Thread devices speak CoAP — small request/response messages over UDP, designed for low power. When packets drop, you feel it as delayed lock status, scenes that take ten seconds, or energy reports that skip intervals. Apps rarely say “CoAP drop”; they say No Response or show stale state.

Diagnosis means correlating drops with mesh parents, interference, and border router load — not running Wireshark on every phone.

Symptoms mapped to drops

Actions succeed on the second try but fail on the first. Automations appear to run “late.”

Battery sensors die faster while reporting Updating often — sleepy devices retransmit dropped CoAP.

Drops worsen when microwave ovens run, when Zigbee meshes heal, or when many Thread routers reboot together.

Only devices two or more hops from a border router show lag; nearby plugs stay instant.

Home-level diagnostics without a lab

Use OEM apps that show Thread parent and last reachable time (Eve, Nanoleaf, some Aqara). Parent flapping correlates with CoAP loss.

Test the same action from vendor app vs. Apple Home vs. Google Home. Drops only in one controller suggest fabric or cloud path issues, not RF.

Power-cycle one border router and time how long until far devices respond — long gaps imply mesh reconvergence dropping packets.

Temporarily reduce mesh hops by adding a mains-powered Thread device midway.

Mitigations that reduce drops

Improve 2.4 GHz cleanliness: move Zigbee channel, relocate border routers, add wired backhaul.

Limit simultaneous bulk operations — updating twenty bulbs while running heavy Zigbee scenes stresses airtime.

Update firmware on border routers and lagging accessories together.

Remove ghost Matter devices that generate background polling.

When drops need vendor or network help

OTBR logs in Home Assistant showing sustained CoAP timeouts on a flat, wired network deserve support tickets with timestamps.

Corporate Wi-Fi with aggressive multicast filtering breaks more than CoAP — whole local control may fail.

Confirm drops decreased

Pick a far Thread lock or sensor. Trigger it ten times; measure subjective delay. Successful diagnosis shows consistent sub-two-second responses and stable parents over twenty-four hours without rising battery drain on sleepy nodes.

Official sources