Model accuracy & calibration
Every v4.1 forecast horizon and every consensus market is graded against the official PSX close. Nothing is hidden — including the misses. Trading guide · Scanner
Graded forecasts0
Direction hit-rate—coin flip = 50%
80% band hit-rate—target = 80%
Brier score—lower is better
No forecast horizon has matured yet. Scores appear automatically once forecasts reach their resolution dates — one session for the 1d horizon, about a week for 1w — and every result is published here, misses included.
Scanner data-integrity log
| Run | Watchlist | Scanned | Rejected | Signals | Avg quality |
|---|---|---|---|---|---|
| 51m ago | breakouts | 25 | 0 | 22 | 70 |
| 51m ago | value | 25 | 1 | 14 | 68 |
| 51m ago | oversold | 25 | 0 | 14 | 68 |
| 51m ago | kse30 | 30 | 0 | 25 | 70 |
| 51m ago | momentum | 30 | 0 | 19 | 70 |
| 52m ago | liquid | 40 | 0 | 32 | 70 |
| 1h ago | liquid | 30 | 0 | 25 | 70 |
| 1h ago | breakouts | 25 | 0 | 22 | 70 |
| 1h ago | value | 25 | 1 | 14 | 68 |
| 1h ago | oversold | 25 | 0 | 14 | 68 |
How to read these numbers
- MAPE — mean absolute percentage error between the projected median and the actual close. Lower is better, but it says nothing about direction.
- Band hit-rate — how often the close landed inside the published 80% band. It should sit near 80%. Much higher means the bands are uselessly wide; much lower means the model is overconfident.
- Direction hit-rate — how often the sign of the move was right. Anything durably above 55% on a large sample is genuinely useful; 50% is a coin flip.
- Brier score — calibration of the stated probabilities. 0.25 is the score for always saying 50%. Below 0.20 across hundreds of resolutions is a real edge.
Track-record statistics are computed from this platform's own stored forecasts and are not audited by a third party. Past accuracy does not imply future accuracy. Not investment advice.