You could make a table where row is guessed probability(input to Kelly criterion) and column it's actual probability and cells contain expected value and expected log value. And pick something which looks reasonable.