Inputs and a complete first forecast | A consistent output structure | Choosing a baseline engine | Method and interpretation | Evaluation and its limits | Marginal probabilities and joint events | Updating the forecast | Positive prices and log transformation | Reading the four-panel plot | Choosing computational settings | Common questions and troubleshooting

A smaller runtime | A runnable example | What is learned, and where | Migration and reproducibility

Numeric example | Reproducibility | Distance methods and parallel execution | Changes in 2.0.1 | Input requirements

A first forecast | Optional features

Overview | Numeric forecasting | Categorical forecasting | Reproducibility

Overview | Forecasting | Reproducibility

What you can do with rollcast | 1) Fit rollcast, aka anchor first, argue later | 2) Peek at the anchor mix | 3) Forecast paths, not just a line | 4) Try fixed mode versus search mode | 5) Residuals: pure anchors or local texture | 6) Minimal plots, tiny but telling | Conclusion

CRAFTing future trajectories without pretending to own the future | The CRAFT package works this way | A tiny demo, because forecasts should have manners | 1. Fold the history into trajectories | 2. Fit a smooth distribution from trajectory draws | 3. Fit CRAFT end to end | 4. Predict again, because the future keeps moving | 5. What to look at when the model returns | Final thoughts