If your process isn't sufficiently documented so that your professional peers could reproduce your results with your data, then your process is unreliable. We berate academics all the time for improper or incomplete documentation of manual experiments. I can't see any reason why source code should be held to a lesser standard.
Well, really it's the algorithm that needs to be published so others can reproduce the results. In principle, publishing code is similar to publishing the sequence to put together your experimental apparatus, which isn't done either.
