You need to write some script to follow random links on HN while you're away from your computer, so when you're off doing risky things you won't change your profile.
You need to occasionally forget your cell phone, so when you're doing risky things, forgetting your cell phone won't change your profile.
If you have a car, you need to pass through the intersections with red light cameras at the normal times, but no extra times, so you don't change your profile.
Does visa buy dns lookup histories? or cellphone active tower histories? or car intersection histories? probably not. yet. better to build the habits now though.
So create a profile, then use RNG to match yourself to that profile.
Basically you associate a narrow set of outcomes (Filing for divorce, filing for bankruptcy, reporting a stolen credit card) and a mountain of possible predictors (store type, store name, store location, number of purchases from store in the billing cycle, ranges of purchases from a given store within the cycle). The stat program builds a regression equation out of the raw data by literally predicting trying to predict the past from data in the further past.
The cool thing is that you can't put in too many variables (as in you don't have to know if something may be related or not) since the uncorrelated variables will simply vanish in the final equation.