You can pass the quiz but you can't explain it out loud
There's a specific gap between recognising the right answer and being able to produce the idea. It has a name, it's measurable, and most study tools are blind to it.
You scored well. Then someone asked you to explain it — a friend, a supervisor, an interviewer — and what came out was fragments. Words in roughly the right area, joined by "so basically" and a long pause.
If you have felt that, you have run into something real, and it is not a confidence problem. It is a gap between two different abilities that assessment almost never separates.
Recognition is not production
A multiple-choice question hands you the answer and asks you to pick it out of a line-up. That is recognition, and it runs on partial knowledge. Three options look wrong, so the fourth wins — and you never had to hold the whole idea in your head at once.
Explaining it out loud gives you nothing to point at. You have to produce the structure yourself: what causes what, in what order, and why the obvious alternative is wrong. That requires the model to be assembled, not merely recognisable.
So it is entirely coherent to be at 85% on the quiz and unable to say a sentence about it. The quiz was never measuring the thing you just failed at.
Two failures that look identical from outside
We track this as a divergence between two signals, and it splits into two distinct problems with opposite fixes.
Silent knowledge — your scores are strong, your spoken explanations are thin. The knowledge is genuinely in there; it has never been forced into a form you can produce. This one is good news, and the fix is uncomfortable rather than difficult: explain it out loud until it stops collapsing. It usually takes two or three attempts.
Verbal overconfidence — you speak fluently and confidently, and your scores say otherwise. This is the dangerous one, because fluency feels exactly like understanding from the inside. You have the vocabulary and the cadence of someone who knows the material, which suppresses the signal that would otherwise make you go back and check. People in this state are frequently the last to find out.
Same surface, opposite remedy. Telling them apart requires measuring both channels, and nearly nothing measures the spoken one.
What an explanation reveals
When you talk a concept through with Amamihe, it pays attention to the shape of the explanation, not just whether the words are right.
Two things in particular give the game away. The first is whether you use the field's real terminology or gesture around it. The second is whether the causal joints are there — because, therefore, which means. You cannot say "which means" convincingly unless you have a model of the mechanism, and vague language is where a missing model hides. "It kind of affects the thing" is what an absent step sounds like.
Worth being straightforward about the limits: this reads structure and specificity, not correctness. It cannot tell whether you are right — that comes from the quiz side. The value is in the comparison between the two, never in either on its own.
Both are tracked per subject, because the divergence is subject-specific. Confident and correct in mechanics, fluent nonsense in thermodynamics. Only a per-subject view catches that.
Why a chatbot cannot tell you this
Ask a general assistant to quiz you and it will. What it cannot do is notice that your spoken explanations of one topic have been steadily weaker than your scores for six weeks. That is not a reasoning limitation — it is a memory one. It has no record of your last hundred answers, so there is no baseline to diverge from. Every session starts from zero, which means every session can only assess you against itself.
The two-minute test
You do not need us for the diagnostic. Pick something you believe you know. Set a timer for two minutes, and explain it out loud — actually out loud, to a wall — without notes.
Where you stall is the answer. Not the parts you got wrong; the parts where you slowed down, hedged, or said "basically." Those are the joints where your model is missing a piece, and they are almost never the parts you would have predicted.
Then go and look at exactly those. That is the highest-return twenty minutes available to you this week.
Stuck at 11pm? Your lecturer isn't awake.
Amamihe teaches the thing you're stuck on, out loud, using something you already care about. One minute, no account needed.
