Why does the pre_rmspe returned by the program sometimes differ greatly from the hand-calculated value?

The _pre_rmspe_ stored in the data by **synth_runner** is consistent with the result returned by **synth**. However, in some cases (which is quite common), it differs from the rmspe calculated using the “_Y_treated” and "_Y_synthetic" values returned by **synth** (within the pre-treatment period), sometimes with large magnitude differences. But sometimes they are exactly the same. Why is this?