- log(.01) - log(1 - .01) = -4.6
- log(.99) - log(1 - .99) = 4.6
- log(.2) - log(1 - .2) = -1.4
- log(.9) - log(1 - .9) = 2.2
The formulas for probabilities were likewise converted -
log(a / (a + b)) - log(1 - (a / (a + b))) = log(a) - log(b)
Note how much simpler the formula on the right is. I think it's best to think of this heuristic as based on scores, with bayesian analysis being an intuitive motivation, rather than a rigorous basis.
ifile is a noteworthy related project.