When the outcome of an action will be determined by a graduated-success oracle (i.e. "no, and," through "yes, but" to "yes, and"), things can get swingy and unpredictable. Using a likelihood mechanic to bias the results, particularly in a way that honors the logic of previous outcomes, reduces the chaos while still producing organic deviations in narrative.

In Mad Hermit, an assigned likelihood drives the selection of one die from a roll of three six-sided dice. The single die produces a range of results as shown in the table above. By selecting the likelihood of each question consistent with the outcomes of previous questions (e.g., if the previous question resulted in a hero hanging from the edge of a cliff, then the likelihood of the hero leaping back into action and immediately slaying her opponent should be "very unlikely"), the story flows more naturally than if it were just a single die being rolled each time.