IQ tests are notorious for working poorly above ~135, and I’d say they only really work well for −20 to +0 relative to the designers, with a somewhat wider range for teams.
How is the designers’ IQ relevant?
The problem with designing IQ tests for high values is that it makes calibration costly. To put it simply, if you want to figure out whether e.g. “getting 94 out of 100 questions right in your test” really corresponds to “only one person in a thousand can do it”, you need to calibrate your test on maybe ten thousand randomly selected people—which means you need to pay them (otherwise you are limited to the statistically non-random subset of volunteers) and some of them will refuse anyway. The costs grow exponentially with the top measured IQ, so at some point the authors decide that it’s not worth it.
This is unrelated to the myth spread by many “high IQ societies” that having a certain value of IQ allows you to create tests for given value of IQ but not higher. That is nonsense. A person with IQ 160 would face exactly the same budget problem as everyone else when constructing an IQ test for values 160 and higher. (Unless their “test” is just some nonscientific guesswork, in which case of course the budget is just the cost of pen and paper.)
Also, in order to test IQ in the upper range, you need items that are difficult enough to be informative in the upper range, e.g. items where people with 130 IQ would usually answer wrong but 150 IQ would usually answer right. But such items would be wasted on the vast majority of test-takers who would get them wrong due to being lower than 130ish IQ. So statistically you get very little value from improving their accuracy in this range.
How is the designers’ IQ relevant?
The problem with designing IQ tests for high values is that it makes calibration costly. To put it simply, if you want to figure out whether e.g. “getting 94 out of 100 questions right in your test” really corresponds to “only one person in a thousand can do it”, you need to calibrate your test on maybe ten thousand randomly selected people—which means you need to pay them (otherwise you are limited to the statistically non-random subset of volunteers) and some of them will refuse anyway. The costs grow exponentially with the top measured IQ, so at some point the authors decide that it’s not worth it.
This is unrelated to the myth spread by many “high IQ societies” that having a certain value of IQ allows you to create tests for given value of IQ but not higher. That is nonsense. A person with IQ 160 would face exactly the same budget problem as everyone else when constructing an IQ test for values 160 and higher. (Unless their “test” is just some nonscientific guesswork, in which case of course the budget is just the cost of pen and paper.)
Also, in order to test IQ in the upper range, you need items that are difficult enough to be informative in the upper range, e.g. items where people with 130 IQ would usually answer wrong but 150 IQ would usually answer right. But such items would be wasted on the vast majority of test-takers who would get them wrong due to being lower than 130ish IQ. So statistically you get very little value from improving their accuracy in this range.