Learning analytics that ignore power and signal will mislabel African students as ‘disengaged’. This study links AFU platform logs with self-reported generator and network windows for 1,104 students in Zimbabwe, Mozambique, South Africa and Botswana. Login frequency was a poor predictor of assignment quality. Submission success tracked whether the portal accepted a file under 8 MB and whether the deadline fell inside a known power window. A simple design change — accepting chapter uploads instead of one large portfolio — raised on-time submission by 23 percentage points in a difference-in-differences estimate. The University of Edinburgh should publish a connectivity standard beside every module guide: maximum file size, offline packet, and a deadline policy that treats a documented outage as a material circumstance. Completion is an engineering and registry problem, not a character judgement.