This repository is a static, read-only survey of academic papers on data contamination in large language model (LLM) benchmarks. It contains no executable code, no datasets, and no pipelines. The only substantive content is a README.md that lists references to static and dynamic benchmarking methods, categorized by domain (math, knowledge, coding, etc.) and mitigation approach (canary strings, encryption, post-hoc detection). The repository includes three images used for illustration. There are no scripts, notebooks, configuration files, or environment definitions. The last commit appears to be dated in 2026 (likely a metadata error), but the content shows no evidence of active development. This repo serves as a centralized bibliography for researchers interested in the transition from static to dynamic LLM evaluation; it has no practical reuse value for data engineering or software development tasks.