The University of Edinburgh teaches in English and French because the continent does. This study tests whether a dual-language rubric, rather than a translated exam paper, reduces construct-irrelevant variance for students in Dakar, Yaoundé, Kigali and Nairobi. Two hundred and sixty master’s candidates completed the same operations case in their stronger working language. Markers used a bilingual rubric with shared performance descriptors. Inter-rater reliability rose from 0.61 to 0.79 after marker calibration in both languages. Students whose home language was neither English nor French still under-performed unless they received a glossary of technical terms before the task. We recommend that AFU publish glossaries with every high-stakes assessment, run marker training in both official languages of the university, and never treat a machine translation of a case as equivalent to a case written for that language community. Quality in a continental digital university is a language question before it is a platform question.