Howdy all, I’m working with the /call/{callId} update endpoint and have been successfully PATCHing metadata onto an active call (e.g., campaign, userId, address). What I’m trying to clarify is whether it’s expected that metadata injected mid-call can be referenced by the assistant as variables in the system prompt / conversational context, or if metadata is only read once at call initialization. Specifically: If metadata is added or updated after the call has already started, should the assistant be able to access and reason over that data? Or is metadata only ingested at call creation / before the first assistant turn, making mid-call updates usable only for logging, webhooks, transfers, etc.? In my testing, metadata PATCHed after the call begins appears to be stored on the call object, but the assistant does not reference or appear aware of it in conversation, which makes me think the system prompt is not recompiled mid-call. I’d appreciate confirmation of the intended behavior here, and whether there is a recommended pattern for dynamically injecting data that the assistant should actively use during an ongoing call. Thanks in advance — happy to provide example requests or call IDs if helpful. Best, Riley