Track Record
How our predictions have actually performed, measured against what the market did next. Published in full — including when the results are unflattering.
Based on 44 trading days of live predictions · last updated September 16, 2026
| Horizon | Observations | Correct Direction | 95% Range | Better Than Chance? |
|---|---|---|---|---|
| 5min | 500 | 50.1% | 49.3% – 50.8% | No |
| 30min | 500 | 50.1% | 49.1% – 51.2% | No |
| 60min | 797 | 49.8% | 48.4% – 51.2% | No |
| EOD | 10522 | 50.6% | 49.1% – 52.1% | No |
What these numbers mean
Across every horizon we measure, our model's directional accuracy is not yet statistically distinguishable from chance. Each 95% range above includes 50%, which is what you would expect from coin-flipping.
We publish this because you deserve to know it before you pay us. The Service is a research and analysis tool — it shows you what the model expects, how stocks rank against each other, and which signals are firing. It is not a proven way to make money, and we do not claim otherwise. If that is what you are looking for, please do not subscribe.
We continue to collect data every trading day, and this page updates as the evidence grows — in whichever direction it goes.
How this is measured
- Every prediction the model publishes is recorded automatically, with a timestamp, before the outcome is known.
- Each one is later compared against the actual price at four horizons — 5 minutes, 30 minutes, 60 minutes, and end of day.
- "Correct Direction" is how often the up/down call matched what the price actually did.
- Observations count independent symbol-days, not raw rows. Many predictions on the same stock in one day share the same outcome, so counting them separately would overstate confidence.
- The 95% range comes from a bootstrap that resamples those symbol-days, which is what keeps the confidence honest.
- Nothing here is back-tested or curve-fitted: these are live, forward-looking predictions scored after the fact.
We also test challenger models the same way. One was rejected in validation for failing to beat this baseline — challengers are only promoted on evidence.