What comprehension data tells you that completion cannot
Completion says they clicked. Everything you actually want to know is somewhere else.
Part of a single argument: the engagement gap is a workload problem.
Completion rate is the number every system gives you for free, which is most of why it is the number every organisation reports. It answers exactly one question: did the learner reach the end.
It is close to uncorrelated with whether they can now do anything differently, and reporting it as though it were a measure of training effectiveness quietly misleads everyone including the person reporting it.
What completion cannot see
It cannot distinguish the learner who read carefully from the one who opened every page in a new tab on Friday afternoon. Both completed.
It cannot tell you which lesson failed, because it aggregates to the course. A course at ninety per cent completion may contain one module that nobody understood, and the completion figure will look healthy throughout.
It cannot tell you whether the material was already known, which means it cannot tell you that you spent three weeks teaching a cohort something they arrived knowing.
Four numbers that do more
Comprehension by topic. Correct-answer rate broken down per lesson rather than per course. This is where the actionable information lives: one lesson everybody fails is a content defect you can fix this week.
Drop-off position. Not how many finished, but where the ones who stopped, stopped. A cluster at a single point is a local content problem. An even spread is a length or motivation problem, and they need opposite responses.
Prior knowledge. What the cohort knew before you started. Without it you cannot distinguish a course that taught well from a course that was unnecessary.
Re-look rate. How often people return to a lesson after finishing it. A little is healthy and means the material works as reference. A lot, concentrated in one place, means that lesson did not do its job the first time.
Read them together or not at all
Any of these can mislead alone. Comprehension without drop-off tells you the survivors understood, which is not the same as the course working, and can look excellent for a course that lost half its cohort in the first module.
Report them as a set, for the same cohort, on the same page. Four numbers is few enough that people will actually look, and it is the combination rather than any single figure that identifies what to change.
The reason this belongs in a blog about workload
Because comprehension data is what makes improvement cheap.
Without it, improving a course means reviewing the whole thing and guessing, which is expensive enough that it does not happen. With it, improvement means rewriting the one lesson that everybody failed, which is an afternoon.
That is the difference between a course that gets better each time it runs and one that is rebuilt from scratch every two years because it has drifted too far to fix.
If you change one thing
Add one comprehension check to each lesson and report the results per lesson rather than per course.
It is a small piece of authoring and it converts your training from something you hope worked into something you can debug.