Replication lag read inconsistency

You run an active-active multi-region datastore (region-east, region-west) with asynchronous cross-region replication. At 04:00 the inter-region network link degrades, not a clean cut, but packet loss pushes replication lag from <1s to several minutes and flapping. Dashboards: each region still serves local reads/writes fine, but users who write in one region and read from the other (e.g. mobile clients hitting different edges) see their own writes 'disappear,' and a few cross-region operations conflict. Recent context: a backbone provider reported degradation. How do you triage and mitigate?