In some simulations, each rerun produces different results as you’re simulating random events (like lightning formation) or using a non-deterministic algorithm (like Monte Carlo sampling). Just “saving the random seed” might not be sufficient to make it deterministic either, as if you do parallelized or concurrent actions in your code (common in scientific code) the same pseudorandom numbers may be used in different orders each time you run it.

But repeating the simulation a large number of times, with different random seeds, should produce statistically similar output if the code is rigorous. So even if each simulation is not reproducible, as long as the statistical distribution of outputs is reproducible, that should be sufficient.

Very interesting. I was thinking of software as most similar to apparatus, and secondarily to procedure. You raise a third possible comparison: calculations, which IIUC would be expected to be included in the paper.

There are some kinds of code (a script that controls a sensor or an actuator) where I think that doesn't match up well at all. There are plenty of kinds of code where they are, in fact, simply crunching numbers produced earlier. For the latter, I'm honestly not sure the best way to treat it, except to say that we should be sure that enough information is included in some form that replication should be possible, and that we keep in mind the idea that replication should involve human interaction.

This is not clear at all. It depends on the "result" in question. If I wrote a paper describing a super numerical algorithm for inverting matrices, and no one is able to replicate the superior performance of my algorithm despite following the recipe in my paper, then whether they can run my code or not doesn't seem to be of the highest priority.

Edit: more careful phrasing.

> whether they can run my code or not doesn't seem to be of the highest priority.

On the contrary; in that case there are four possibilities:

a: your algorithm doesn't work at all, and your observations are a artifact of convenient inputs or inept measurements.

b: your algorithm works, but the description in the paper is wrong or incomplete

c: your algorithm works as described, but the replicater implemented it incorrectly

d: other

Having the original implementation code is necessary to distinguish between cases a and b versus case c, and if the former, the code for the test harness is likely to help distinguish a versus b. (Case d is of course still a problem, but that doesn't mean it's reasonable to just give up.)

I agree with the case analysis, but disagree with the implication that the code needs to be runnable (which is seems to be the point of the discussion at hand). In many cases having source code, even if it no longer runs, should be sufficient.

I complete agree with your case analysis, but disagree with the conclusion that the code needs to be runnable for it to be useful -- I thought this was the point of the discussion at hand? In most situations, having source code, even if it no longer runs, would be sufficient to conduct the analysis you describe.

I'm all for more transparency, and this includes making codes and data public as much as is reasonable. But the real test is if someone can independently replicate the result, and how to incentivize replication studies (in both computational and experimental science) is also important, and in my view should not be divorced from discussions of reproducibility.

Edit: rewrote to clarify my position.