid, Generator, Consumption, Outage
Generation, distribution, load and failure.