When I said short-term testing, I assumed one takes, say, 100 units from random positions in a manufactured lot, and stress-tests them for 100 hours.

There's no point in beating this dead horse. The fact is, nothing's perfect. Testing and maintenance of statistical data will reveal any necessary future improvements, and also will reveal hazards in manufacturing, not to mention possible design errors.

The only thing that MUST be ABSOLUTELY perfect is the code. It can be simulated, with proper tools, and one could expect you to create them if you can't buy them, such that events can be automatically or manually inserted in a simulation. I'd guess that, if it takes 2 weeks to code and debug a project, at least 10 weeks of thorough rigorous simulation in the pre-manufacturing phase would be indicated. Even that won't guarantee that the code is perfect, but it will certainly be closer than if it's simulated for 10 minutes.

Many coders believe that, once their code "runs" their job is done, just as many hardware guys think that, once the autorouter is finished, they can send the gerbers to manufacturing. That's the stage where the hard part begins, and not the end.

RE