[Portugal] Nature on "The case for open compute programs"

[Giovanni Manghi]

http://bit.ly/KlcjeA


"...Nature’s policy on code release implies that algorithmic
descriptions using mathematical specifications, equations, formal
algorithmic descriptions or pseudocode (simplified version of complete
code) may be required. But there is no guarantee that such tools can
avoid ambiguity, and even if they could, we have shown above that
implementation and numerical errors—possibly compounded by differences
in machine architecture—will still arise. So, even if complete code is
made available, exact replication or even reproduction of central
results may fail. A reasonable observe might therefore ask why code
should be made available at all. Our response is that the alternative is
far worse. Keeping code closed ensures that potential uncertainties or
errors in a paper’s conclusions cannot be traced to ambiguity, numerical
implementation, or machine architecture issues and prevents testing of
indirect reproducibility. Although it is true that independent efforts
to reproduce computational results without recourse to the original
source code constitute an important approach, the all-too- common
treatment of code as a black box unnecessarily slows and impedes valid
efforts to evaluate model results. We therefore regard the
non-availability of code as a serious impediment to reproducibility..."