Yeah so many factors involved and little things that aren't ever mentioned in the paper. Implementing a few models and algorithms from papers made me realise how poorly they're written a lot of the time. Code publishing really is essential for validation