Obviously the best way to prevent recidivism among violent criminals would be to simply imprison them for life following their first offense (we'd call it the "one strike and you're out" law), but thanks to that antiquated document known as the Constitution, we're forced to coddle convicts by ensuring that they don't endure "cruel and unusual punishment" to pay for their crimes. Well if we can't lock up all the muggers, jaywalkers, or tax cheats and just throw away the key, the next best option would seem to be predicting criminal behavior before it happens -- and though it sounds like straight-up science fiction, that's exactly what University of Pennsylvania criminologist Richard Berk intends to do. Using a data set consisting of 30 to 40 variables derived from anonymous Philadelphia probation department cases over a two-year period, Berk and his colleagues were able to craft an algorithm that supposedly separates those folks likely to re-offend from ex-cons not deemed to be as dangerous. The point of this program is not to make an end run around sentence limitations or to arrest people before they've committed a crime, but rather, to help probation units decide how to best allocate their resources among hundreds of potential re-offenders. To wit, each subject's variables are plugged into the software in order to create a so-called "lethality score" -- and although the aggregated data is still relatively small, Berk points out that childhood exposure to violence has already floated to the top as the single most likely predictor of murder. Another, less well-known, but equally accurate predictor: jailhouse tattoo across an inmate's chest that reads "Die, ___, Die."