That's not exactly what's going on here. The point is to use the sensory memories to improve the interviewee's ability to recall, and then see who actually recalled better the second time around. Yes, you could probably compare the two for inconsistencies as well, but that's not what he was measuring for this—just the change in level of detail between the two versions.
That's right, and another difference is that word frequencies change. E.g., in a lie there might be many references to "the green car", while a truthful story would refer to it variously as "the old green car", "the yellowish hatchback", "that ugly old car", etc. Yes there are inconsistencies in those descriptors. That is a property of memory, not of guilt.
Which shows yet another deficiency in LEOs' standard methods. They interrogate until they get an inconsistency, which they know the courts will interpret as guilt, even though this research shows that it is no such thing. Therefore intelligent guilty people can talk to the police with no fear, while no innocent person should ever talk to them.
