What is known, how it is known, and when.
Reality Graph is a temporal and geospatial operational model connecting real-world entities, events, relationships and decisions while preserving their evidentiary basis.
- OBSERVE
- UNDERSTAND
- SIMULATE
- RECOMMEND
- AUTHORIZE OR REFUSE
- DISTRIBUTE
- ENFORCE
- EXECUTE
- EVIDENCE
The model
Reality Graph holds the operational world as connected objects rather than as tables of records. A vessel, a facility, a report, an ownership link, a mission and the decision taken about them are all first-class objects, each with a location in space, a position in time and a record of the assertions that support it.
- EntitiesPeople, organizations, vessels, vehicles, facilities and systems, with identity history.
- EventsObserved and reported occurrences with time, place, source and uncertainty.
- AssetsOperational and infrastructure assets, their state and their dependencies.
- LocationsPoints, tracks, areas and routes in a common geospatial reference.
- RisksAssessed exposures linked to the entities, events and assets they concern.
- MissionsThe scope that bounds access, analysis, communication and action.
- Temporal stateThe graph as it stood at any past moment, not only as it stands now.
- Geospatial relationshipsProximity, containment, movement and exposure across space.
- Object and relationship historyEvery change retained, attributed and ordered.
- Source-backed assertionsEach property and link carries the assertion and the source behind it.
- Contradiction representationDisagreement between sources held as an explicit, inspectable object.
- Decision contextThe graph state that was in front of a decision-maker, linked to the decision.
Statements, typed
Reality Graph records not only what is believed but what kind of statement each belief is. The type is assigned when a record is created and does not change afterwards: an inference never becomes an observation by being repeated, and generated text reports what was reported, inferred, assessed or predicted rather than what simply is.
| Type | Question it answers | Written by AI |
|---|---|---|
| Observation | What did a sensor, system or person directly detect? | No |
| Source claim | What does a source state, without having observed it? | No |
| Machine extraction | What does this content say, as parsed? | Yes |
| Inference | What follows from these records, by this method? | Yes |
| Assessment | What does an accountable person judge this to mean? | Drafts only |
| Prediction | What may happen, under these assumptions? | Yes |
Fact is not a type. A statement can be treated as established for a purpose when it meets a configured corroboration policy — and that treatment is itself an assessment, displayed with its basis and open to revision.
Reconstruction
Because history and decision context are part of the model, the graph can answer the question that matters after an event: what did the people who acted know at the moment they acted? Atlas uses the same temporal state to replay an area of interest; the Evidence Ledger links each decision to the graph state it was taken against.
Access to the graph is governed object by object. A person or AI agent sees only what their mission, label ceiling and purpose allow — and data they are not entitled to is not sent to their session at all.