Definition
Knowledge graph maintenance is the work of keeping entities, relationships and facts true as the organization changes. Temporal drift is what happens when the graph is treated as a snapshot.
The problem
Contracts are amended. Teams move. Revenue is restated. A graph built last quarter is already wrong in the places that matter. If agents query a drifted graph, they act on yesterday’s organization with today’s permissions.
What exists today
- Periodic rebuilds, which leave a window where the graph is silently false.
- Human curators, who cannot keep pace with operational change.
- Append-only extraction, which records new facts without retiring the old ones.
Deca10 research question
What verification loop — generate, check, update — keeps an organizational graph inside a known error bound as source systems change?
Evidence
Maintenance metrics are being defined around drift detection latency, stale-edge rate, and the cost of a full versus incremental update. Results will be published here when they are stable.