Who Gets Flagged More
Flag rates are not evenly distributed, and the pattern is consistent enough across institutions to be predictable before deployment.
Analysis
Who gets flagged
Flag rates differ by group for reasons unrelated to conduct, and the pattern tracks disadvantage that already exists.
7 notes
Disability and Neurodivergence
Proctoring models a narrow range of bodies and behaviour as normal. Students outside it are flagged for existing, and the obligations here are legal as well as ethical.
Analysis
Skin Tone and Face Detection
Face detection and matching perform measurably worse on darker skin. In proctoring this produces more failed check-ins, more lost-face flags and more accusations.
Analysis
Home Conditions and Bandwidth
Proctoring assumes a quiet private room and a stable connection. Students who have neither are flagged for their circumstances.
Analysis
Caring Responsibilities and Interruption
Students with children or dependants cannot guarantee three uninterrupted hours. The system treats each interruption as an anomaly and each anomaly as a question.
Analysis
Religious Dress and Head Covering
Detection and identity matching perform worse with head coverings, and removal requests create a specific and avoidable harm.
Analysis
What to Do About the Disparity
Measuring the gap is the easy part. Four responses that actually reduce it, and two that only appear to.
Procedure