Repo under interrogation: Heban-7/Data-Contract-Enforcer Files in scope: report_final_pdf_ready.md , contracts/ai_extensions.py The question, anchored You have two questions stacked on top of each other in the same artifact: (a) effective sample size : the report says Sample size: 251 but the implementation caps embeddings at 200 . Which n is the statistic actually computed over, and what does the discrepancy cost you? (b) evidence chain : when drift_score = 0.0 from a centroid-based cosine method, what additional evidence do you need before writing "Text content is semantically stable" in the report? Both have a single answer-shape: a centroid is a first-moment summary, and a first-moment summary is silent on everything else — sample size, dispersion, multi-modality, model identity, and fallback behavior.…