Given the ADE ICs used in the Shellys I doubt it’s a sampling rate issue - those things typically sample at 8000 samples/sec.
I think that’s likely to be the cause of your discrepancy. AFAIK there is no spec on how quickly an energy meter should respond to a power reversal. From a revenue point of view I suspect energy providers would say “the faster the better”. I think the diverter folk refer to this as their “energy bucket” - basically how much slop the meter gives them before it starts counting in its Import or Export register. It’s possible your Shelly has a much smaller bucket than your revenue meter. When your Shelly reports that you’ve imported energy during diversion, does it also show an identical export of energy?
There some more detail here on what’s going on inside any meter (Shelly or revenue) trying to deal with this. That detail refers to the ADE7816 which is a very close cousin to the ADE ICs used inside the Shelly. Scroll down a few posts to see a scope trace of how quickly the IC detects a power reversal.