Q1 — Weak contribution. The goal is learning outcomes, not attendance precision. Existing attendance works — this optimizes a non-problem.
Q2 — Heavy cost, shaky tech. Cameras in 60+ rooms + GPU servers; accuracy degrades with masks, lighting, look-alikes. Disputes cost more time than roll calls.
Q3 — No inbound API. The academic ERP can't receive the data — someone reconciles CSV exports manually anyway.