P3204 — GH#598 _TZE284_9ern5sfh "Unable to connect" (2026-11-02, Paris)
User report presumed true (docs/rules/USER_REPORTS_PRESUMED_TRUE.md). 10 leads:
- Payload: replayed
homey app build+npm run prepare-publishat tag v9.0.1604 —_TZE284_9ern5sfhsurvives in lcdtemphumidsensor (with TS0601). Not a compaction loss. - One owner: lcdtemphumidsensor only (compose + app.json). OK.
- pid: interview says TS0601, listed. OK.
- Endpoints/clusters: MISMATCH. Node ep1 in = 0, 4, 5, 60672, 61184; out = 10, 25. lcdtemphumidsensor declares bindings 1, 1026, 1029. The Homey SDK docs state one bind request is made per
bindingsentry during pairing; none of the three exist on the node. Before P2823 the couple lived on climate_sensor (nobindings) and paired. - onNodeInit crash: none found in either driver (Tuya listeners wrapped).
- Mandatory 0xEF00 / capabilities at pairing: none in code; the only pairing-time step is the manifest binds of lead 4.
- Test 9.0.1604 contains the couple (lead 1).
- Prefix: Z2M lists
_TZE284_only; interview string is exactly_TZE284_9ern5sfh(case OK). - Sleepy end device (receiveWhenIdle false): failed/unanswered binds extend the interview while the device sleeps — consistent with "Unable to connect", and with a Bridge radio path.
- Pair template: driver uses default Zigbee pairing; learnmode text fine.
Fix: move the couple (_TZE284/_TZE204/_TZE200_9ern5sfh, canonical case only) to lcdtemphumidsensor_3, whose bindings 4/5/61184 all exist on the node and whose DP map is Z2M's (DP1 temp/10, DP2 humidity, DP4 battery %, request-driven time sync). Pins, sacred-keep, curated removal, removal evidence, fp shard follow. Already-paired units elsewhere keep their driver.
Open risk (not changed): other EF00-only TS0601 couples in lcdtemphumidsensor (e.g. _TZE200_vvmbj46n) get the same 1/1026/1029 binds; check with an interview before moving any.