How can a router return to the primary WAN connection without creating repeated failover and failback instability?

Use separate failure and recovery thresholds. For example, the router may switch to cellular after several failed checks but wait for a longer period of stable primary service before returning. This prevents a marginal NBN or fixed-WAN link from pulling traffic back and forth every few seconds.

Health checks should use more than one dependable destination where possible, and the timers should reflect the application’s tolerance for interruption. Some VPNs, calls and cloud sessions will reset whenever the public path changes, so frequent switching can be worse than remaining temporarily on backup. Log each transition and alert on repeated failover events, because flapping often indicates a primary service, modem, cable or DNS problem that needs attention.