In this case, however, there was one honest-to-God bug (in AFNI). The other software modeled something in a way that sounds fairly reasonable, but didn't quite line up with the way that nature (or, actually, the scanner manufacturers) behaved. Unlike the fish issue, I think you could legitimately try to do things correctly and still get bit by this sort of bug.
One way to read scientific papers is to deconstruct the apparatus in the experiment to see if it could actually produce the data in the paper, and if so, what would have to be true for it to do so. Good papers go through that in experimental design section, bad papers gloss over it or toss it out as a stipulation "We used the same instrument that everyone else uses to read brain activity the fMRI scanner." (bad) "We used a technique called fMRI scanning which measures the magnetic resonance of hemoglobin in the blood, specifically it measures the difference between oxygen rich and oxygen poor hemoglobin." (much better) Then as a reader you can see that any change in oxygen, whether it is brain activity or bacterial decomposition would show up in such a resonance scan.
http://prefrontal.org/files/posters/Bennett-Salmon-2009.pdf
Choice quotes:
"[the salmon] was not alive at the time of scanning"
"The salmon was asked to determine what emotion the individual in the photo must have been experiencing."
You fuck the formatting for mobile users.