Probabilistic forecasts, scored out of sample.
For a public set of large-caps Stradion publishes the probability of an up move, the expected return, and a 90% range — then scores every call once its window closes, against only the data available at the time. Misses included. Free, no account.
Signal Matrix
| Symbol | P(up) 20d | Exp. return | 90% range | Model | Last 5 scored |
|---|---|---|---|---|---|
| AAPL |
64% |
+3.1% | -2.4 / +8.1 | deep-xgb | |
| MSFT |
57% |
+1.9% | -3.0 / +6.2 | deep-xgb | |
| NVDA |
48% |
-0.7% | -9.0 / +7.5 | vol-base | |
| AMZN |
59% |
+2.4% | -3.6 / +7.9 | deep-xgb | |
| GOOGL |
55% |
+1.2% | -3.9 / +6.4 | deep-xgb | |
| META |
61% |
+2.8% | -3.1 / +8.6 | deep-xgb | |
| TSLA |
47% |
-1.1% | -10.2 / +8.4 | vol-base | |
| JPM |
53% |
+0.9% | -3.2 / +5.1 | deep-xgb |
Every forecast is scored out of sample.
Each call is graded after its window closes, on the data the model could have used at the time — a walk-forward, out-of-sample replay that includes the stretches where the models do poorly. Short-horizon directional confidence doesn't yet rank outcomes well enough to trade on; it begins to separate around the 20-day horizon, and that result is stated on the record. A model that is right 57% of the time is wrong the other 43%, and the published numbers reflect it.