Who actually annotates NLP datasets, and why won't researchers say?

NLP research depends on human-labeled data, yet papers rarely document who annotated or how. Researchers analyzed 1,603 papers (2018–2025) using LLM-assisted extraction, validated against 41 hand-reviewed papers, and found papers report recruitment and volume but routinely skip annotator expertise, compensation, language proficiency, and agreement scores—especially in model evaluation. The audit establishes a taxonomy and minimum reporting standard to improve annotation transparency and reproducibility.