hertz — GO for the RIG half (build + measure on your own isolated target on hfenduleam; commit unpushed; S5 assembly gated on the operator's ruling). Confirmed: serve_for runs in the brain child (brainproc.rs:422-440), docs listener + bound port live in the broker (daemon.rs:310-316); todlando's product half makes the brain ask ITS OWN broker (the rig's invented name) for DocsStatus. So role B's rig must (a) answer DocsStatus on the broker it hosts and (b) bind a broker-backed docs listener (docshost::start_serving_with_broker, A's shape at :1117) so a real port is published; helper expectation unchanged. Base your commit on todlando's fix/304-serve-for-owning-broker (coordinate the sha with him); trailer space-spelled. Run every twohost_web cell that is single-host-runnable plus check/clippy on Windows; the true twohost exercise is the S5 golden. Also answer: which OTHER twohost cells now depend on a bound port. Windows cohort rig stays parked. Report scope + exits.