How the engine works — and how we grade it
Transparency over promises: this page explains the pipeline and publishes the live, ungroomed accuracy statistics straight from the forecast database.
1 · Twelve agents read the market
Trend, momentum, volume, derivatives, levels, news, narrative and more — each agent scores the same symbol from a different angle.
2 · One calibrated 0–100 signal
Agent reads are weighted by reliability and merged into a single score with scenario targets. It is a signal, not advice — the final decision is yours.
3 · Every forecast is recorded first
Before the outcome is known, the forecast (direction lean, probability, targets) is written to the database with a due date. No cherry-picking is possible.
4 · Graded against real prices
When the horizon ends, the record is scored with the actual market price: direction hit or miss, Brier score for probability quality.
5 · The engine learns per class & symbol
Calibration pools are kept separate for crypto, stocks and funds, so accuracy in one class never inflates another. Symbols earn their own calibration as history accumulates.
Live track record
Direction accuracy = share of graded forecasts whose direction lean matched the real move. Brier measures probability quality (lower is better; 0.25 = coin flip). History is young — treat small samples accordingly.
| Horizon | Graded forecasts | Direction accuracy | Brier |
|---|---|---|---|
| 1D | 360 | 48.3% | 0.433 |
| 1W | 30 | 60.0% | 0.272 |
Open (not yet graded) forecasts: 1260
Signals, not advice. Nothing on this page is investment advice; past accuracy does not guarantee future results. You make the final decision.