[mrow84]

It's important to recognise that a lot of the model runs are done on super-computers, and the runs themselves aren't always easy to replicate from a practical point of view, because of things like resource limitations and configuration issues.

I certainly agree that things should be open sourced, probably allowing for some kind of embargo system for publication, but we shouldn't allow a rhetoric to develop suggesting that people should be making things easily replicable, because that essentially means making very portable and resource-flexible code, which is not trivial, even in less complex domains, as I'm sure many people on this forum recognise.

As an example, many models make use hundreds of cores via MPI (along with preposterous amounts of RAM), and whilst I am no expert, I get the impression that porting such things to more commodity hardware is essentially impossible.

As an aside, I have seen more bad-faith practice (i.e. manipulation of results) in industry than in science, and have spent approximately equal periods of time in both.

[gaius]

Today's supercomputer is a GPU in 5 years time.

[mrow84]

Maybe in terms of various metrics (though I'm not so certain things are progressing that quickly), but not in terms of source compatibility, which was the point I was trying to make.