Very cool. Did a very little testing for Infocom back in the day and knew a lot of the folks involved.

It has been a while since I've played interactive fiction, so I can't make a specific suggestion, but modern games seem to be better at keeping the directions consistent (or at least providing clues when they are not). As others have noted, older games broke directionality to serve as puzzles -- failing to acknowledge that some people have a sense of direction while following twisty paths!

Newer games also tend to follow some quality of life rules in their design, things like avoiding arbitrary deaths and avoiding situations where the player cannot progress because they missed something earlier in the game.

Not ZMachine related, not even with Inform6, where n_to, s_to and w_to e_to are pretty much self-explanatory. It's just that Dungeon/Zork/Zork-I-II-III were made that way.

Nothing to do with the VM. The game was designed that way, navigation is part of the puzzle.

This was never a technical oddity. This was generally a tool of verisimilitude: the real world isn't built on a clean square grid. You might have a diagonal hallway or a road that curves or a passageway between two "rooms" with a strange incline.

On the one hand, you could encode the hallway itself or the road curve or the passageway as their own weird segments in the grid. But then maybe you bog the player down in a lot of liminal spaces that don't really add much to the game. On the other hand, you could ask the player to bring or build their own map and pay attention to descriptive text like "to the north is a passage that seems to bend to the east" or far more subtle variations of such.

Zork and many other IF games were built on the premise that people would map things and getting lost or confused by grid breaks was part of the fun. (Going back to, as neighboring comments point out, the original Adventure which was modeling caves and caves have always had strange three-dimensional twistiness that doesn't fit a square grid. Part of the fun was discovering that disconnect between the game mechanics using square grid compass terminology and the digraph of the game spaces being more confusing than that.)

Don't imagine a grid.

Instead imagine that "GO EAST" takes you onto a winding road.

The road arcs southward and to go back you'd have to "GO NORTH".

Further... some travels in Zork even drop you through a hole (though I forget if it tells you so).

So going East might put you in the basement and there's no way to climb back up.

Perhaps you could try making a map.

It was impressed on me in college how much the MOS 6502 is sort of a fascinating line in the sand for a "clockwork computer". You can visualize all of the 6502 registers and all of its easily addressable RAM in a bitmap you can see on a modern screen and its general clock speeds were so much more relatively close to human scale so you could watch such a visualization compute in (near) real-time.

I remember building my own mini-6502 debugger of my own for an assignment I was having difficulty debugging on the breadboarded hardware in our lab (and also to save me from having to spend all of my waking hours in that lab). I want to say it was in something stupid like PHP at the time, but it was just a quick and dirty hacked together thing I did in like an hour during a lecture for a different class.

I've seen some really cool projects on HN of people building visual/visible 6502 boards. LEDs have gotten so cheap and the 6502 is so easy to replicate in FPGA and other hardware spaces that you can build real neat 6502 computers. Which I think the original homebrew computer club hackers would still respect all these years later. (The 6502 fueled the original home computer revolution and was in so many machines, from the Apple II to the Commodore 64 to the NES and more. It was such a key workhorse chip.) I'm still glad I had a lab course on the 6502 in my college education. (The follow up one with the Motorola 68k was maybe less successful, and mostly taught us why the 68k is the last real breadboardable CPU in so many ways including the way that it frequently burned through breadboards and made most debugging a hardware issue.)

Infocom came out of the MIT AI Lab which was very LISP-centric at the time. Depending upon you you listen to, the gaming part was something of a side effect as opposed to the ultimate goal.

Inform 7 source code, at first glance, looks like plain English. It makes you start to wonder whether Inform 7 is actually an LLM and not a compiler. Then as you look at more code, you quickly start to see the structure of the language, and realize that it is indeed a programming language merely structured to look like English at first glance. But it's very cool that you can do things like this (example from the Inform 7 documentation):

    Definition: a room is neighboring if the number of moves from it to the location is 1.
    Every turn:
        if a random chance of 1 in 2 succeeds:
                    let current location be the location of the lurking critter;
                    let next location be a random room which is adjacent to the current location;
                    if the lurking critter is visible:
                        say "The critter [one of]slouches[or]slithers[or]shambles[or]lurches[at random] away.";
                    move the lurking critter to next location;
                    if the lurking critter is visible:
                            say "A critter [one of]oozes[or]staggers[or]ambles[or]creeps[at random] into the room.";
        [Whether or not the critter has moved, we need to adjust the sword-glow, because the player may have moved.]
        if the lurking critter is in the location:
                  adjust sword-glow to glowing brightly;
        otherwise if the lurking critter is in a neighboring room:
                  adjust sword-glow to glowing faintly;
        otherwise:
                  adjust sword-glow to glowless.

This snippet omits the definition of "the lurking critter" (a standard NPC) and the sword (a standard object with a custom defined method called "adjust sword-glow to"), but it should give you a good idea of what Inform 7 source code looks like. (Inform 6 looked much more like a traditional programming language).

TADS is hardly modern. It’s actually the oldest VM based platform from the early 90’s.

I wonder if you could make a "choose-your-own-adventure" book uning LLMs.

It would write the book ahead of you, kind of like how railroads could send trains with supplies out to the track as it was built.