The article linked explicitly cites a study that disproves that assumption. People cannot tell liars apart better than chance even when watching videos of murderers' interrogations.
Scenarios in the article may be cherry picked and videos are not the same as being in the same situation in real life. Also in article they didn't have a chance to provoke subject into verbal contradictions as they were only looking at visual cues.
Even in the article they mentioned 85% success rate after some training.
I don't quite understand what point you're making. The article explicitly says that "spotting" a liar, that is detecting lies based on non-verbal clues does not work. However, other interrogation techniques, such as getting the subject to talk more freely to give them chance to make contradictive statements does work. That's what the 85% rate is about, don't use nonverbal cues, use verbal techniques.
I bet even if the stakes were "if I find out the number, you go to prison for one year" he wouldn't be able to.
My point was not that detectives are useless, interrogation does work wonders -- in some cases making the criminal confess, in others making the innocent confess to something he did not do.
My point was getting convicted based on some lie detection bullshit or just some confession alone is evil.
But if you knew you did something that could land you in prison and you were supposed to hide it, that would be a completely different scenario.