D09 Forecast Models · guided laboratory · 8 scenarios · 1,440 synthetic observations
AutoReg: forecast first, reveal the holdout second
Choose a scenario, change one model parameter, and step through a 36-observation holdout. State, benchmarks, and metrics update without letting future actuals alter the frozen forecast.
Frozen forecast and causally revealed actuals
trainingforecastrevealed holdoutfrozen benchmark
18-row origin and holdout audit
Training state ends at index 143. Only holdout rows through the selected horizon are shown. The benchmark is frozen at the same origin.
| phase | index | value | forecast | benchmark | error |
|---|