Leaderboard

Fair score (the default order)

Every settled call scores from the probability you locked at publication — nothing else. Calling an unlikely outcome correctly is worth more; being highly confident and wrong costs more. Maximum 20 points per prediction, and the reward and penalty always add up to 20, so no probability is a better strategy than another.

correct = 20 × (1 − probability) · incorrect = −20 × probability

Scores are computed by the system from the locked probability once a call is objectively resolved. Members, moderators and the AI cannot adjust them.

Track record

Resolved accuracy, discounted for a small sample. Five correct calls out of five ranks below fifty correct out of fifty, because one settled call proves far less than fifty.

score = accuracy% × (1 − 1 / (1 + resolved calls))

Example: 80% over 4 resolved → 80 × 0.80 = 64.0 · 80% over 40 resolved → 80 × 0.976 = 78.0

The metrics behind it

• Accuracy — correct ÷ resolved (Correct + Incorrect). Predictions still pending or expired never count for or against you.
• Resolved — how many calls have actually been settled, by the AI outcome scanner or a reviewer. This is what earns the sample-size weighting above.
• Predictions — total calls recorded in the selected window. Volume alone never lifts a rank.
• Recent accuracy & trend — accuracy on calls settled in the last 30 days, and the gap against everything settled before that.
• Followers — shown for context and used only when you sort by Followers.
• Avg conviction — the average 1–10 conviction on your calls. Displayed, never scored, so hedging low doesn't game the ranking.
• Rising contributor — 3+ contributions (predictions, posts or comments) in the last 7 days.

Windows and sorting

All time, Last 7 days and Last 30 days each recompute every metric from the calls made inside that window. The sort buttons reorder the same ranked set — Accuracy ignores the sample-size discount, Volume orders by total calls, Followers by audience, and Rising surfaces this week's most active members first. Ties fall back to prediction count, then followers.