Important point is if the parameters in different models/datasets are linked (set to be treated as one) and you assign different values to it using assign_parameter_values() for two different datasets consequtively, then the last loaded parameter value will be placed in the corresponding position of all-parameter vector but NOT in the best-fit parameters of the first dataset. This way fitting will start with correct (last) initial parameter but if you want to examine initial guess before you fit you have to force the last version of this 'common' parameter onto the first dataset. The way to do it is to load parameters for ALL datasets and execute the method transfer_all_parameters_to_individual_datasets(). This method sends all parameters from all-parameter vector (alongside with the parameter ranges) into individual datasets. Now individual plots may be made (dataset_plot_fit()) to see quality of an initial guess in every dataset.
For a specific usage see TotalFit.m