It's sometimes said of murder mysteries that success is equal to the sum of the probabilities the reader intuitively assigns to the top suspects.

If the butler seems 20% likely to have done it, the cook 30% likely, and the maid 15% likely, and no other character is a plausible candidate for the crime, then your book has a score is a lousy 65%. But if the butler seems 60% likely, the cook 85%, and the maid 30%, then you knocked it out of the park with a score of 175%. You've made your readers hopelessly credulous, which is exactly your job as a storyteller.

I think the next step I should take toward reasoning with quantitative precision is to stop caring whether my probabilities sum to one.

There's this very familiar mental dance I go though when making predictions. First, I think, "what is the probability?" Then I think of alternative hypotheses, and assign probabilities to those. I notice the sum of my probabilities approaching one (or going past it), get uneasy, and check how many alternative hypotheses are left. When I find the answer is "a lot", I move a little toward revising the probabilities I've already stated, feel a combination of embarrassment and dissonance, and give up.

Or, occasionally, I go all the way through the dance, and end up with ad hoc probability assignments that make sense together but have nothing to do with the model I use to predict things in real life.

What I need is *two separate procedures*: one for gaining explicit knowledge of my implicit model, and another improving my model.

Gaining explicit knowledge of my implicit model should usually come first. The backtracking dance results from feeling that my automatic implicit models should already make sense. Which is absurd; I'm currently running on an ape.

Right now, it seems like there are two groups: 1) people who are good at this stuff and trot out their lovely, consistent numbers with confidence, and 2) people who are scared shitless of saying numbers out loud, because numbers lack plausible deniability, so if you say a dumb thing with numbers people will laugh at you.

I'd really like to see people publicly assigning greater than 100% on their first pass at probability assignments. Then they can walk us through what they think their brain is doing and why, and we can learn about the mistakes people are making and how to fix them.

"Right, so, I think a huge chunk of that 85% for the cook came from the scene where he seduces the pool boy. The pool boy scene was *super salient*, and I... might have gotten a little carried away. On further reflection, seducing the pool boy doesn't mean much for the case one way or the other. My new estimate is about 60% for the butler."

If I don't have a lot of training in quantitative probabilistic reasoning, and I end up giving about 100% probability to the top candidates in a *really good* murder mystery when asked to assign numbers, I probably need to ditch the pretentious formality and quantify the crazy mess I really feel.