FREE f :

From the set of reflections that satisfy all selection criteria, chose randomly a fraction f and use it as a test set (ie for the calculation of the free R value). Because the number of reflections used for the minimisation has to be relatively small (to make the program practical), the free R value will be calculated over a very small number of reflections. This has two consequences : (i) the standard uncertainty (due to statistical fluctuations) of the free R value is large, and, (ii) because the number of reflections is small, we can not independently scale F_os to F_cs for the test set (which may bias the R_free ?). The take-home message is that the free R values produced by QS should not be taken at face value.