On the cost of inference at the boundary of a regulated network.

Inference at the boundary of a regulated network is a different problem from inference inside one. The boundary is where capability becomes regulated. Until the system crosses it, the system is a possibility. After, it is an artefact.

The cost is not the cost of the call. The cost is the cost of the apparatus that turns the call into a record. A record must be signed, retained, retrievable, and refutable. A call that cannot become a record is not yet useful.¹

We have learned to design the apparatus before we design the call. The model is selected last; the trace format is selected first. This is the method, laid out elsewhere.

The studio operates on the position that determinism is a service contract. A non-deterministic system at the boundary of a regulated network is not an engineering object. It is a hazard.²

What follows is not a recommendation. It is a description of what we have done and what we have observed. The reader may file it under whatever heading is convenient.