Streaming clients and the self-hosted Replicator hold a local copy of entitlement state, so flag checks run in memory with no call to Schematic in the request path. If that connection drops during a deploy or a network interruption, the client resumes where it left off and replays only the updates it missed, instead of tearing down and re-syncing the entire dataset.

Local state stays current, enforcement keeps running without a gap, and reconnects stop triggering a cold re-sync. This matters most for high-throughput deployments that run on local state and cannot absorb a resync stall on every reconnect.