Here’s where we are:
Loads on the grid side of the blue phase were measuring incorrectly while the inverter was producing the blue phase voltage reference. It was both out-of-phase and lower than the independent grid side blue phase. Using derived reference for the grid side loads has resolved that problem and those loads are now measuring correctly.
The Blue phase VT is on the output side of the inverter. When the inverter is producing the power, it is a steady presumably clean 230V and the IoTaWatt is successfully monitoring power on the offgrid side.
When the inverter switches from supplying inverter power to providing grid power, the IoTaWatt is intermittently unable to accurately sense the voltage cycles and ceases to record power and voltage. These lapses can be seconds, minutes or hours in duration.
Attempts to obtain samples of the failing voltage signal have been unsuccessful because of the catch-22 that the IoTaWatt will not measure a signal it cannot sense.
My suspicion is that the inverter passes the grid through using solid-state switching that is altering the zero-crossing timing. I would need to see the actual failed voltage signals to be able to understand the sampling problem and determine if it is just a little outside of the parameters of the sampling routine, which possibly could be relaxed, or if there are multiple ambiguous zero-crossings that are defeating the fundamental way that the sampler works.
Another perhaps more remote possibility is that the multiple mechanical switches in the path of the blue phase are arcing or otherwise corrupting the clean grid sine wave. Limited plots of power anecdotally suggest that the problem could be triggered by larger current draw (the water heater?). Unlikely as this may be, a simple test would be to change the transfer switch while the problem is occurring to see if it goes away. Loads on the non-blue off-grid side will be measured with the wrong reference, but we would be interested only in seeing if the blue voltage reference freezes up.
I could figure this out given the ability to examine the equipment and modify the IoTaWatt configuration to extract more diagnostic information. There are unadvertised resources in the firmware that could be of help here. But being around the world and 15 hours out of phase makes doing anything like that very difficult and risky.