Two fleet commander loops through one shared sidecar host.
A retained development fixture assigned each fleet a separate commander identity, admitted both through one two-slot supervised child, recorded accepted typed intent proposals for both sides, and required clean worker and process teardown before match artifacts were published.
The responder was deterministic fixture code, not a decision model. The run proves application wiring, authenticated transport, shared residency mechanics, identity separation, and lifecycle ordering. A named GGUF/model run is still required before any claim about model behavior or paired AI fleets.
See the deployment acceptance path ↗