Originally posted by Daniel Kimberg:
As part of an effort to organized data internally I actually encountered the same question. I think eventually, something along the line I am doing will be adopted

Here are the guidelines I used:

• The researcher must document every step. The results must be traceable back to the source data. The instruction must be strong enough to reproduce the result from the source data.
• Applied (1) recursively, until you can offload the responsibility of tracing to a third party. So, for fMRI data for example, you must map your final activation map back to the raw fMRI data you get from the scanner. Then the process stop because the scanner people will take over the responsibility of making sure your raw fMRI data withstand scrutiny.
  
• Anything you used along when performing (2) has to be documented. If you use program, keep a copy of the published source  code if available (FSL/SPM).  If proprietary program used, e.g. Matlab, note the version number used. Also, if appicable, the Operating System version must be documented, such as if the researchers make a detour to a supercomputer to process some of the intermediate data.

The idea is you document everything you do until you can pass the buck to someone else. As I am dealing with internal data management, I can insist that the proprietary program are archived as part of the project. From a journal's perspective they will be prepared to accept that you name the proprietary software and its version, and (if need be) explain the steps you had taken inside the software to process the data.

I don't think for neuroscience they will want Linux source code ... unless you modify the Linux source code. ;-)

HTH
Cinly