WTI & Brent crude — official daily data (FRED / U.S. Federal Reserve).
Forecast — gradient-boosted model, 80% confidence band
| Benchmark | Horizon | Forecast | Low (10%) | High (90%) | P(up) |
Backtest skill* | Direction acc. |
🔮 Tomorrow’s Call — news-aware self-tuning ensemble (next trading day)
Each prediction blends three “experts”; the weight bar shows how much the system currently trusts each, which auto-adjusts from past accuracy. Naive=price unchanged · ML=price model · News=headline sentiment.
📊 Self-Tuning Scorecard — how each expert has actually performed
| Benchmark | Scored days | MAE naive | MAE ML | MAE news | MAE ensemble | Ens. direction |
Lower MAE (mean abs $ error vs next-day actual) = better. The ensemble continuously shifts weight toward whichever expert is winning — so if news sentiment proves predictive it earns more influence; if it’s noise, its weight decays automatically. Over the seeded history the naive “unchanged” baseline is hard to beat at 1-day, which is the honest reality of crude; the news expert’s real test begins with live scored days from here forward.
💵 Paper Trading — would the signal actually make money? ($10,000 start, no real money)
📰 Latest Oil News & Sentiment — AI-scored (Gemini)