'Customary' uses of predictions of dangerousness are widespread and well-accepted in the criminal justice system. Customary uses of prediction include judicial sentencing, police arrest decisions, and prosecutorial decisions about case dispositions. 'Exceptional' uses of prediction are those involving state action which could not be supported absent the prediction. Such uses included preventive and pretrial detention, 'career' criminal investigations, and selective incapacitation. The problem with exceptional uses of prediction is the absence of guidelines. Customary uses of prediction have pre-established limitations. Critics of statistical predictions of dangerousness argue they are not sufficiently accurate to be the basis for depriving persons of their liberty and that they violate the presumption of innocence and notions of individual autonomy. Customary predictions of dangerousness are inevitable in the criminal justice system, and they should be made as accurate as possible (use statistical instead of intuitive methods) and limited in their impact, given the error factor. Punishment involving a prediction should not extend beyond that permissible independent of the prediction. When offenders with similar backgrounds and a similar current offense receive significantly different sentences based on a prediction of dangerousness, the base expectancy rate for violence of the 'dangerous' offender must be reliably substantially higher than the offender given the less severe sentence. 28 notes.