It is true that you don’t want anomalies to be anomalous only with respect to some arbitrarily chosen (or sought) feature of the reference periods. The solution to this is to do sensitivity analysis of the anomaly to these ‘arbitrary’ features (length, starting/ending points). With most climate things, anomalies will persist. This is a standard thing to do.