The research behind SAIL

The formal proofs, the empirical audit on 13,192 real ICU stays (11,354 under the published Sepsis-3 table, used as a sensitivity cohort), and the honest account of what's still open — including the pre-registered statistical test, now run (Prediction A; see Status). Nothing here has moved; every link below is this project's existing page at its existing address.

Mechanisms

The two proven mechanisms — treatment-confounded severity construction and treatment overlap — explained in depth, with the intuition behind each.

Read →

Proof

The formal theorems and propositions, with full derivations you can check line by line.

Read →

Evidence

The empirical audit's numbers, experiment by experiment, including what's still open and why.

Read →

Guided walkthrough

A six-step, plain-language guide — try the live SOFA calculator yourself.

Try it →

Visualizer

One synthetic patient, three synchronized views — causal graph, state vector, and predictability gauge moving together in real time.

Try it →

Showcase

The whole finding in three acts, from the core question to the six-variant chart Variant F is the point of.

Read →

Status

What's confirmed, what's open, and the honest history of bugs found and fixed along the way.

Read →

Reproducibility

Every extraction query, the full analysis notebook, and the complete pipeline — public.

Read →

Arrived here first, but actually want the tool? Use the package →