(Mostly) AI solution

[iamaran]

I asked Copilot to solve The QEGS Adventure (https://solutionarchive.com/game/id%2C7 ... C+The.html) by uploading an UNGACed plain text file. It did pretty well and with my occasional corrections did just that. I've uploaded "our" solution.

I don't think it will be the most efficient (although it kept boasting that it was) because it got stuck in directional loops several times, eg. going N,S,N,S,N,S,E instead of just N,S,E. I removed these where they were obvious, but others may remain. It also made the occasional mapping error.

However, the solution posted is fully working.

I did this mostly out of curiosity, but also seeing as I didn't enjoy the game when I looked at it a few years ago I thought without help it may be unsolved for all time.

[iamaran]

So I tried QEGS II https://solutionarchive.com/game/id%2C7 ... C+The.html and it did not fare well, unable to map the game at all.
However, it did summarise the objectives and how to achieve them, which enabled me to solve it very quickly.
No, it wasn't fun. But I'd decided a few years ago that this game wouldn't be great to play anyway, so an ideal candidate for AI to assist with.
Very interesting.

[iamaran]

Part of the problem (for me) was Copilot's undeserved (over)confidence.

"I will produce a precise, complete walkthrough".
"I will give you a perfect, step‑by‑step solution with no guesswork".

Umm... No. But what was provided was very useful.

[Exemptus]

The idea has some merit, but I'm not surprised output is inconsistent because it's very unlikely that models have been trained on this kind of material (decompiled source to sequenced solution). Theoretically, nothing prevents a model from doing a reasonably good job when the adventure is totally deterministic, as it would be able to parse the whole game logic given enough tokens, but it is a niche task for a general purpose engine. It's surprising enough that the system does even half well at it.

[iamaran]

Agreed. I was similarly surprised at its success. It did really well at QEGS I. Although, if you look at the solution, it's barely a game at all really. However I was impressed by the understanding of the use of the teleport over the standard map.

I decided against keeping the transcript because there were too many backtracking moves and discussions/corrections by me to make it worthwhile.

But it was an interesting experience that I may revisit.

[iamaran]

And to be honest, having looked at these games in the past, a human shouldn't have to be forced to solve them! Though, of course, my input was necessary.

[Mr Creosote]

Would you have the "UNGACed plain text file"? Sorry, I'm not so familiar with these formats, but I'd love to experiment a bit on my side.

[Strident]

I had it look at a Quill source file (produced with unPAWS).

It struggled to give a workable solution, in the game I chose, but what was produced would've helped someone solve it.

It seemed to do okay when I asked it to produce a list of what all the flags (the numbered variables) did...

example:

Flag reference map (Quill variables)

Flag 0 – Darkness / lighting state
Purpose: Whether the current area is dark.
Used in: Under-stage movement and torch logic.
Behaviour:
SET 0 when entering dark areas.
CLEAR 0 when leaving or lighting the area.
Effect: Triggers “Everything is dark.” system message when set.

Flag 5 – Trapdoor linkage state
Purpose: Synchronises trapdoor state between stage and under-stage.
Used in: Lever logic.
Behaviour:
Set when lever is pulled.
Cleared when lever is pulled again.
Effect: Determines whether the trapdoor object is open or closed.

Flag 11 – Sofa searched
Purpose: Prevents repeated sofa searches.
Used in: SIT command in Green Room.
Behaviour:
Set after sitting once.
Effect: Stops repeated creation of the King of Clubs.

Which would be a help when looking at debugging or solving a game.

[Strident]

And, things like this are definitely useful...

Can you produce a breakdown of what changes the score flag, flag 30?

Flag 30 is the score accumulator, and in The Magician it is modified in very deliberate, theatrical beats. Every change is a clean ±10, and every scoring action corresponds to either professional stagecraft or unprofessional behaviour.

Below is a complete breakdown of every place in the response table where Flag 30 is modified, grouped by intent and outcome.

Positive scoring actions (+10)
These all represent correct preparation or performance discipline.

WEAR TAILCOAT — PLUS 30 10
Dressing correctly for the act.

GET RITA (after she changes) — PLUS 30 10
Successfully preparing your assistant.

LOWER CAGE (flies) — PLUS 30 10
Correctly rigging the dove cage.

WEAR TOP HAT — PLUS 30 10
Completing the magician’s costume.

DROP SAW (on stage) — PLUS 30 10
Placing the sawing-in-half prop.

DROP BOX (on stage) — PLUS 30 10
Placing the sawing box.

DROP PACK (on stage) — PLUS 30 10
Preparing the card trick.

DROP RABBIT (on stage) — PLUS 30 10
Preparing the rabbit trick.

DROP BAG (on stage) — PLUS 30 10
Preparing the egg bag.

DROP EGG (on stage, bag present) — PLUS 30 10
Completing the egg bag trick correctly.

These ten actions are the only way to reach exactly 100 points, which is required for Message 35.

Negative scoring actions (−10)
These represent clumsiness, impatience, or poor stage etiquette.

GET CAGE (on stage) — MINUS 30 10
You shouldn’t be handling heavy props during setup.

GET SAW (on stage) — MINUS 30 10

GET BOX (on stage) — MINUS 30 10

GET PACK (on stage) — MINUS 30 10

GET BAG (on stage) — MINUS 30 10

GET RABBIT (on stage) — MINUS 30 10

CLIMB LADDER while wearing or carrying top hat — MINUS 30 10
The hat falls and is lost.

These penalties are subtle but brutal: one careless pickup can permanently block the best ending.

[iamaran]

Would you have the "UNGACed plain text file"? Sorry, I'm not so familiar with these formats, but I'd love to experiment a bit on my side.

Sure:
https://www.dropbox.com/scl/fi/7ym2jgw7 ... pitpp&dl=0

[Strident]

The AI is a little confused about some of the aspects of this game, but I asked it to use a source code file to produce a playthrough of a game from start to finish, as if it was a short story from the point of view of the main character...

I stood in front of the mansion with my collar turned up and my patience already worn thin. Every year it was the same. The place was supposed to be empty, locked up, quiet. Instead, light flickered behind the windows and the sound of music and shouting drifted out into the night. Pesky kids. Again.

I wasn’t about to march straight in. My leg wouldn’t take it, and even if it did, there were more of them than me. So I did what I always do best: I walked the long way round and started thinking.

Out in the garden I found the usual clutter. A bench, bins, a ladder someone had left lying about. I shifted the bench closer to the wall, testing it with my weight, then climbed over into the neighbour’s garden. Someone clearly took pride in their roses. One bush was tied back with twine, knotted tight. I couldn’t undo it, so I snapped a broken tile off the ground and cut the twine free. That would come in handy.

Back on my side, I leaned the ladder against the mansion wall and checked it twice. No sense rushing. From halfway up I could see an upstairs window standing open, a tattered net curtain hanging loose. Inside, a group of kids were huddled together, whispering and telling ghost stories. I smiled to myself.

I pulled the curtain free, nearly losing my balance, then climbed down and fetched a white party balloon I’d spotted earlier. I blew it up, tied it off, and fixed the twine to it. Then I added the net curtain, stepping back to admire my handiwork. It wasn’t much to look at, but it would do.

Back up the ladder I went. I lowered the homemade ghost past the window and swung it gently. At first nothing happened. Then someone screamed. Chairs scraped, feet pounded, and the room emptied in seconds. I hauled the ghost back up before anyone got a good look at it. One room down.

From there I climbed higher, through the window and into the attic. It was dark, so I flicked on my flashlight and swept the beam around. That’s when I saw it: a nest, crawling with spiders. I don’t mind spiders, but I knew exactly who would. I scooped the nest into an old ice-cream tub, snapped the lid shut, and climbed out onto the roof.

Smoke drifted from the chimney below. I tipped the nest in and listened. Moments later, screams echoed up from inside as spiders poured out into the fireplace room. Another group of kids fled, shouting about things crawling in their hair. I allowed myself a quiet chuckle.

Next I turned my attention to the kitchen. Through the window I could see kids preparing food, dipping apples into caramel. Nearby, a hot water header tank sat against the wall. I dropped a dried-out red marker pen into it and waited. Soon enough, there was a scream as someone turned on the tap and saw what looked like blood pouring out. The kitchen emptied fast.

With the room clear, I slipped inside. On the table were onions, sticks, and a pan full of sticky caramel. I dipped the onions, pushed sticks into them, and set the fake apples in the fridge to harden. I didn’t have to wait long. When the kids came back and bit into them, the screams told me the trick had worked.

There were still a few holdouts. I unlocked the cellar hatch and climbed down, finding a freezer and, inside it, a dead rat. I wasn’t touching that without gloves. I put it into an empty pizza box, carried it back up, and placed it neatly on the doorstep. Then I rang the doorbell and hid in the bushes.

The door opened. A kid picked up the box. A moment later, there was shouting and the sound of something being flung down the drive. I stayed hidden until it was quiet again.

For good measure, I lobbed a pair of eyeballs into the punch bowl through an open window. That finished off the last of them. One by one, the rooms fell silent.

At last, the mansion was empty. I stood in the doorway, listening. No music. No laughter. Just the creak of the old building settling. Another year sorted. I turned away, already thinking about the paperwork and the cleaning crew I’d have to call in the morning, and headed home, satisfied that the night had ended exactly as it should.

[Mr Creosote]

Very interesting! I fed this into several different models. Each of them found the winning condition with ease. But when it came to actually writing down a step-by-step solution, one model just laid out generic, high-level descriptions. Two others said "OK" and started printing commands, but apparently got stuck in endloess movement loops at some point and timed out. One managed to provide a full (untested) solution, which contains a few objects which are not mentioned in yours (e.g. the kipper tie, the trousers and the crystal). The answer contains a dry remark: "Note: The game files contain a massive amount of repetitive maze data and redundant room connections in the provided source text."

For QEGS 2, the same model also gave a full step-by-step solution which looks reasonable.

I then tried to achieve something similar with decompiled Z-Code games. Though it seems those are all too big. The models give up with a "maximum number of tokens exceeded" message. I tried fiddling around a bit by simply removing message strings etc., but not successful yet. The models recognise the test game's setting and various objects, list possible interactions, but it seems they cannot even identify the winning condition, let alone surmise how to get there.

The ridiculous number of locations aside, those QEGS games seem to be very sparse on objects, possible interactions etc. That makes them easily parseable, I guess.

[iamaran]

The ridiculous number of locations aside, those QEGS games seem to be very sparse on objects, possible interactions etc. That makes them easily parseable, I guess.

Yes, they are long on location and short on interest or puzzles. Which presumably explains why they were unsolved.

In QEGS 2, there is one location that has an EW location typo. I.e. west from 33 to 44 (or whatever) but from 44 east went to 38 instead of 33 (for example). That may account for something, but shouldn't have made it loop like it did because the model simply reads the code and follows it, regardless of what the human eye immediately sees is wrong. And it was irrelevant to the gameplay.

[iamaran]

Very interesting! I fed this into several different models. Each of them found the winning condition with ease. But when it came to actually writing down a step-by-step solution, one model just laid out generic, high-level descriptions. Two others said "OK" and started printing commands, but apparently got stuck in endloess movement loops at some point and timed out.

Did they all boast how great they are, like Copilot!?

[Mr Creosote]

Kagi "Quick" Assistant:

Code: Select all

Solution to the QEGS Text Adventure

Based on the provided text adventure file, here is a complete solution to successfully finish the game.
Objective

The goal is to obtain both the Report (object 14) and Card (object 29) and bring them to room 60 (Headmaster's office) to receive a "Good Record" and complete the adventure 【1】.
Step-by-Step Solution
1. Starting Position

Begin in the Junior School Hall (room 1). The immediate goal is to collect essential items and navigate through the school environment.
2. Obtain the Kipper Tie

    Go SOUTH to the Gym (room 2)
    Go EAST to backstage (room 3)
    Go EAST to Headmaster's Garden (room 4)
    Go SOUTH to Trendy Trevor's Games Office (room 5)
    Go SOUTH to the Fields (room 11)
    Go EAST to College Grove Road (room 6)
    Go EAST to Preen's Shop (room 187)
    Go EAST to the Subway (room 188)
    Go EAST to the Bus Station (room 189)
    Go SOUTH to the Bullring (room 194)
    Go EAST to JAT Records (room 195)
    Go NORTH to Outside Tesco (room 190)
    Go SOUTH to Westgate (room 191)
    Go SOUTH to Rooftop Gardens (room 196)
    Go SOUTH to Casanova's Nite Spot (room 201)
    Go NORTH to Rooftop Gardens (room 196)
    Go NORTH to Westgate (room 191)
    Go WEST to Outside Tesco (room 190)
    Go WEST to the Bus Station (room 189)
    Go WEST to the Subway (room 188)
    Go WEST to Preen's Shop (room 187)
    Go WEST to College Grove Road (room 6)
    Go SOUTH to the Fields (room 12)
    Go WEST to the Fields (room 11)
    Go SOUTH to Db's office (room 17)
    Go EAST to room 9, Mr. Furniss's room (room 18)
    GET TIE (object 42) 【1】

Then it details steps for "3. Trade for Trousers", "4. Trade for Antifreeze Crystal", "5. Obtain Frogs Legs". When it comes to "6. Access the Headmaster's Office", it goes into an endless death spiral.