Every prediction, scored in public.
We publish falsifiable claims with a confidence level and a target date before the outcome is known. After the target date, each claim is resolved against measured signal data and scored — including the misses. Outcomes are never edited after the fact.
CALIBRATION · CONFIDENCE VS REALITY
No resolved predictions yet — the calibration curve appears once claims start reaching their target dates and being scored. Until then there is nothing to plot, and we won't invent it.
FORECAST RECEIPTS · WHAT WE SAID VS WHAT HAPPENED
Every night we store 30-day forecasts for each trend. Once a forecasted date passes, the prediction is compared against the composite signal that actually landed — and the error is shown here, unedited. Receipts appear as soon as the first forecasts have aged enough to be checked.
METHODOLOGY
Claims are issued with a confidence (0–100%) and a target date. An automated evaluator flags each claim once its target date passes; resolution is recorded with notes and, where possible, a link to the evidence.
Each resolved claim contributes (confidence − outcome)² to the Brier score, where outcome is 1 if correct and 0 if not. Lower is better: 0 is clairvoyance, 0.25 is what coin-flipping with 50% confidence scores.