(duplicate of the comment above — posted twice due to a submission error)

Follow-up on your 2026-04-15 suggestions — both delivered in the third FB22476701 update (2026-04-21): filtered QUIC PCAP (GCS host only) capturing the deadlock window, plus pre/post system state snapshots (brctl queue, nettop, 30-min cloudd log, sample stacks showing missing-wakeup in mach_msg2_trap). Three repros within 50 minutes confirmed h3/GCS is back on this path. Thanks for the pointers.