We have experienced similar reliability issues multiple times over the past month. Examples include: * Vapi voice model outage: a fallback model was configured, but it did not activate, leaving customers in silence. * Deepgram transcriber outage: a fallback transcriber was configured, but it was not invoked, so the agent could not hear the customer. * Full-scope outage today: the health-check alert email was only received after the Vapi server had already recovered. fallback numbers are configured but not invoked. The main concern is that the fallback and alerting mechanisms are not behaving as expected during real outages.

i will be sharing all of the feedback from the community with the team once this incident is fully resolved. I definitely understand the frustration