In any standard FRPG, if the PCs find a castle then that castle will have existed in the setting prior to them finding it. And it would have existed even if they had not. In the BW game in which my characters found Evard's Tower, the tower existed before they found it. How else could they have found it?! In fact my PC found evidence in the tower - ie letters apparently written by his mother as a child - which implied the tower had existed from well before he was born.
So it is not at all distinctive of a sandbox that a castle, or tower, or any other relatively permanent thing should exist in the setting independent of who finds it.
Similarly for Emerikol's dungeons: presumably if Emerikol decides to use a campaign world for another campaign, set a year or so after the previous campaign, and drops in a new dungeon that is 1,000 years old, then that dungeon existed in the world during the last campaign too (and was about 999 years old when that old campaign finished).
So when Emerikiol says it may not have existed in the sandbox earlier I think that means the GM hadn't thought of it yet, and so hadn't written it down. And when Bedrockgames says it should exist in a sandbox whether or not they find it I think that means the GM should have thought of it already and written it down, so that (eg) the GM can narrate signs of its existence.
I don't know how those two claims - and related claims about verisimilitude, feeling "real", etc - are to be reconciled.