• The docstring for linear_truncated and the class docstring describe what the behavior would be if the default were linear_truncated=False. However, the default is True. The model uses a LinearTruncatedFidelityKernel, which isn't described at all.
• It is not clear how data fidelities are to be interpreted. Does a higher number indicate higher fidelity? Can they lie outside [0, 1]? Do they have a cardinal or ordinal interpretation?

This came up in #2393 , and I spent a while myself trying to understand how fidelities should be specified and how they affect the kernel.