When certain values in one observation file are required more than once and 'merge_across_dims' + 'split_multiselected_dims' are used, the values are misplaced. It might happen when reading experimental data with many start dates, and the corresponding observations are required to have the same data structure.