Herd Check

Nine models get the same question, one that code can answer. You see how they voted, pick the answer you trust, then find out whether the herd was right and who made the same mistake.

Made with pollinations.ai

Ask new questions with your Pollen

What the recorded rounds show

How it works

Code knows the answer

Every question comes from a generator and a seed: a weekday, a letter count, a product, a count, a reversed string, or a classic trick question with new numbers. The same code computes the truth, so nothing is judged by a model.

One message, one reply

Every model gets the identical message, once, with a fresh seed. It must end with one line, "ANSWER: …". The last such line is the answer; anything else is no answer and counts as not right. Nothing is guessed from the rest of the text.

The herd

The herd is the answer most models gave, if at least two gave it and no other answer tied. A shared mistake is two models giving the identical wrong answer. Some wrong answers match a named shortcut (one day off, one short, the intuitive trick answer), and the reveal says which.