# Predictive Graph Digital Twin
**Entity class:** Analytic model
## Definition
A **predictive graph digital twin** is a continuously updated probabilistic representation of an entity embedded in a temporal network of people, devices, accounts, vehicles, locations, organizations, transactions, communications identifiers, and events. New observations update uncertain identities, relationships, latent states, and conditional forecasts.
It is a twin of an investigative hypothesis, not a duplicate of a human being. It should preserve provenance, competing explanations, uncertainty, and legal purpose rather than collapsing association into a single threat score.
## Relationships
- **structure:** [[wiki/Temporal Heterogeneous Graph|Temporal Heterogeneous Graph]].
- **updates through:** [[wiki/Data Assimilation|Data Assimilation]] and [[wiki/State Estimation|State Estimation]].
- **relationship layer:** [[wiki/Identity and Relationship Analysis|Identity and Relationship Analysis]].
- **computational acceleration:** [[wiki/Reduced-Order Modeling|Reduced-Order Modeling]].
- **ethical boundary:** [[wiki/Association Is Not Guilt|Association Is Not Guilt]] and provenance-aware uncertainty.
## Sources / Provenance
- [[research/Oden Institute - Theater-to-Person Predictive Scale|Oden Institute: From Theater-Scale Fields to Person-Scale Receptors]]
- [ODNI — Data Mining Report CY2021–2023](https://www.odni.gov/files/documents/CLPO/CY2021-2023_Data_Mining_Report_FINAL.pdf)