@Francois regarding the 'raw' and 'lfp' labels:
unfortunately although NWB is a standardization format, they give too much freedom in labeling.
I had this discussion in the past with the developers and this will be problematic for support by us unless we are more strict with what should be used (but then, if a file is already created and not made based on our "stricter" guidelines it won't be loaded). They suggested maybe to have a brainstorm compatibility check, but this has not been done yet.
Too much freedom is what's going on here.
As you can see in in_nwb_read, I take the "most important" signals as a guide (electrophysiological - raw, lfp), and use their sampling rate as the main one. The behavioral - 'less important' signals, are downsampled or upsampled to match the sampling rate of the raw/lfp. The only way to know which signals should be considered as the guide for the sampling rate is if I force it to be with the label raw or lfp.
If you want this to be semi-automated, I would suggest there is a pop-up when the file is originally imported (in in_fopen_nwb) that asks the user to select one of the present keys within the nwb files as the "sampling guide".
I'll check if I can find a file.