Can off-the-shelf time-series foundation models forecast how a process's directly-follows behaviour evolves? Explore the paper's bundled backtests, or upload your own log for a live forecast of its next week.
The real last week is held out and forecast from the rest, then scored against what actually happened. The right-hand graph is that same-period actual week — the held-out ground truth — so ER / MAE / RMSE are genuine accuracy.
Dataset
One of the paper's four event logs
Model
The time-series foundation model that produced the forecast
Forecast — the held-out last week, predicted
Actual future — the same week's real DFG (held-out ground truth)