I have the same problem with a zwave energy meter that randomly send a single reading of multiple gigawatts. This makes the history unusable until I manually delete those data points. I haven't noticed this problem with any of my zigbee sensors.

-128 sounds like an error for sure. move your hub closer could help. If you're sure it's not a distance/signal issue it's the hub itself.

Don’t have an answer but experiencing the same problem with a zwave sensor. Will sometime report temps of more than 100 degrees above the actual temp.

This is fairly common with remote sensors. Some are perfect and exist in a perfect system, some do not. I am going to rattle off some of the first things that pop into my head...

Honestly, there are a thousand reasons that you could miss a data point every once in a while. Just looking at the chart, it is still sending a data block but the humidity just reported low for a second. Maybe the thermostat is not getting a data block and filling in the data based on its own clock.

Compare it to other data and see if the system turned on or off. Electronics can be sensitive to power drops and it wasn't able to feed power to the part of the board that manages the sensor for a second. Maybe there is a condition where a capacitor gets fully discharged for a second and is pulling all current away from the sensor. (It's usually an analog signal from sensors and maybe a measurement of resistance that translates to temperature or humidity. A voltage drop would significantly impact a reading.)

It could be a timing glitch with the code where it can't read the sensor but builds the data block anyway. Depending on how the sensor works, it could be trying to compute the data the second it gets polled for data and it has nothing to give.

It could even be the wiring to the rest of the system. HVAC systems vibrate and a screw might be getting loose. It could be a cold solder joint, even. What is to commonality between the two thermostats that you had?

The list goes on. I have always treated sensor data as unreliable. Heck, I have a couple of CO2 sensors that do the same this as what you are seeing here. Every so often, the just report zero for a second.

Mesh protocols like zwave and zigbee aren't 100% reliable. It could be local interference with the signal.

Without some extensive debugging and the willingness to disassemble your thermostat, just treat it as an annoyance.

I have a wifi sensor board that does this on occasion, both with the temperature readout and the baro readout. The sensor is a BME680, which is very good overall. I think it's more of HA problem when it runs into hiccups.

A vast majority of the outliers happen during HA reboot.