The need to update our estimates of probabilities (e.g., the accuracy of a test) given new information is commonplace. Ideally, a new instance (e.g., a correct report) would just be added to the tally, but we are often uncertain whether a new instance has occurred. We present an experiment where participants receive conflicting reports from two early-warning cancer tests, where one has higher historical accuracy (HA). We present a model showing that while uncertain which test is correct, estimates of the accuracy of both tests should be reduced. However, among our participants, we find two dominant approaches: (1) participants increase the more HA test, reducing the other; (2) participants make no change to either. Based on mixed methods we argue that both approaches represent two sides of a ‘binary’ decision i.e., (1) update as if we have complete certainty which test is correct and (2) update as if we have no information.