there is no "slightly out of range". Its either in the range or outside. Valid or invalid, when it comes to critical tests like these.
if temperature deviation is outside of acceptable deviation range then thats a system fault and result should have been considered invalid.
Back in the days there was much less regulatory oversight and products like that could slip through the cracks, Resulting to deaths, missed diagnosis etc etc.
The same with Teslas AP: its either 100% confident in the situation or not. If the car is not confident in whats happening - it should shut down. If that happens too often then the AP feature should be removed / completely disabled.
How many more people have to get into accidents? I know, if Musk's mom (knock on wood) was a victim of this feature then things would be taken more seriously.
You do realize that I gave a much simplified view of the situation as this is a web forum discussing a related subject, not an actual design review of the instrument, right?
To any process I can set multiple "acceptable ranges" depending on what I want to accomplish. There can be a "reject" range, "ok, but warning" range, "perfect, no problem" range, or a "machine must be broken" range
Everything is context dependent; nothing is absolute.
I like that term. When I was involved in the medical instrumentation field, we had a similar concern: it was possible for an instrument to produce a measurement, e.g., the concentration of HIV in blood serum, that was completely incorrect, but plausible since it's within the expected clinical range. This is the worst-case scenario: a result that is wrong, but looks OK. This could lead to the patient receiving the wrong treatment or no treatment at all.
As much as possible, we had to be able to detect when we produced an incorrect measurement.
This meant that all the steps during the analyte's processing were monitored. If one of those steps went out of its expected range we could use that knowledge to inform the final result. So the clinician would get a test report that basically says, "here's the HIV level, but the incubation temperature went slightly out of range when I was processing it, so use caution interpreting it."
Like most software, the "happy path" was easy. The bulk of the work we did on those instruments was oriented towards the goal of detecting when something went wrong and either recovering in a clinically safe manner or refusing to continue.
With all the research into safety-critical systems over decades, it's hard to see how Tesla could not be aware of the standard practices.