Originally Posted by
Notanatp
Interestingly, none of the reporting has said just how likely the test scenario was to occur in flight, other than to say it was "esoteric," or "theoretical," or "extremely improbable."
The test simulated the effect of 5 bits being simultaneously flipped in the FCC's memory due to cosmic rays or some other unnamed cause. The 5 bits were independent. One bit was flipped to tell the FCC that MCAS was active when it wasn't (this disabled the yoke cut-off switches). Another bit told the FCC to incorrectly issue a nose-down trim command. The other 3 bits weren't described but apparently were necessary to create the runaway trim scenario.
Assuming a 5-bit event, the odds of these 5 specific bits being flipped depends on the size of the memory. If the FCC has 1-megabyte of RAM, then the odds of those particular 5 bits being flipped in a 5-bit event should be on the order 10**-32. I don't know how frequently the FAA estimates a 5-bit event will occur on an aircraft with a 1-meg memory, but even if it approaches 1 (i.e., it can be expected to happen each flight), the likelihood of this particular 5-bit event occurring on any single flight is on the order 10**-32. You could fly all the 737 Max's that will ever be made 3 times a day for a billion years, and you'd still be looking at probabilities on the order of 10**-16.
As for the rest of the reporting, I cannot find it now but I've read one account of the test that said all three test pilots recovered the airplane if it was assumed they recognized the problem within 3 seconds (the time to respond to an autopilot runaway pitch trim). The FAA wasn't satisfied with that so it ran additional tests were it allowed the failure to go longer before being recognized, and one of the three pilots either couldn't recover or didn't recovery fast enough. My understanding is that it was on the basis of this addition test where it was assumed that only an exceptional pilot would have recovered, therefore, the catastrophic classification. Has anyone else seen this account, and if it is correct, then doesn't it sound like this is rigged against Boeing?
There's a significant assumption here that bit flip events are what are referred to by statisticians as IID (independent and identically distributed), but if they are bits within a single word of data in memory or a single register within the processor they could be physically extremely close together. The 286 is based on 1.5um technology so those bits may be only tens of um apart. For memory devices they could be closer. I don't know enough about cosmic ray events to know whether a single event could affect multiple bits within a small area like that.