Replicating Hu et al. data generation / preprocessing should be easier

I am trying to replicate the data processing in Hu et al. but it's proving to be tricky. Should I be using "V4" inputs? The variables in data.utils (e.g. v4_inputs) are not actually described. For example, what are the "tm" variables, e.g. "tm_state_t_dyn"? I could not find these variables in the table of the ClimSim paper either.

Also, how do I preprocess the inputs in the same way as Hu et al. ? The only input preprocessing I found in data.utils is using a variant of min-max, but not all inputs in Hu et al. were normalized using min-max, some of used x/std and some used 1 - exp(-λx).

Should I first run ds_input = (ds_input - data.input_mean)/(data.input_max - data.input_min) and then manually preprocess select variables using the other two methods? But won´t the above command normalize everything with min-max?

Start by reading data.utils, especially the v4_inputs and tm_state_t_dyn definitions, then compare them with the Hu et al. preprocessing description and the ClimSim paper table. Document the meaning of the variables, identify the intended input version, and describe the complete preprocessing sequence so replication produces the expected inputs.