Think twice / 05
The Last Seat
Twelve seats. Twenty regulars. You think tonight will be quiet. What if everyone else does too?
How to play
You are one of twenty regulars deciding whether to visit a twelve-seat music venue. You can forecast the crowd, share advice or offer reservations. Your challenge is to enjoy the evening while understanding what your message does to everyone else's choices. Complete a twelve-night season, then compare a different approach from the same start.
- Look at recent attendance, enter your estimate of tonight's total crowd, and choose Go tonight or Stay home. Start with your forecast private so you have a comparison.
- After the reveal, tap a guest to inspect their forecast and decision. Read the four message alternatives for this one night, then press Plan next night.
- Try a shared forecast, two group forecasts or reservations. After twelve nights, compare enjoyment, crowded nights, empty seats and forecast error. Use Replay this season to test another approach with the same starting conditions; look for an improvement and name its cost.
What to notice
A quiet-night forecast sounds useful. But if everyone hears it and decides to go, the venue fills up. Predict a crowd and people may stay home instead. The forecast changes the very evening it describes.
You are one of twenty regulars hoping for one of twelve seats. Look at recent attendance, make your forecast and choose Go or Stay. After the night is revealed, inspect why individual guests acted as they did. Try sharing a forecast, then try reservations. A reservation can allocate a seat; a prediction only suggests what might happen.
Brian Arthur's El Farol Bar Problem explored this loop in 1994. His fictional guests tried different ways to forecast attendance and learned from the result. If everyone followed one apparently good rule, their agreement could make that rule fail. Advice about traffic or a popular event can create the same kind of feedback.
Our messages, reservations and no-shows extend Arthur's idea. A quieter night, fuller venue and better personal outcome are not always the same result. The alternatives table compares one night from the same starting point; a different season would have to unfold night by night. Are you improving coordination, or changing the choices you hoped to predict?
Read more: Brian Arthur, Inductive Reasoning and Bounded Rationality.
Recent attendance
Night 1 / 12The doors opened
What if you had used another message?
Same night, guest habits, availability, message reach, forecasts and your Go/Stay choice. Only the message mode changes. Later nights would learn from different outcomes.
| Message | Arrivals | Empty seats | Standing | Your enjoyment |
|---|
Tonight at the little venue
Tonight’s crowd is unknown. Your choice comes first.
Explore the nineteen guests’ reasoning
The other nineteen regulars
Each has three forecasting habits. After a night, inspect which one they used and whether your advice changed their choice.
Your season so far
| Night | Message | Your choice | Forecast / actual | Enjoyment |
|---|
Rules & model
What makes a good night?
There are nineteen simulated guests plus you, and twelve seats. Up to twelve arrivals is comfortable; more means some guests stand. Nobody is denied entry. You earn +1 enjoyment for going to a comfortable night, −1 for going to a crowded one, and zero for staying home. Your forecast’s accuracy is shown separately as how many people it missed by.
How the guests decide
Each guest carries three simple forecasting recipes. They use whichever has recently been closest to the real crowd, averaging its last four misses. Ties keep the earlier recipe in their portfolio. All their private recipes are checked after the night, including when the guest followed your advice or could not attend.
See the seven forecasting recipes
- Use last night’s count.
- Average the last three nights.
- Continue the recent trend: twice the latest count minus the previous count.
- Expect the opposite: twenty minus last night’s count.
- Use the count from two nights ago.
- Always predict ten.
- Average the last six nights.
Predictions stay between zero and twenty. The first six attendance counts are explicitly fictional starting history, not earlier rounds secretly played for you.
An available guest goes if their chosen estimate is twelve or less. A message recipient uses your shared or group number instead. Guests do not anticipate how everyone else will respond to the same advice.
Why reservations may still leave gaps
Each guest has a 15% chance of being unable to come. Message reach is a chance per guest, not an exact quota. In reservation mode, Go claims a place for you; remaining invitations rotate among guests without knowing who will be unavailable. Message followers come only if invited. Non-followers still choose privately, so they can create a crowd. No-shows are not replaced by a standby list.
What the comparison can tell you
The alternatives reuse this night’s starting guest habits, forecasts, availability and message draws. Only the message mode changes. Later nights would learn from their different outcomes, so this table does not predict an alternative whole season.
Replaying restores guest learning, starting history and the random seed. There is no live audience or saved profile. These are fictional decisions that make feedback visible, not a forecast of real people or a judgment about your personality.
Where this began
W. Brian Arthur’s El Farol problem, introduced in 1994, explores attendance choices based on forecasts of crowding. Our twenty regulars, three-rule portfolios, four-night scoring window, no-shows, public messages, reservations and personal enjoyment points are an adaptation. These simulated choices are not a prediction of real people’s behaviour.