WHAT IS THIS?!

Step right up. Make a terrible choice.

Welcome to my booth. Five dilemmas. Two doors on each card. Both doors have teeth. The pigeons have the keys.

Pick A or B and I slide the next card under the glass. Survive all five and I will read your choices with the confidence of a carnival barker who has never once checked his credentials. You will also see how other browsers chose. Send the same set to a friend and compare your wreckage after they finish.

Who built these traps?

One AI writes all five in a fresh context. It gets no previous sets, no information about you, and no vote totals. Luna and Haiku take turns, switching at midnight UTC. If the model circles back to dragons, doors, or a suspicious number of pigeons, we keep the footprints.

After the five cards are written, a second call to that same model chooses colors, gives each option a broad label, and supplies a few deranged lines for your final reading. It cannot rewrite the choices. Then, once per set, it gives all 32 possible five-choice paths a ridiculous title and proclamation. Your path picks yours. No extra AI call for each visitor.

Curious how this contraption behaves when nobody is looking? Enter the evidence tent ↗