I just had this idea. It rocks.
Let's decide a token set based on significant daily life data, such as wgets from key webpages (newspaper, weather report, relevant blogs, incoming e-mail, whatever, even real horoscopes).
In the evening, one judges the day and files the morning tokens under the right training set.
Over time, one gets a reliable morning prediction on what's going to happen :)
This is definitely going to be covered in my LinuxDay talk on Nov, 26. The talk title is "Advanced tools for cazzeggio" where "cazzeggio" roughly means "creative ways of wasting time".
Serious note: of course this wouldn't work, because reading the prediction in the morning would alter your perception of the day, creating feedback loops that would interfere the results.
It could still be useful to be read not in the morning, but only at the end of the day: that would be an interesting tool to tell you if your life is becoming too predictable. :)