The core issue is a priority conflict the transcriptionEndpointingPlan (start speaking plan) is only used if smartEndpointingPlan is not set and the transcriber doesn't have built-in endpointing capabilities. If the transcriber has its own EOT detection, it takes over automatically. (Vapi) So if Scribe v2's realtime VAD/endpointing counts as built-in transcriber-level detection, Vapi should defer to it but you'd want to confirm that Scribe v2 is actually recognised as such by Vapi's pipeline, otherwise there could be a conflict where both are active. Are you seeing the assistant cut off too early, respond before the user finishes, or something else and is this on web or phone calls?