Observe · Remember · Reason · Verify · Converge
The Bayesian World Model
Every belief here is a probability with an explicit uncertainty and a first-class source. Nothing is asserted without evidence — claim, evidence, confidence, and source travel together.
The model at a glance
Each domain score is a precision-weighted combination of its grounded beliefs — confident beliefs count more. The shaded band is the 1σ credible interval; the bar tip is the mean.
Every belief, with its receipt
One row per grounded belief: the raw measurement (evidence), the normalized flourishing posterior (claim), its uncertainty (confidence), and a link to the exact feed it came from (source).
What to ground next
The loop ranks beliefs by uncertainty × leverage — how much grounding one belief deeper would tighten its whole domain. These are the next questions worth the model's effort.
Known gaps — not live-grounded
How a number gets here
Posteriors come from real yearly public data. A raw indicator is mapped to a 0–1 flourishing score by a chosen linear map between a "bad" floor and a "good" ceiling: posterior = clamp01( (raw − floor) / (ceil − floor) ) That mapping is a value-laden choice, not an objective truth — so each belief carries an uncertainty that already includes a penalty for the choice. Domain scores combine beliefs by inverse-variance (precision) weighting.
Primary source: World Bank Open Data — keyless yearly world aggregates.