Lightweight calibration tracker for probabilistic forecasting. Log a prediction with a confidence level, resolve it when the outcome lands, and see whether your stated probabilities actually match reality.
Calibration is a trainable skill: if you say "70%" on a hundred things, about seventy should come true. brierly keeps you honest about that — without becoming the kind of overbearing tool it's meant to avoid. It's a mobile-first PWA: React + IndexedDB, no backend, your data never leaves your device.
- Brier score — mean squared error of your probabilities.
0is perfect,0.25is what you'd get always guessing 50% (the bar to beat),1is worst. - Log loss — punishes confident-and-wrong far more harshly than Brier; one cocky miss costs a lot.
- Bias — average of stated minus observed: positive means you're overconfident, negative underconfident.
- Reliability diagram — your probabilities binned and plotted against the 45° line of perfect calibration; points below the line at high confidence reveal overconfidence.
Open predictions past their resolve-by date are surfaced as due, so logging doesn't quietly turn into a graveyard of unresolved guesses.
npm install
npm run dev # local dev server
npm test # scoring tests (the math is the one thing that must be right)
npm run build # production build (PWA: installable, works offline)Open the production build over HTTPS on a phone and use Add to Home Screen to install it like an app — it runs full-screen and offline.
Everything lives in your browser's IndexedDB; nothing is sent anywhere. Because that's per-browser, the Export JSON button is the real backup — export periodically, and use Import JSON to restore or move between devices.
MIT.