Finally your newly reformatted data files must be tested. You should ensure that the programs properly understand your model. You should check the worst items of bad geometry with the geometry command and calculate the R-value of your model with the rfactor command. If you have wildly bad bond lengths or bond angles either your model is in error or the stereochemistry has been incorrectly defined. In either case the error must be corrected before refinement begins. It is quite important to check the bad contacts too. Any overlooked geometry restraints will result in inappropriate bad contacts being identified. Often forgotten disulfide bonds will be found in this fashion.
The check by rfactor involves some subtle checks as well as the obvious calculation of the R-value itself. Of course, the R-value must be reasonable. If your model was produced via MIR model building your R-value should be less than about 45%. A model produced by MR could have an R-value in the upper 50%'s and still be good. At this point it is too early to condemn such a model. A Molecular Modification model should be below 25% R-value unless large changes occurred during the modification.
In addition to the R-value some other values should be checked. The completeness of the observed data set should be consistent with your expectations based on the progress of data collection. If the completeness is too low or the completeness of the ``matches'' is not the same as that of the observed data (which it must be because the calculated structure factors are 100% complete) it is likely that you have the wrong asymmetric unit in reciprocal space. This error is corrected with the correct command.
If you have reserved a test set, and your model is truely unrefined, the starting working R and free R's should be roughtly equal. Some fluctuations are to be expected because of the small size of the test set, but if your test set contains at least 1000 reflections the free R should be no more than 2the working R. Of course, if your model was refined to an isomorphous crystal's diffraction pattern and your R's are equal you have a problem. This situation indicates that your test set is not a true test set -- in such cases you must select as you test set exactly the same reflections as the original refinement.