Yes, as you say they can be controlled/messed with by someone that knows how. But that's not exactly dave who's just been asked by his Mrs if he's shagging mavis next door!
IIRC it's not even by someone who knows how they work.
There are simply too many variables between people and the actual equipment and test questions do nothing to zero out most of those, so someone might
feel guilty about what has happened (or even something completely unrelated) and that will throw the results off, or someone might feel have done something else and thus the results get thrown off, or the person may simply show more/less response than the baseline which throws the results off, and even then your results are being interpreted by a human who most likely has zero real, repeatable training in it.
It's one of the "junk" proofs that has littered US law enforcement over the years, often based simply on previous successful cases (where a conviction was gained*), and thus the person who got the initial successes is able to "train" up more people and they can refer to the original person's track record, then when their evidence is questioned they call on their trainer or someone else trained by him to back them up. The result is a self reinforcing feedback loop until people start to look at how many of the cases ended up being overturned and why, or Judges actually started to look into the basis of the "science" rather than take a prosecutor's experts word for it (with blood spatter that started to happen about the time it turned out large numbers of cases that rested extremely heavily on the evidence of one guy and his "training school" were being overturned with things like DNA proving it wasn't possible for the convicted person to have done it, unfortunately a number of times after the death sentence had been carried out).
*Please ignore those cases where the key evidence from the expert was found to be nonsense 20 years later.