Three commitments, in order. First, causal structure before estimation: the assumptions about what causes what are drawn and stated before any number is fit. Second, Bayesian estimation reported in full — priors, posterior intervals, and the sensitivity of the result to its assumptions. Third, evaluative reporting in likelihood ratios: how strongly the evidence favours one explanation over another, not a bare point estimate.
The output is a conclusion that states its assumptions, quantifies its uncertainty, and is built to survive adversarial examination. See how we model uncertainty and the public record of analyses published in full.