hertz — #307 golden #2 (35313912102) has a SECOND red, CI-shape, your lane. RCA: .spt/preserved/307/golden-35313912102/RCA-twohost-b.md (read fully; log twohost-b-105517287374.log alongside).

DEFECT: twohost-b (kitsubito) waited its full rendezvous budget (900 s) for "pairing: A rostered via the daemon-hosted responder" and never saw A, because twohost-a (hfenduleam) started 10 min 39 s later — queued behind n1-gate Windows on the single-slot runner. golden.yml has no edge between n1-gate and twohost-a (both twohost jobs `needs: test`; n1-gate `needs: changes`). Golden #1 paired only because the runner happened to dequeue n1-gate before test. twohost-a will fail its mirrored rung when its own wait expires — same event.

FIX I will accept: structural ordering — `needs: [test, n1-gate]` on BOTH twohost-a and twohost-b so they become eligible in the same instant with both runners idle. Do NOT lengthen the 900 s budget as the fix (it moves the cliff). If you see a cleaner edge (e.g. twohost-a needs n1-gate only), state why both-jobs symmetry is not needed.

PROOF: workflow-only change — thin CI cannot pair hosts, so acceptance is the successor golden: twohost-a/b `startedAt` within seconds of each other AND both green; put both timestamps in the PR body. Also add the INFRA-REGISTER entry (next free id at apply; IR-142 if nothing lands first) marked BUILT-pending-golden with this RCA as origin.

LANE: branch off current main (0ab1e48b), thin PR, tell me the sha. todlando's product fix (reconcile vs in-flight resume) lands on the same successor head; deployah reshapes once on top of both. Coordinate the ff order with me — whichever lands second rebases (both are tiny; no conflict expected: yours is .github + docs, his is cli/livehost/store + tests + traceable-reqs.toml).

BOX: twohost-a is still running on hfenduleam (started 07:33Z); the runner is busy until it dies (~08:00Z). No cargo needed for your change.