the trouble with the physical examination is that there simply isn't a physical test that allows us to diagnose (aside from the behavioural measures that we have already). and with respect to 'but future science will enable us to do this' i really don't think that this will be the case. the current diagnostic categories are conceptually confused and do not map onto real distinctions in nature. the distinction between 'mentally ill' and 'not mentally ill' is similarly confused and doesn't map onto distinct categories in nature. with respect to predicting behaviour... we really don't do so very well and a considerable amount of the predictive leverage we have already is probably due to confirmation bias / self fulfilling prophecy.

the best predictor of future violence... is past violence...

but that being said often people do change.

the trouble is that we simply don't know whether someone is likely to be a risk or not. we guess... but not so good at it. rather like... trying to predict the weather.