LLM-as-a-judge: how to evaluate AI without fooling yourself

LLM-as-a-judge from first principles — when to use it, how to design rubrics, the three biases that skew scores, and when to use something simpler.