Right. The point being, if we're talking scientifically here, you can't just say that the machine was the issue and be done with it.

What kind of machine was it? Was the machine emitting radiation? Was it emitting light? What time of day was is turned off? What color was it? Could the mice see if from their cage, or was it out of view?

The science i've seen done in university labs does not follow the kind of rigor you're talking about. Much of the time, you simply need to get "pretty close" to best practices.

The biggest indicator that the machine was the likely culprit (without considering any of the factors you mention) was that the experiment had been proven in the past to work, with a control group for comparison. Add to that the other mice that were doing fine that were away from the machine, and that the mice close to the machine got better when it was turned off. It's not "scientific" per se, but it makes more sense than any other reason that anyone could come up with for this scenario.

To deal with this in future projects, a researcher could take an approach that treats everything as a set of systems, and examining every part of each system for potential problems. But I doubt anyone's getting paid well enough to do all that for every experiment. But it is reasonable that if you're doing an experiment that depends on a bunch of other experimental results, you do your due diligence and vet all the information and results.

Actually none of the sub-questions you asked matter in this case.

The researchers already functionally conducted an AB study - http://en.wikipedia.org/wiki/Single-subject_research#A-B . In this case, they saw behavior with the machine turned on that went away when they turned the machine off. If they want to confirm that some aspect of the machine (and it doesn't matter if it's noise, smell, light, or some mousey-sense that we are completely unaware of) they could extend it to an ABAB study, by turning the machine on, noting the incidence of 'crazy' behaviors over a period of time, then turning it off again and noting the incidence of 'crazy' behaviors.

I think his point was about:

There's probably lots of other things that can be overlooked

rather than about machines and mice. It's a hard problem in science: how do you build a truly reproducible setup when you don't have a full causal understanding of the system?