These questions are are based on your misrepresentation of what others have said and are answered by undoing the misrepresentation.
When is a DM allowed to do a thing?
"Allowed" is a misrepresentation.
"Thing" is overly broad as a means of another misrepresentation.
If we unravel it back to the honest question we get the following answer:
If players (including the GM) in a playgroup have preferences for how much player agency (actual meaning, not Cheesegear's misreprentations) they would have, then it makes sense for the campaign to fall within those preferences OR inform the players that won't want to participate so they can opt out. Obviously sometimes informing all the players is the most effective way of ensuring those that won't want to participate were informed.
Already we notice a few things that were glossed over.
1) Notice it is preference dependent and thus player (including the GM) dependent. One style is not "bad" but imposing a style on players that won't want to participate is bad. This cuts both ways, some player will not want to play a game deep into the emergent style.
2) However despite there not being a "bad" style, there are clear player preferences that you have asked for. Just because different people have different preferences does not mean nobody has any preferences.
3) Fully understanding the answer requires understanding and not misrepresenting Player Agency. You have made a concerted effort to misrepresent it every time it was explained to you. Obviously you are misrepresenting it on purpose. So all you have to do, is stop misrepresenting it.
Now if we track back through previous answers, we come to Tanarii's definition of Railroading as "negating Player Agency". Given the negative connotation they ascribe to Railroading, it is clear that Tanarii's playstyle preference excludes the negating of Player Agency. So it comes to nobody's surprise that I predict Tanarii's playgroup runs campaigns that don't include the negating of Player Agency.
That is the complete answer to the honest version of your first question.