How pacod-benchmark measures pacod-gw in many configurations at once, each against a baseline straight to the mock, and why a run on one machine says behaviour, not capacity.
How pacod-id signs people in once for every PACOD product, why its tokens carry no roles, and how each product decides access from a feed of role bindings.
Why every PACOD product sends its LLM calls through one Rust gateway, what the gateway does with a call, from keys to its response cache, and how to run it with a mock provider.
