All of them optimize for those well defined regulatory test cases. It's all shades of gray as long as they are just tuning parameters to yield favorable results

All of them optimize for those well defined regulatory test cases. It's all shades of gray as long as they are just tuning parameters to yield favorable results in the test scenarios without bothering with possible drawbacks in other situations. But things suddenly turn solid black when a method of optimization is employed that is explicitly disallowed in the rules. It's not unlikely that this infraction was enabled by "motivated disbelief" questioning that software alone can be a "defeat device". And without court-actionable "defeat device" this thing would clearly be in the gray range, even if the "up to forty times" that are being talked about would probably make it one of the darker shades. They might have even checked this with some lawyers, who can sometimes be very eager to look at the law in the way most favorable to the client: "Yes, it may hold in court" (and if not I will still get my money, also, the negative PR caused by the lawsuit even if you wirll will be your problem, not mine)