Do Our Confidence Numbers Mean Anything?
When we say 70%, does it happen 70% of the time? This page grades every signal’s confidence against what actually resolved — updated nightly, wins and misses alike. No cherry-picking: every fitted signal source is shown, including the ones that started out badly miscalibrated.
See the live performance scoreboardThese are the same signals the paid Discord posts.
Join the free Discord to see it fire live, or go Pro for every emission with a Polymarket deep-link. $15/mo, cancel anytime.
Signal Confidence, Audited
Every signal source is graded against its own predictions. “Raw gap” is the worst mismatch between predicted and actual win rate before fitting. “Calibrated gap” is the same measurement after our nightly Platt/isotonic fit. When the gap shrinks, the confidence number actually means what it says.
Each signal source’s raw predicted win rate is compared against the actual resolved win rate within confidence buckets (the reliability diagram). A nightly Platt or isotonic fit is applied per source, and we report the maximum bucket gap before and after fitting — the “proof of fit.” Markets resolve retrospectively, so buckets fill in over time as pending signals close out; nothing is filtered or excluded once a signal source has enough samples to fit.
Resolution labels are live; a multi-source on-chain verification audit is in progress.