Inside an Evidence Check
Every mastery node ends in an Evidence Check. It is not a test with a mark; it is a diagnosis. Here is exactly what it captures, how the diagnosis is made, and what happens next.
Not a score. A reading of how you think.
A single correct box tells you very little. A learner can be right by luck, right on shaky ground, or wrong for five different reasons. Maestro reads four signals together on every check, so it can tell a deep answer from a shallow one and a real grasp from a good guess.
The check is deliberately not graded and gives no correctness feedback while the learner answers. That protects the diagnosis — later answers are not contaminated by feedback on earlier ones — and it removes the incentive to fake: there is no mark to win, only support to route.
The four signals
Response
What the learner actually answered — the selection or decision itself.
Reasoning
Why they chose it, in their own words. A required field, not an optional one — this is what an outside tool cannot supply.
Confidence
How sure they were. Confidence calibrated to correctness is a strong sign of real understanding; confident wrongness signals a misconception.
Process
How they got there — response latency against their own baseline, focus changes, paste-versus-typed input, revision count. Weight context only, never accusation.
Five ways to be “wrong” — each handled differently
The same mistake can mean five different things, and each needs a different response. Lumping them together is what makes ordinary quizzes useless for learning.
- Secure mastery. Understands it, and knows they do. Ready — proceed, or take an optional challenge.
- Slip. Knows it; a careless mistake. Light feedback, move on.
- Fragile. Right answer, weak ground under it. Reinforce and re-check.
- Knowledge gap. The concept has not formed yet. Route back to the missing prerequisite.
- Misconception. A confident, specific wrong model. Trigger content that confronts that exact wrong model — not a repeat of the lesson.
Restraint before it acts
The bigger the intervention, the higher the bar. A clear result is acted on immediately, but a costly reroute — say, on a suspected misconception — earns one confirming probe first: the same idea in a slightly different form. If the error returns, it is real; if it vanishes, it was a slip. Maestro stays conservative under uncertainty.
What the learner sees is a coach's readout, never a score: what is strong, what is missing, what to add next. Underneath is a real diagnosis; on the surface it reads as coaching, which keeps a hard moment from becoming a discouraging one.
The mechanics under the hood
Technically, each check is an instance of an evidence-capable template. The instance carries a diagnostic band map — a deterministic, validated mapping from every possible signal combination to exactly one band: secure, fragile, knowledge gap, misconception, or not-sure. Renderers never interpret signals themselves; they capture and emit, and the band map decides.
“I'm not sure” is a first-class band of its own, always available and penalty-free. It routes like a knowledge gap but is recorded distinctly, because honest uncertainty is a calibration signal the learner model values.
The design law
An Evidence Check is never a simple right/wrong quiz. Response, reasoning, and confidence — always. No correctness feedback in-flow. And its outcome updates a belief; it never directly produces a grade.