Mystery Interactive Fiction Needs a Fixed Answer

Mystery interactive fiction only works when the answer exists before you go looking for it. A story that decides who did it at the moment you accuse someone is not a mystery; it is a mood with a suspect list attached. What the form is good for — noticing, doubting, being wrong in a way you can trace back afterward — needs a fact sitting under the prose that your attention can be measured against.

• A mystery is a promise that a fixed answer exists and can be found by paying attention.
• In the published source for Infocom’s Deadline, suspects follow a MOVEMENT-GOALS table of rooms and times that runs whether or not you are watching.
• The same code advances the clock one minute per turn and ends the case when it passes 1199.
• Deadline’s default reply to arresting a suspect is that you lack the evidence to convict, so naming the culprit is not the win condition.
• A narrator writing live holds no sealed answer, so a mystery read that way depends on a record you can hold it to.

What does mystery interactive fiction demand that other stories don’t?

It demands that you be able to fail on your own merits. In a novel the detective’s attention is the author’s attention, and the two arrive at the solution on the same page. Mystery interactive fiction hands the attention to you instead, which means the story has to contain something you can miss: a detail placed early, left alone, and still there when you come back for it.

That is a harder promise than it sounds. Almost everything else in a story can be improvised in front of a reader without anyone being cheated — a room, a stranger, a change in the weather. A solution cannot. Once the answer is allowed to form at the moment of the question, every hour you spent watching was spent on nothing, and the story has no way to tell you so.

This is what the fair-play convention in detective fiction is protecting. The rule was never that the reader should guess correctly. It was that the reader should be able to, from what the page has already shown, and that the author should not be able to move the answer once the reader starts closing in.

The three places a culprit can be decided

Two of those three are honest. An answer written by the author and an answer drawn at random before the first scene are equally findable, because in both cases the truth is older than your first question. Randomizing which of several arrangements is true costs the reader nothing; it only costs them the ability to look the answer up. The third case is the one to watch for, and it is the easy one to build.

Where is the answer kept in a classic detective game?

In machine state, and Deadline is the clearest published example. Infocom’s 1982 murder investigation has had its ZIL source posted on GitHub for education, discussion and historical work, and read from that side the mystery turns out to be a timetable, a clock and an evidence check.

The suspects keep a schedule. A MOVEMENT-GOALS table gives each character a room and a time; the routine that reads it queues the next move as a timed interrupt, and adds a small random variation to the minute it fires. The randomness is in the timing, never in the truth. If you are somewhere else when a suspect walks upstairs, you missed it, and the house does not replay itself for your benefit.

Time is one global. The clock routine adds a minute to PRESENT-TIME every turn, and when the value passes 1199 an inspector arrives, tells you your time is up, and puts you in a waiting police car. The twelve hours of the title are not atmosphere; they are a numeric comparison you can read in the source.

The ending is gated on evidence rather than on the name. Arresting is its own verb, taking a person and the thing you are arresting them for, and the default response to arresting somebody is that you realize you do not have enough evidence to convict, so you resolve to continue the investigation. Being right is not the same as being able to show it, which is a rule that belongs to detective stories more than to games.

Why does a live narrator struggle with a whodunit?

Because nothing is holding the answer on its behalf. Authored interactive fiction keeps its truth outside the prose in a form the story can test. Inform describes itself as a programming language for creating interactive fiction using natural language syntax; ink, inkle’s scripting language for interactive narrative, declares variables with VAR and treats a visited passage as “an integer variable containing the number of times the content at the address has been seen”. The story asks a number, and the number does not have opinions.

A generated scene has none of that unless somebody put it there. It is consistent with whatever it was handed on that turn, which is why limits in AI stories only survive when they are written down where the story can check them. A culprit that was never recorded is not being kept from you. It is being made when you ask, and made again, differently, tomorrow.

The visible symptom is that nothing can be a red herring. A red herring is a deliberate wrong answer, and deliberateness requires a right answer to be wrong against. Without one, a misleading detail and a contradiction look identical from the reader’s chair, and the reader learns to stop taking notes. That is the real loss: not a broken puzzle, but the end of the reason to pay attention.

How can you tell a fair mystery from a bluff?

Test the three things a mystery cannot fake: persistence, independence and refusal. Go back to an early detail and see whether it is still exactly what it was. Wait somewhere and see whether the world moves without you. Then accuse the wrong person early, and see whether the story lets you be wrong.

What to test	A fair mystery	A story bluffing
Re-examining an early clue	The detail comes back unchanged, in the same words	The detail shifts, softens, or is gone
Letting time pass	Suspects keep their own schedule and are elsewhere	The world waits politely at the door
Accusing the wrong person	The accusation fails and the case continues	Every accusation lands and the story agrees
Ending the case	You must produce the evidence, not the name	Saying the name is enough

The third row is the sharpest. A story that cannot let you be wrong has nothing hidden, and a story with nothing hidden cannot reward the kind of slow, re-reading attention that interactive fiction is best at. One failed accusation early is worth more than an hour of atmosphere.

Keeping your own case file

We build Fabledrift, an interactive fiction app for Android, and we will not tell you it holds a sealed envelope with a name in it. It does not. Every scene is written live by an AI narrator that is handed the story’s structured memory — facts, characters, places, promises — on every turn, and what that record is good for is consistency after the fact rather than a solution planted before it.

That still gives an investigation something to stand on. You can write your own notes into the record, and when a note you wrote contradicts the story’s version, your note wins. You can ask the narrator a question without advancing the story, which is how an interview should work. Stepping back to your own earlier scene is free, so re-reading a room costs you nothing but time.

If you want a case-shaped premise, Ironreach Station opens with a station that should have a crew of four hundred and does not. Read it the way mystery interactive fiction rewards being read: write down what you saw and when, keep your own timeline, and treat the story as something you are accountable to rather than something that will agree with you. The mystery you get is the one you insist on.

Sources: the published ZIL source for Infocom’s Deadline, the Inform 7 repository, and ink’s language documentation.