If their justification is "we want to see how your algorithm made its decision", you should perhaps have your model output more stats and infos about that. As

If their justification is "we want to see how your algorithm made its decision", you should perhaps have your model output more stats and infos about that. As a user of data tools myself, I am generally suspicious of any black box model, and would like to understand the model well before using it. For instant if your model is a deep neural network, I'd like to know the structure of the network and the activation of the layers when I run my data through it. If they're really interested about the guts of the model they will agree to a solution like that. Having the source code will certainly not help them understand this as it is highly unlikely that anyone will dig into this. Note that asking for the source code is fairly common practice in finance cause people generally distrust black boxes (at a firm I worked for they specifically chose MySQL over some other tools because they had the ability of looking through the code if they needed to). There is actually no point for them to look at your code... You can agree to something where if you go bust, then you'll give them the source code.