Gerald Ajam

Completion is not competence

Completion is the first figure on most reports. It is easy to count, easy to raise and easy to read as success. It records that a learner reached the last screen. Whether they can now do what the last screen was for is a separate question, and it has to be asked with the help switched off.

Status Position

Published November 2026

All notes

In short

Completion shows that a learner reached the end with the help still on the screen, so competence has to be checked separately, with the help switched off.

The shift it turns on

The usual reading

Completion as the outcome

They finished, so they learned it.

The reading it moves to

Completion as exposure

They reached the end. Now see what they can do alone.

01The situation

The figure that goes on the slide

When a learning product is reviewed, completion is usually the first number on the slide. Modules finished, lessons completed, assignments submitted. It has every quality a reporting figure needs. The software collects it, it rises when people use the product and it fits on one line.

It is also the easiest figure to raise without anyone learning more. A reminder raises it. So does a badge, or a shorter lesson. A button marked “done” raises it fastest of all. Once completion is the target, every one of those changes looks like progress.

Coe (2013) listed coverage of the curriculum among his poor proxies for learning. Completion is coverage, counted by software.

02The tension

Research suggests

What did they finish it with?

What if the trouble lies less in the number than in the conditions it was collected under? A learner finishes a module with the hint button, the worked example, the retry that waits until the answer is right and sometimes a chatbot in the next tab. Completion records that the learner and all of that support got to the end together. It cannot say how much of the work was the learner’s.

The evidence on practice says to expect a gap. Bastani and colleagues (2025) gave high school students a standard chatbot to practise maths with. They scored higher in practice and lower in the exam that followed, once the chatbot was gone. A version set up to give hints instead of answers largely avoided the harm. It did not raise exam scores either.

So the more help a product gives, the less its completion figure says about the learner. Better help can widen the gap, because it makes finishing easier for everyone, including the learners who have not yet learned the thing.

Completion records that the learner and all of that support got to the end together.

03The move

Check competence where the help stops

  1. Call it what it is

    Report completion as completion. “Reached the end” is an accurate label, and a useful one. It shows who got through, where people stopped and which modules nobody finishes. Labelled as learning, the same figure carries a claim it cannot support.

  2. Take the support away at the end

    Put a short unaided check after the last screen. Two or three items, with no hints, no retries and no worked example. Make them a little unlike the practice items, so they test the idea and the memory of a particular item does not help. This is the cheapest evidence of competence there is, and many products never collect it.

  3. Look again later

    Give a sample of learners one or two items of the same kind a few weeks on. Completion is measured at the moment of finishing, when performance is at its highest and says least about what will last.

  4. Report the gap

    Set completion and the unaided result side by side, and never merge them into one score. The gap between them is the most informative figure in the report. A product with high completion and a wide gap is teaching people to finish.

04Where it breaks

Sometimes finishing is the point

Completion is the right measure where reaching the end is the goal: reading a safety notice, acknowledging a policy, watching a briefing everyone must have seen. Holding those to a competence test adds cost and little else.

And completion is a precondition everywhere else. A learner who stops halfway cannot show competence on the half they never reached. Low completion is worth acting on, even though high completion proves little.

05Where it lands

Two numbers where there was one

The change is small to describe. Where a report had one figure, it now has two: how many finished, and how many of those can do the thing unaided. The second number is lower. How much lower is the finding.

It also changes what a design review argues about. A change that raises completion and leaves the unaided result flat has made the product easier to finish. That may be worth having. It should not be reported as better learning.

And it puts one question to every new form of help, chatbots included: what happens to the unaided result when this is switched on? Until someone measures that, nobody knows whether the help is helping.

The point

What transfers

The reason this page exists, rather than the story that produced it.

  1. Completion records that a learner reached the end with the support in place. It says little about what they can do once the support is gone.
  2. The more help a product gives, the less its completion figure says about the learner.
  3. Put a short unaided check after the last screen, a little unlike the practice, and look again a few weeks later.
  4. Report completion and competence side by side. The gap between them is the finding.
  5. Where finishing is the goal, measure finishing. Everywhere else, treat it as a precondition.
Singapore