I see researchers simultaneously defend IQ tests as the most predictive tool in psychology and also acknowledge they are imperfect measures of intelligence. Those two claims seem in tension to me. If the tests are imperfect, how much should we trust the predictions they make? And what exactly are the limitations people are pointing to when they say IQ tests do not fully capture intelligence?
Imperfect in measurement theory just means there is some noise between the observed score and the true underlying ability. IQ tests have that, but they have less of it than almost any other tool in social science. The predictive validity survives because the signal is strong enough to overcome the noise. Legitimate limitations like test anxiety or format familiarity introduce random error at the individual level but do not invalidate the construct or the group level predictions.
Imperfect and invalid are not the same thing. A ruler with slight warping still measures length. The question is whether the error is systematic and biasing or just random noise that averages out. For well designed IQ tests the evidence points mostly to the latter.