Why prior vs posterior matters
The prior is the population baseline — what the model expects before any case-specific information arrives. The posterior is the personalised prediction after conditioning on what is actually known. The shift between them measures the information value of the evidence.
A node that barely updates despite strong evidence signals a missing pathway in the graph. A node that updates sharply even when not directly observed reveals indirect propagation through the network. For decision-making under uncertainty, the posterior is what matters — but the prior provides essential context for judging whether an update is plausible or the evidence is genuinely surprising.
Gray bar = prior mean (population baseline). Teal bar = posterior mean (updated for this evidence). Values shown on the model's internal latent scale.