> It is illegal discrimination to make hiring decisions based on IQ scores. You can only use a test that shows those with higher scores perform better at the job.
There is more scientific evidence that IQ predicts job performance than for most of the more qualitative criterions that employers use in practice. The source of the illegality of using IQ tests is not legislative (intelligence is not a protected class[1]). It comes from the precedent of Griggs v. Duke Power Co., in which the IQ test was being used as a barrier for jobs where IQ was not a good predictor (not programming), and prior to thorough research on the subject. There have been hundreds of studies showing that IQ tests predict job performance (though not super strongly) at the beginning of careers, even though this fades once they have more experience. This[3] meta-analysis approaches these studies very critically, and concludes that many inflate the effect, and that it doesn't validate the IQ test (i.e. there may be non-cognitive reasons for it) but that the correlation still exists. (For what it's worth, I suspect they're right about the non-cognitive part since the correlation has become much stronger over time, particularly around the 1970s. However, when your goal is to test for something and not to explain it, correlation is good enough.)
> Basically you can not ask questions that test skills that would not be used in the job the candidate is applying for. ... You must be able to defend the metric you use for hiring and that it is not discriminatory.
No, the burden set in Griggs and later elaborated in a modification to the Civil Rights Act is not to prove that a test is not discriminatory at all. Instead for tests that are discriminatory in practice (though not obviously in design, like IQ), you must show that the test is truly indicative of job performance. If this were not the case, you couldn't ask if people went to college, due to the racial disparity in college graduates. Also you can ask things you can't prove are related to job performance if it isn't discriminatory (not that this is really useful).
> If you give an IQ test that is unrelated to the job and say no one below 100 IQ will be hired you must prove that someone with 101 IQ can do the job but someone with 99 IQ can not properly do the job
Of course you don't have to show that anyone who fails your test will 100% be worse on a case by case basis than someone who passes. If that were the case you couldn't justify pretty much any test other than "applicant is not dead or in a coma." You just need to show that it is clearly less likely. I would find it really surprising if there wasn't some kind of IQ threshold effect for programming (i.e. the closer you are to the middle of the spectrum, the better a predictor of programming performance IQ is). That would make it easy to justify using an IQ test as a predictor, especially in combination with the more general studies.
I don't think IQ would be terribly useful to weed out programmers because I suspect there are more related ways to judge applicants that would already weed out those in the region where IQ is useful, but it is not as simple as "IQ testing for a job is illegal."
[1]: https://en.wikipedia.org/wiki/Protected_class
[2]: https://en.wikipedia.org/wiki/Griggs_v._Duke_Power_Co.
[3]: http://www.tandfonline.com/doi/full/10.1080/10888691.2014.98...