That's one element. This first couple of things I'd check would be: - Double-blind: What if you tell students it is live, but it is actually pre-recorded?

That's one element. This first couple of things I'd check would be: - Double-blind: What if you tell students it is live, but it is actually pre-recorded? - Are the live+archive and archive-only videos the same? - Are the live+archive students actually watching the video when it is live, or are they using the archive feature? - Do archive-only students who consistently watch at the same time do better? - Does time/day of watching affect results? (archive-only students get the link at least [duration of class] later.) You also need to control other interactions with the participants, track the results faithfully, avoid bias because instructors are more familiar with students from "live" interactions, ensure cohort size and result size are large enough to be meaningful, that group composition isn't biased by confounding factors, etc. But random assignment is a good start!