Joining happened once, during start up, and the control handle was dropped straight after. Nothing could rejoin because nothing still owned the handle that joining needs. That held up until the access point went away. The mesh access point this controller reaches the network through is switched off overnight, and the controller never came back: it sat transmitting to an access point that was not there for five days, while the fans, the button and the LEDs carried on working perfectly. Only the network was gone, and nothing on the device could say so, because saying so is what the network is for. So keep the handle and give it to a routine that watches the link and joins again when it drops. The MQTT session already reconnects when the broker goes away; this is the same idea one layer down, and without it that reconnect loop retries forever over a radio link that can never work again. Joining also waits between attempts now. Retrying as fast as the chip can answer is harmless for the one attempt at start up and a busy loop for the whole night once it can run whenever an access point is off. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>