Table 2. Modeling Strategies Evaluated in the Study^a.

strategy	aggregated models	proper training set	calibration set	exchangeable
CCPnew	10	Anew	Anew	×
CCPold	10	Aold	Aold
CCPpool	10	(Aold ∪Anew)	(Aold ∪Anew)
ICPoldnew	1	∀Aold	∀Anew	×
CCPAT	10	Anew ∪ ∀Aold	Anew	×
CCPAT2	10	Aold	Aold ∪ ∀Anew

^aThe A and A notation should be interpreted as 1 or 9 parts of a 10-fold split of the data set A, where the folds are shifted for each ICP model in a similar fashion as in cross-validation test-train splits. The last column indicates whether the model’s calibration set is exchangeable with the test data, i.e., theoretically guarantees valid models.