It is one of the five fields of a simulation, and it does more work than its length suggests. The scenario is what makes a twin reason as someone in a moment rather than as someone answering a survey.
Concrete beats abstract
The most common weak scenario is a topic rather than a situation.
Thinking about tipping
That gives a twin nothing to be in. It will produce an opinion, and opinions are weakly related to behaviour.
You have just placed a food order on DoorDash and you are at the checkout screen
Now there is a moment, a place, and something in front of them. Twins reason about their own situation rather than about the subject in general, and the answers separate.
The test: could you take a photograph of it? If not, it is a topic.
What belongs in it
Where they are and what just happened. Enough for a person to know what moment this is.
What is in front of them, if that matters. The screen, the shelf, the email.
The immediate specifics of this instance. The order was forty minutes late. The pack is the new smaller one.
What does not belong
General facts about the world. Prices, norms, what is available, what is usual — those belong in the environment, where every twin gets them consistently and each one carries a source. Putting them in the scenario means retyping them for every simulation and having no record of where they came from.
The options. They belong in the action space.
The question. It belongs in the question.
Anything that argues for an outcome. "You have just placed an order and the fees already feel high" is not a scenario, it is a nudge. It will move the result, and it will move it in a way that looks like a finding. Keep the setup neutral and let the twins bring their own reaction to it.
Length
Two or three sentences is usually right. A scenario long enough to include everything is long enough to bury the thing that matters, and the details that belong in a long scenario usually belong in the environment instead.
When you are testing a change
If the point of the simulation is a change — a price cut, a new pack, a rival doing something — put the change in the scenario and hold everything else constant, then run the same thing without it. That pair is a comparison, and it is the only way to attribute the difference to the change rather than to everything that moved at once.