Accuracy scorecard
How often we're right.
Every prediction is falsifiable. We score ourselves against what actually happened. Recomputed weekly, with the methodology public.
Calibrating · public scorecard launches once we have enough data
calibrating
We won't publish accuracy until we have enough data to publish honestly.
Depth of sampleDozens of settled predictions
Spread across timeSeveral distinct weeks
We hold publication until there are enough settled predictions, across enough separate weeks, to rule out a lucky run. Until then, this page shows no number at all.
How we score
A prediction is only real if it can be wrong.
Every reasoning chain states a falsifying observation up front. When the window closes, we check what actually happened and mark the prediction right or wrong. The tally on this page is the unfiltered result, recomputed weekly. We publish the number or we publish nothing.