Sadly, the added nuance here is that we actually have multiple multi-model bundles for different LGBM models based on the business context (loan type), and we wanted to move the decision for which model to call when, as well as the data transformation work needed to support it, to a compute plane outside of SF core. We have two different orgs that need access to the same endpoint as well. That's really where the Lambda came in as the brokerage layer.