Ok so I stopped with Eye of the Beholder 3 for now, that game's too broken to be fun, and it's just a slog. Really the only fun element I found was manually mapping it out, and that's saying something since there are in fact auto-mapper plug-in apps for the Eye series.
After that I've started going through old Amstrad games from when I was a little kid, finishing ones that screwed me over and I never completed. And this has lead me to some very poorly designed text adventures (the better designed ones were difficult but could be finished).
Completed a text adventure called The Secret of Bastow Manor that me and my sisters pulled our hair out trying to solve. We didn't get very far, and looking back at this has brought up a raft of design issues, as well as outright *broken* things that would have prevented us completing the game even if we did navigate its stupid parser. There are a lot of objects in this game that you can see on the screen, and have to guess their name, or even *guess they even exist*. And the LOOK command isn't helpful enough. The game has pictures, and you get stuff for example a square on the screen and you're meant to work out that this is a "CASE" and do "OPEN CASE", then you can LOOK CASE and see what's in it. However if you LOOK CASE before opening it, you just get a generic message "You see nothing special", but this message also appears if you type LOOK and random gibberish. There are well-designed text adventures and this is not one of them. If there are random nouns you're meant to work out, there need to be clues. This game doesn't even give you LOOK ROOM or the equivalent to tell you what things are called or what you can interact with.
The real problem here is that you don't know what the nouns are until you type them with the correct verb: LOOK, MOVE, GET, PRESS, READ, OPEN, UNLOCK etc. if you do the correct noun, but the wrong verb, it just says a generic message about how you can't do that. A better design here would have been that if you "LOOK XYZ" then it checks if there's a combo of VERB NOUN for XYZ at your location and then gives you a different message "you can see XYZ". Then you'd know you're on the right track. But you get nothing, so often you did the right noun but the wrong verb and then you try something else, never hitting the exact right combo. A better designed text adventure mitigates this parser-guessing by making failure give you a little information. If you say PRESS XYZ but there's no XYZ it could say "you can't see XYZ" but if XYZ is recognized but not pressable it says "You can't press that" instead. I ended up following a walkthrough to beat this game however, since doing the guessing thing is only fun for so long (hence why we gave up).
All that would be hard enough, but do-able. However, going through the walkthrough I got to a blood-boiling moment. The game is ported and it looks like they re-did the artwork. And whoever ported it decided that some plot-critical objects weren't worth drawing on the screen. There's ARMOR as an object, and you can't get past a certain screen without the armor. But there are no clues that you need the armor, and the room with the armor seems to contain neither visual nor textual clues that the armor exists, or that it's in this room. I can only assume that it was drawn in the original version of the game. You can in fact look at the source code of the game and see GET and ARM as tokens, but then you gotta still brute-force it by typing GET ARM in every room of the game, and that's presuming you knew that's the combo you were missing. So the game was in fact not just punishingly cryptic in terms of guessing the games for things, it was just broken and unfair.