Oh well, the issues with sending out code on the forum is that everyone can see the big errors you’ve made in the code 
I have noticed a “bug”.
On the second scan I output
vec2
, but I then use the following line:
vec2_ = vec[trials2, subj2,0] * beta[subj2]
Which explains why the model can’t find an ok alpha. Correcting that issue - everything seems to work well.
Thank you (again) for the great help you provide. It is almost impossible doing these things without this forums.