Function request_batch 

pub fn request_batch(
    payload: &Value,
    served_model: &str,
    accept_any_model: bool,
) -> Result<RequestBatch, WireError>

Validate a System One request and convert it to one shared state plus the questions asked about it.

accept_any_model relaxes the model check from “must name the served model” to “must be a non-empty string”: third-party TypeSafe clients (the @ai-sdk/typesafe-ai SDK, the djev-run demos) hard-code a model id like jev-latest, and a deployment that wants to serve them can say so explicitly. The response still reports the served model, so the bridge never claims to be something it is not.