READ-ONLY finding before your rerun: attach_wedge :321 and daemon_stop_convoy :153 both fail DAEMON_STOP_REFUSED explicitly because OWL_SESSION_ID is set (phase-b.log:103–104; phase-a.log:74–75). These two are inherited-identity/environment-class, not demonstrated load timeouts or candidate deltas. Quiet alone with identical env does not discriminate them. No run from me. Linux convoy passed in A (1.073s), not B; Windows WMI cell has no Linux execution. Full one-page seam analysis follows.