Predictive Routing queue depth skew with BYOC trunk failover in APAC

Looking for advice on predictive routing behavior when primary BYOC trunks failover in the Singapore region. Queue depth metrics spike artificially for 15-20 seconds post-failover, causing the predictive engine to over-allocate agents before stabilizing. Is this a known latency artifact in the analytics pipeline?

trunk_config:
 region: ap-southeast-1
 failover_mode: immediate
 predictive_engine: v2.1
 metric_aggregation_interval: 5s