Why we publish our prompts
The hardest question about an AI-guided party is who controls the question. Here is our answer.
Ask an AI model a question two different ways and you can get two different answers. Critics of an AI-guided party will point this out, and they will be right to.
If a leader can privately reword a prompt until the machine agrees, then following AI means nothing. It is just a costume for doing what you already wanted.
So our protocol starts with the question, posted in public before any answer exists. The prompts come next, published word for word with the model version and the full answer, including when an answer is inconvenient.
That does not make the process perfect, but it does make it checkable. Anyone can run the same prompt, try a fairer one and show the difference. We count that as help.