How does Centralpoint handle inference failures?

Failed inference calls are retried with exponential backoff against the configured provider, then optionally failed over to an alternate provider if policy permits. All failures are logged with model, error code, prompt hash, and user context so administrators can diagnose patterns. The platf