We study a systematic failure mode in language models: when the true answer to a STEM question is surprising relative to training-data priors, models prefer plausible-sounding distractors over the correct answer. We build a 97-fact STEM benchmark spanning six domains (calculus, physics, chemistry, statistics, linear algebra, constants) and evaluate six models from GPT-2 (117M) to Qwen3-4B using log-probability multiple-choice ranking. Accuracy rises from 16% to 77% with scale, but systematic errors persist even at 4B parameters. We identify four scale-invariant bias patterns (positivity, linearity, missing-constant, truncation) that appear at all scales. A transfer matrix experiment shows zero cross-pattern generalization from single-pattern adapters; mixed training achieves 70-100% per-pattern accuracy. Log-probability margin is a perfect binary oracle: positive margin predicts correct answer with 100% precision and recall on the 40-fact probe set. Margin magnitude tracks domain difficulty. v1.1 changes: Expanded limitations section, replaced informal self-references with DOI citations, strengthened abstract opening, added GitHub link.

Part of the rho-eval / knowledge-fidelity research program. Paper 9 of 9. Code: https://github.com/SolomonB14D3/knowledge-fidelity