L40 Ultra constant 'offline' = firmware TCP/SACK bug, not WiFi

Device: Dreame L40 Ultra — model dreame.vacuum.r2492b — firmware 4.3.9_1752 — Dreamehome — US region (us-west-1)

  — Please read before the standard troubleshooting —
  This is NOT a WiFi/home-network problem, so reboot-router / move-closer / hotspot / factory-reset do not apply. I've ruled the network out with measurements:
  - WiFi signal −34 dBm; association verified solid for 5+ min, zero disconnects
  - ~80 ms RTT to your cloud; handshakes complete <90 ms
  - Gateway: no blocks, no IPS, DNS/NTP fine

  The fault is at the TCP layer, inside the robot's own cloud daemon. I captured it at the gateway.

  — What happens —
  The robot holds a persistent TCP connection to your cloud (us-west-1, :19973). Every ~60–120 s it enters a retransmit/SACK deadlock and silently abandons the connection (no FIN/RST) after ~100 s, then reconnects. Each cycle = an "offline" event in the app.
  Mechanism: the robot keeps retransmitting a segment the server has already SACK-acknowledged instead of the one actually missing; never recovers; ~100 s timeout; reconnect. Looks like a SACK-recovery bug — firmware-fixable.

  — Scope — Matches the most common connectivity complaint across the Ultra line on 4.3.9 firmware; users reproduce it next to the router → shared firmware cause, not WiFi.

  — Full write-up + capture — https://github.com/Tasshack/dreame-vacuum/issues/1639

  — Ask — Please forward to firmware/networking engineering, not standard support. I can provide the full pcap on request.