I’ve probably reviewed dozens of logs where this is happening, and I can reasonably safely say that the problem is almost certainly not in the binding. Spikes like this are normally caused by either the sensor, or from data being corrupted “in flight”. ZWave has quite a poor error detection system which is easily fooled in a noise environment but without seeing logs to understand the issue further, this is the likely cause.
In this case you could use a rule in order to disregard values which are outside the “normal” range. Use a second item that is not linked to a sensor, trigger the rule on each change of the linked item, verify if the changed value is in the normal range, if yes post update this value to the unlinked item. Use the unlinked item for persistence and on the sitemaps.
If that seems to much effort you could manually delete the false values from the used database.
I’ll try this. I was considering noise, but a at least 2 of my four sensors are in the same location as when I had this problem before and it was “solved”. I’ll post again after I’ve had some time to see how things work.
But maybe something else in the network has changed - the noise doesn’t need to be introduced at the sensor itself, but could occur in the mesh when routing through other devices. There are many reasons for this sort of thing and it’s unfortunately quite complex.