Both the current Libre and the Dexcom G6 can be very accurate, but there's little consistency between sensors, or even time of day – could be the sensor, could be the insertion point, could be a whole host of other factors. My wife's G6 has often proved remarkably accurate compared to a finger-prick test, while at other times it's been way, way off – but its trend prediction (fast drop/rise) is usually enough.
I'm guessing this one isn't being billed as a medical device yet because this is a good way to get it thoroughly tested first and, maybe, tap into a new market with money to burn. And hopefully lessons learned here will roll through to approved medical versions, whose official software, at least, leaves a lot to be desired.