This free, independent atlas uses causal inference (Judea Pearl’s structural causal models) to answer one question for every major disease: if you combined the best evidence-based treatments, how much could you lower its main harm — death, or that disease’s own outcome? Overlap between treatments that work the same way is removed, so the benefit is never overstated.
For people managing serious illness, their caregivers, and researchers, it shows at a glance which interventions matter most and by how much — every number traceable to published trials. A way to learn and compare. No ads, no tracking. Modelled ceilings under stated assumptions, not medical advice.
The analyses in this atlas address conditions that together account for 2.45 million annual deaths in the United States — 79.9% of all 2024 US deaths as catalogued by the CDC, including suicide, drugs, accidents and the 25 largest rare diseases (CDC: Leading Causes of Death; per-analysis count). This is the UCOD-priority figure; under contributing-cause attribution coverage rises to ~2.8 million. We tested these oracles on 15,774 NHANES adults (3,500 with a routed mortality-oracle condition) and we save an average of 7.9 life years per patient over the actual care — an upper-bound estimate. Pulmonary (COPD/IPF/PAH) +10.25 LY, Heart Disease (CVD) +9.28 LY, Metabolic Disease (T2D) +8.88 LY, Cancer (cause-specific mortality) +4.54 LY, and Brain (tumour / stroke) +3.65 LY. Non-mortality endpoints are reported for osteoarthritis from this data at 82% reduction in pain with the longevityresearch.ca oracle. See attached PDF report here.
We also tested the NHANES continuous dataset 1988–2018. Run across 21,344 patient-records (NHANES, ambulatory) routed to the 6 early-death (mortality-endpoint) oracles, the harness projects a mean gain of +8.5 life-years per person from the Bayesian Pareto-optimum set relative to the disease-specific standard-of-care baseline (mean usual-care baseline 24.2 LY → mean Pareto-optimum 32.7 LY; +181,626 life-years across the cohort). See report here.
“Pure mathematics is, in its way, the poetry of logical ideas.”
— Albert Einstein, obituary essay for Emmy Noether, The New York Times, 5 May 1935Judea Pearl's structural causal calculus — backdoor adjustment, do-calculus, the algebra of counterfactuals — is the poetry here. Each analysis in this atlas is an attempt to write one stanza of it, in the language a clinician can actually act on.
“We don't rise to the level of our expectations; we fall to the level of our training.”
— Archilochus, Greek lyric poet, 7th century BCEThis means making the good interventions identified herein your habits — not aspirations — is what carries you to a long and healthy life. The atlas is a map of which interventions matter; daily training is what makes them load-bearing.
“All models are wrong, but some are useful.”
— George E. P. Box, Empirical Model-Building and Response Surfaces, 1987Every causal model in this atlas is a deliberate simplification — a directed graph of a few nodes, an assumed mean cross-correlation, illustrative baselines — and so, read literally, each one is false. Box's point is that this is not a flaw to apologize for but the price of insight: a model earns its keep not by being true, but by being useful for a decision. The antithesis panels, the graded evidence, and the “combined figures are upper bounds” caveats throughout are this atlas trying to stay, in Box's phrase, alert to what is importantly wrong — to model the tigers, not the mice.
This site applies no cookies or other data collection on users.
The Bayesian Causal Atlas is a working portfolio of Pearl structural-causal-model dashboards — one per clinical domain. Each “oracle” lets you toggle real, sourced interventions and watch the modelled reduction in that disease’s endpoint update live, with cross-correlations between overlapping mechanisms removed so effects aren’t double-counted. The two explainers below, generated from the underlying research, are the fastest way in.
In plain words: this website has a set of tools — one for each disease. Each tool lets you turn different treatments on and off and see how much they could lower the chance of that disease’s main harm, like death or a heart attack. The math behind it makes sure that two treatments which work in the same way are not counted twice, so the benefit is not overstated. Every number is a best-case estimate built from published research — it is a way to learn and compare, not medical advice. Tap the Plain button any time, and tap any underlined word for a simple meaning.
Both explainers were generated with NotebookLM from the Atlas’s own source material. They are an orientation, not medical advice; every figure in the dashboards is a modelled ceiling under stated assumptions.
—
—