Hi Alistair,
Yes, we know this is a bit confusing currently. The problem is that dwi2fod tries to use the b=0 image here as well, due to the msmt_csd algorithm, but you only offered a response that has no b=0 part in it, due to the tournier algorithm. What you are trying to do here is essentially single-shell single-tissue CSD, but just using the msmt_csd algorithm’s code. That is ok, there’s nothing wrong with that. But, the msmt_csd algorithm then needs to only get the single-shell part of your data, and not the b=0 image(s) as well. To resolve this for your example here, replace your step 2 by:
dwiextract dwi.mif - | dwi2fod msmt_csd - rsp.txt rsp.mif -debug -info
Note that if you do this, you best don’t describe this in a paper as if having done MSMT-CSD; because as mentioned, you’ve only been doing single-shell single-tissue CSD, but while exploiting the benefit of the “hard constraint” in the msmt_csd code.
Again, I/we fully agree this is confusing at the moment; this will eventually become much more clear when the need for a single-shell single-tissue “exploitation” of the msmt_csd algorithm will disappear.
Also, I will post something else below to give you another good option for your scenario: doing MSMT-CSD with 2 tissue types (WM and CSF) on your data. Bear with me for a moment, as I gather some text from elsewhere to explain this… 
Cheers,
Thijs