A remote hotel · 40 people · one storm

Nobody wrote this murder.

Forty people arrive at a mountain hotel with their own histories, debts and secrets. They eat, drink, argue and go to bed. Somewhere in there, one of them kills someone — and not because a designer decided they would.


How most mystery games actually work

A writer picks the killer first, then works backwards, scattering clues that point at the answer they already chose. It reads well once. Play it twice and you can feel the hand that placed everything.

This one runs the other way around. The rule the whole project is built on:

Do not choose a murderer, generate clues, then pretend a simulation produced them. Create the people. Create their relationships. Simulate their behaviour. Let circumstances create the crime. Record what actually happened.


What the simulation is doing

Ordinary life comes first

Guests and staff run on schedules, appetites and moods across roughly three in-world days. Housekeepers clean rooms and sometimes read what’s on the desk. A bartender works until one in the morning. Grudges thicken over breakfast. Nothing is aiming at a murder — most of the simulated time is people simply having a week.

Intent is never announced

When someone finally decides to kill, that decision is internal. It never enters the record. The only trace it leaves is behaviour — where they go, what they pick up, who they wait out.

Witnesses are unreliable honestly

Everyone present perceives an event, then remembers it imperfectly: times drift, strangers go unrecognised, familiar faces get confidently misidentified. Their testimony is built from that memory — not from the truth.

Objects have physics and history

A weapon has fingerprints, blood, an owner and a location. If a killer carries it into the snow and buries it, it is genuinely in the snow. If they wipe it in a hurry, some prints survive and some don’t.

Records only exist where the building has them

Cameras see their own field of view and nothing else. Keycard doors log the rooms that have readers. Walk through the courtyard in the snow and you track melt indoors — which is a clue nobody placed there.


Run one and look at it

Each run generates a population, simulates them, and checks whether a coherent, solvable case emerged. If it didn’t, the entire world is thrown away and another is tried — never patched into shape. You get the case file first; the truth stays behind a spoiler.

Generate a case →