Scored in public

The record

Accuracy alone can flatter a forecaster. So the record shows the lazy baseline, a proper scoring rule, and whether its confidence means anything.

—correct

—lazy baseline

—calls made

—judged

—"survives" calls that survived

—Brier score vs baseline (lower is better)

Calibration

An honest forecaster's 70% should happen about 70% of the time. Bars show what actually happened in each confidence band.

actually survivedwhat it said · bands: % sure it survives

The calls it was most sure about, and got wrong. They stay here.

Green pushes toward "survives", red toward "dies". Re-fit once a day on every judged call; the new weights enter the chain before they are used.