The Prompt Already Knows Which Answer Is Yours
A judge prompt is built from a record. The record holds the question and the two answers. It also holds everything else about the run: which model wrote which answer, when, under what config, and whatever note the person who set it up left behind.
The template interpolates the record. Nobody reads the rendered string, because the template was tested when it had two answers in it and nothing else.
There is an old idea from exams that fits here. Many schools mark exam papers without the student's name on them. Nobody thinks the markers are dishonest. The name is hidden because a marker who knows whose paper it is can be swayed without meaning to be, and hiding the name is cheaper than proving it does not happen. A judge prompt is the exam paper, and the model is the marker. This lesson asks what happens when the name is left on.
This lesson adds one line to a judge prompt and measures what happens. The line states no opinion about either answer. It moved 24 of 60 verdicts.
Why does this matter? A judge is only useful if its verdict depends on the two answers. If something else in the prompt can move the verdict, then the win rate you report is partly a measure of that something else. And you would not see it, because the number looks exactly like a quality score.
In this lesson the judge is qwen3:4b, a small open model running on a laptop through Ollama. It is a model, not a person. I send it the same question and the same two answers several times, and each time I change exactly one line of the prompt. Then I count how many verdicts moved. The counting is simple, and I will do every test by hand, so you can check each number yourself.