A context graph is described as a persistent layer linking telemetry with organizational knowledge, asset ownership, tickets, prior investigations, so an AI system can interpret alerts without querying raw logs from scratch every time. If that's how it works, could a mature context graph reduce how much raw historical data you actually need retained in an expensive SIEM tier, since the graph already holds the relevant relationships rather than requiring a fresh query across raw logs? Or does the graph just sit on top of the SIEM without changing the underlying retention need at all.