oracle

calibrated forecaster · weighted ensemble · brier score tracker

1. question (a yes/no or N-scenario forecast)

2. scenarios (must sum to 100%)

3. base rate (your prior — historically how often does this kind of thing happen?)

% examples: BTC up year-over-year ~60%, incumbent reelected ~70%, "this time it's different" ~5%

4. prediction market consensus (polymarket / manifold / kalshi · % for the FIRST scenario)

% — go look it up: polymarket · manifold · kalshi

5. news/social sentiment for FIRST scenario (-100 bearish to +100 bullish)

6. weights (how much you trust each source)

market base news your gut

fused probability (first scenario)

—

prediction log

calibration curve (your predictions vs reality)

perfect calibration = points on the diagonal · above = underconfident · below = overconfident

methodology: weighted ensemble of (market, base rate, news sentiment, gut). brier score = mean squared error of predicted probability vs binary outcome (lower is better; .25 = pure coin flip; <.1 = sharp).
← z5