It is one of the five fields that define a simulation, and it is the one people fill in without thinking and then get bitten by. Twins reason from the world as it stood on the date you set. Change the date and you have changed the question.
Why it changes the answer
Ask about delivery tipping in 2019 and in 2026 and you get different answers, because the fee structure is different, the norms are different, and the alternatives are different. None of that is the twins being inconsistent. It is them being correct about two different worlds.
Concretely, the date does two things:
It selects the environment. Only facts that were already true at that moment are included. Prices, availability, norms and events are all dated, and anything that had not happened yet is left out.
It sets what a twin can know. A twin reasons as of that moment. It does not reference anything later, and it does not talk about the date as though looking back on it.
Getting it wrong
Nothing announces it. A simulation with a bad date runs normally, returns a clean distribution, and reads exactly like a good one. It is the quietest way to get a confidently wrong result in the entire product, which is why it gets its own page.
The three common versions:
Leaving it as today when you mean the future. You are testing a launch for next spring. Twins asked about today will reason from today's prices and today's competitive set.
Setting a future date and expecting foresight. A date in the future does not mean the twins know what will happen. It means they reason from what is known to be true then — announced prices, scheduled changes, established trends — and from the assumptions in the environment, which are marked as assumptions. A simulation set in the future is a well-grounded estimate of behaviour under stated conditions, not a forecast of events.
Comparing across dates and changing something else too. If you want to know what two years did to an attitude, the date has to be the only thing that moved. See Comparison.
Choosing one
Set it to the moment the real decision happens. If you are deciding a price for a spring launch, the date is the launch, not the meeting where you decide it.
Where the answer might turn on the date, run it twice. Two dates on the same question is the cheapest and most informative comparison available, and the difference is attributable because nothing else changed.