Track record
Are our forecasts any good? These are short-range, checkable forecasts, each with a deadline and a resolution rule, scored in public when they resolve. The A–D probabilities can't resolve, so this is how you can check our judgment.
27 forecasts: 17 open, 2 withdrawn, 8 resolved; next deadline Nov 20.
Open forecasts
The amber bar is our probability; a hollow ring marks prediction-market odds where a comparable market exists.
Calibration
The chart appears once 15 forecasts have resolved.
Resolved
Withdrawn forecasts
Withdrawn forecasts aren't scored. Each shows why it was withdrawn.
How scoring works
Each resolved forecast scores (probability − outcome)², where the outcome is 1 or 0. The Brier score is the average: 0 is perfect, and always guessing 50% scores 0.25. A raw Brier score depends on how predictable the questions were, so we also show a skill score against always forecasting the base rate of the resolved questions (above 0 beats it), and, where a prediction market priced the same question, the market's Brier score on those same questions. Every score shows how many forecasts it rests on; a handful proves little.
When we say 70%, it should happen about 70% of the time. The calibration chart waits for 15 resolved forecasts, because a group of one or two would read 0% or 100% by chance; it then groups them into three bins (five once 40 have resolved), each with its count and a 90% interval, and points near the diagonal are well calibrated.
New forecasts are added in each weekly wrap-up, and nothing is edited after it's made; if a forecast's stated context turns out to be wrong we add a dated correction and still score the original probability. Ill-posed forecasts can be withdrawn unscored; the reason is always shown. How we round and rate evidence: Method.