BETA — invite-only. Every call below is real, immutable, and scored on calibration — read the exact rules.
/calibration & the brier score

Calibration & the Brier score

This is where skill gets earned rather than claimed. A call isn't scored on whether it hit — it's scored on how well the stated confidence matched reality.

// outcome = 1 if the call came true, 0 if it missed
brier_score = (confidence/100 outcome
outcome = true outcome = false
0.000.250.500.751.00025%50%75%100%90% & wrong = 0.8155% & wrong = 0.3090% & right = 0.01brierstated confidence

Because the penalty is squared, being 90% sure and wrong (0.81) costs almost 3× being 55% sure and wrong (0.30). Overconfidence is punished quadratically harder than honest uncertainty — the curve is the whole argument.

CONFIDENCEOUTCOMESCORETAKEAWAY
90%true0.01Confident and correct — the ideal.
55%true0.2025Hedged but correct — modest credit.
55%false0.3025Hedged and wrong — a small penalty.
90%false0.81Overconfident and wrong — punished quadratically.

It's a proper scoring rule — the mathematically optimal strategy is to state your honest belief, no more and no less. That property only holds because confidence does exactly one job here: it is scored, and it is never used to compute a payout.