Open Data Infrastructure
Data Modeling Event-Centric Context for Agents
How event-centric data models help agents reason over state changes, business processes, policy events, and time-bound operational context.
Agents struggle with business context when the data model only shows what is true now and hides how it became true.
Events explain change
State is useful. Events explain why the state changed. For agents, that difference matters because many business questions depend on sequence, cause, policy state, and timing.
An event-centric model records what happened, when it happened, who or what caused it, which entity changed, which policy applied, and which source produced the record. That gives agents a path through history instead of a single flattened fact.
Model the business process
Use event types for orders, approvals, denials, entitlement changes, model releases, incidents, and table promotions. Link events to entities and source systems. Store valid time and observed time separately when the distinction matters.
Core idea: event-centric context helps agents reason over transitions, not just current values.
Give agents time-aware context
A retrieval layer should expose event sequence, lineage, source authority, and policy state with the answer. If an agent recommends action based on a customer, claim, account, or dataset, it should know which events created the current state.
For related ODI context, read event-sourced agent workflows, temporal entity memory, and context graph change impact analysis.
What breaks first
- The model stores current status but not the event that changed it.
- Policy decisions are events, but they are not linked to business entities.
- The retrieval index flattens time and loses sequence.
- Late-arriving facts overwrite history without explaining observation time.
Event model questions
Ask which events define the business process, which entities they affect, which policies govern them, and which time field the agent should trust. If the model cannot answer those questions, the agent will invent continuity.
Sources to start with
These primary sources anchor the technical claims in this guide.
- W3C PROV-O recommendation
- OpenLineage object model documentation
- dbt semantic models documentation
- NIST AI Risk Management Framework
Agents need the story of the state, not only the state itself.