Everything the IDL section renders, in one read.
The three halves fail independently and are reported independently: the chain read, the release read, and the legacy account. A GitHub outage must not hide what is on chain, which is the fact an operator is most likely here for.
One RPC, one network — the network is the service's, not the caller's.
A deployment is the console for exactly one network: the route 404s on the other one and every action re-checks it, so a map keyed by network could only ever hold a single entry and the "not configured for that network" branch would describe a state two upstream guards already make unreachable. Taking the network as construction state instead makes a caller asking this service about the wrong chain unrepresentable rather than merely refused.