PS: also note that this issue of computer science can be seen as an analogue of the inter-rater variability. There is simply no system in the world that reproduces the exact same results consistently, whether softwares, medical machines (MRI, EEG, etc) or humans. Users tend to think that computer science is rock solid and perfectly reproducible, but that's far from the truth. Computer science might be more reproducible than other methods, but it is surely not perfect.