Put the assumptions on the table

What Bayesian reasoning adds—and what it cannot supply.

Calling a claim “unlikely” does not avoid making an assumption. It expresses one without specifying its size. Bayesian reasoning asks us to distinguish where we start from what the evidence changes.

The core idea is comparative. An observation supports one explanation over another when it is more expected under the first. Showing that a favored explanation can accommodate an observation is not enough if its rival can accommodate it just as well.

Writing down a prior and an evidence weight makes disagreement more precise. Two readers might agree about the documents but disagree about the starting plausibility of concealment. Or they might share a starting point and disagree about whether two witnesses supply independent evidence.

AI assistance makes the arithmetic and bookkeeping easier. It can calculate alternative outcomes, maintain a ledger, and flag repeated evidence. That creates more room for substantive scrutiny, but it does not manufacture reliable inputs.

A beautifully calculated result can still rest on a poor reference class, a missing alternative, or an exaggerated likelihood ratio. Sensitivity analysis matters because it shows which judgments control the conclusion.

The Centaur Papers use this approach to make arguments inspectable. The invitation to the reader is not to defer to the number. It is to identify the assumption that should change, explain why, and see what follows.