Evaluation Bias and Measurement Reliability

Covers systematic judging biases such as output-length preference and other methodological caveats affecting evaluation validity. Scope is measurement bias and reliability limits, not model capability; boundary excludes absolute scores, rankings, latency, and cost.