When I published with an Elsevier journal they explained the "draft" thing pretty clearly. They said you retain copyright on whatever you write yourself. But once you have reviewer feedback and incorporate feedback from the journal editor(s) then it's no longer entirely your work and you couldn't distribute it at your own will. You could pay the journal a huge amount of money (in effect for the work done by the editors), and then the paper would be open access. The fees might not be reasonable (it's kind of hard to judge) but the overall logic made sense to me

So the first draft before review - the one you wrote all on your own - is what you can put up online. I'm not in CS but I assume the logic is similar in other fields.

They also provided a separate link of the final published PDF that you could use to distribute the paper to colleagues and interested parties. This link worked for a sufficiently long period of X amount of days/months. After that it was paywalled and in their garden