No single future is knowable — so the Oracle simulates 2,000 of them (a Monte Carlo
path ensemble on the coin's own moves) and prices the whole distribution: a median path target,
50% / 90% ranges, a probability ladder, and a transparent direction ensemble — then
records the forecast daily and scores itself 24h later. BTC · ETH · SOL · HYPE.
Calibrated probability, not a crystal ball.
📊 Probability ladder — the chance price finishes ABOVE each level, next 24h
simulating…
🧭 Direction ensemble — why the lean
loading…
🎯 Calibration & reliability
▸ How it works — and why it's honest. This is the credible version of a "quantum"
predictor: since THE future is unknowable, we simulate 2,000 possible futures — block-bootstrapping the
coin's own fat-tailed hourly moves, rescaled by an ensemble volatility forecast (close-to-close EWMA +
the Parkinson and Garman-Klass range estimators, blended with Deribit's DVOL — the options market's
own implied-vol index — at a walk-forward-validated weight for BTC/ETH) with hour-of-day seasonality
— and read probabilities straight off the ensemble. The corridor is conformally calibrated
against its own realised hit-rate so coverage tracks the 80% target, and the direction lean comes from a
transparent ensemble of weak signals (funding crowding, trend/regime, mean-reversion, RSI, MACD, live order
flow) whose votes are shown above — bounded [12–88]% because daily direction is ≈ a coinflip and pretending
otherwise would be selling you noise. Everything is recorded daily and scored: pinball loss on the
interval, Brier on direction (vs the always-50% baseline of 0.25), coverage with a Wilson CI, and sharpness —
a wide band that always "hits" is worthless; tight-at-target is the skill.
Prediction history — every daily forecast, logged with its date and resolved against the realised price