Brier score calculator
Score your own binary forecasts — mean Brier score, reliability buckets and skill score, computed in your browser.
Line format: probability (0 to 1), then the realized outcome (1 happened, 0 did not). Spaces or commas both work. Rows that are unreadable are reported, never guessed.
Paste a sample and press “Score my forecasts”. Everything stays in your browser.
Synthetic example rows are clearly synthetic — they carry no forecaster's record. Tollbridge's own published record, its refusals included, lives on the track-record page.
The Brier score, in one formula
For each binary forecast, take the square of how far the probability landed from the outcome:
Brier (one forecast) = (p − outcome)², with outcome 1 when it happened and 0 when it did not
The mean over all eligible rows is the score people compare. 0 is perfect. Always predicting 50% scores 0.25 no matter what happens. An overconfident repeated miss reaches 2.
Skill score, and when it is undefined
The tool also computes a skill score against the sample's own outcome share — how much better your probabilities do than someone who always answers with the share of yes outcomes in your sample, which is the least any forecast should beat. When every realized outcome is identical there is no such reference, so the skill score is reported as not defined instead of a fake number.
Why sequencing matters
A forecast is gradable only when it was recorded before the outcome. Scores computed after the fact measure memory, not forecasting. Tollbridge therefore records the probability first, settles it against the market's own clock, and grades every decision — refusals included — on the public track record. The fundamentals are on the Brier score explainer, and the tool above never clips through a missing market: it scores exactly what you paste and reports the rest.
This page scores your sample only — it is not a claim about Tollbridge's published performance. That record, with its sample sizes and publication floors, is public.