The question at the end

Listening to a whole lecture feels like having understood it, and the only thing that separates understanding it from having heard it is trying to recall it without looking. So when the slide ends, before moving on, the teacher asks one thing.

It is not an exam and it does not keep score. You can turn it off in settings, and if the question is not generated yet the slide ends the way it would without them: you are never made to wait for one.

Four ways, and they take turns

Choosing among four options is recognition; filling the gap is producing the word; answering the open one is saying the whole thing; explaining it out loud is giving it back. Only one of the four, across forty slides, would be the same exercise every time.

• Multiple choice. Four options, and the wrong ones are the topic's typical confusions, not absurdities you rule out at a glance.
• Fill the gap. A word is missing from a sentence and you type it. Typing it is quite a bit more than recognizing it among four.
• The open question. Answered from memory and in words. Nobody marks it: you see the answer and say whether you had it.
• Explaining it out loud. Every fourth slide, and only if you downloaded the model that listens.

The other three are dealt out by page number, and the spoken one falls every fourth slide precisely so it does not always eat the same kind of slide: every twelve slides, each of the four ways comes up three times.

Telling it in your own words

You talk for thirty seconds and it tells you which ideas you covered and which you did not: "you explained 3 of 5 ideas", plus a couple of sentences on what you left out. No mark and no score: the only thing worth knowing is what you said nothing about.

What you say is transcribed on your Mac, with the model you downloaded for it, and compared against the script's ideas, not against the slide's words. Explaining well means not repeating: if the idea is about coupling and you say "how much one module depends on another", that idea is explained. Counting matching words would have called it missed.

Naming is not explaining: saying the term without saying what it means does not count. And when in doubt it does not count in your favour, because having one extra idea to review is no big deal and being told you explained something you did not is.

A question that fails the check is not shown

The question is asked about what was explained and the answer has to be on the slide. Both conditions are in the request to the model, but neither is taken for granted: what comes back is reviewed in code and a question that fails the check is discarded silently.

With the answer in front of you, answering is reading

While the question is on screen the script goes away and the slide blurs. Both come back when you see the answer, and the slide comes back with the frame placed where that answer lives, which is what turns "I didn't have it" into "it was right here".

What you got wrong gets noted

Getting the question wrong marks that part of the slide as not understood, in the same list as the ones you mark by hand from the script. There are not two lists -"what I marked" and "what I got wrong"- because when you come back you are not looking by how it got marked, you are looking for what is still not understood.

From the document's home screen you jump back to the first marked slide.