It sounds like the simulators all this testing were on didn't actually simulate MCAS, so the "fault injection" was done manually. Testing in the way you describe wouldn't provide meaningful results without an MCAS-capable simulator.
Hmmm. Then unfortunately we'll have to wait for that to be implemented I suppose. If the algorithm is as it has been described, it may work, but at this point is moot. The damage has been done.
