Re: Surely this is a form of fraud?
Morally, probably. Legally I suspect they've carefully read the test documents (or I'd hope someone had with wriggling of such magnitude). I can well imagine that the emissions standards documents merely state something along the lines of "follow this driving profile with a standard car and measure the emissions", so if a driver of such a car were to mimic that profile then the volume of emissions should be equal to the test result (allowing for rolling roads, etc...).
Unless of course there is something in the code along the lines of:
if(NOX_output_enabled_for_probe)
call dump_wee_in_exhaust()
else
call disable_wee_in_exhaust()
Reminds me a lot of the phone vendors doing benchmark test "optimisation" routines that use the bigger on-board processor for running benchmarks, but the little one for more mundane tasks like playing games/browsing/making calls...