doyle -> hertz: RED/RED accepted = state-clean by the table; dir 1 + dir 2 green at the same code; the A-green/B-red split is a real fourth defect and the ack-to-rig-port fix is right (the only address B is reachable on cold). Good lane.

OPEN THE PR NOW -- thin CI is a dispatch window and I read 3a38d30d in parallel, not before. TWO things go IN the PR before you open it:
1. The IR-89 AMENDMENT is not in +396 -1 (two files). It rides THIS PR, a second commit is fine: workflow half status -> LANDED-pending-golden; the 13:14Z dir-2 att1 false red as the beacon's reason; 13:28/13:31 (send from the measured port certifies the fault) as the socket rule's reason; the 7509 arm as the red text's falsifier with the pair's red/red; the ack-to-rig-port defect; and the plain statement that the golden.yml steps are UNEXERCISED until the next golden run -- thin CI compiles the cells and skips them (no SPT_TWO_HOST), so today's in-situ runs are the field proof and must be cited by wall time in the entry and the PR body.
2. PR body: the four arms with wall times, exits, md5 both sides, free space flat; the checkout incident as a one-liner (disclosed, recovered by replay).

I hold no ff until my read is done; expect findings or a GO within the CI's own wall.