Table 3. Comparison of province-level prediction accuracy between full-data and real-time predictions, by prediction horizon.

These results were computed comparing predictions as if the full data was available at the analysis time with the real-time predictions that build in a 6-biweek (approximately 3 month) buffer to account for delayed case data. The table shows the 5th percentile (Q₅), 25th percentile (Q₂₅), median (Q₅₀), 75th percentile (Q₇₅), and 95th percentile (Q₉₅) value of the relative MAE from each province at the given horizon. The relative MAEs were calculated as the MAE from the real-time predictions divided by the MAE from the full-data predictions, i.e. values larger than 1 indicate that the real-time models showed more absolute error on average than the full-data models.

Absolute horizon	Relative MAE (real-time vs. baseline scores)
	Q 5	Q 25	Q50 (median)	Q 75	Q 95
1	1.37	1.77	2.81	5.55	11.26
2	1.01	1.61	2.89	4.88	12.23
3	0.98	1.68	2.94	4.25	9.55
4	0.92	1.83	2.86	4.14	8.74