Methods · Metacognition & self-regulation

Calibration training, at the level of a whole assessment

Predict your own grade before you get it, and close the gap, which is what makes every decision about how much work to do a decision rather than a guess.

When to use this: When checking yourself
Learning impact: Strong change
Learning evidence: Strong research tradition
Where you are with the topic: Checking readiness
How often to use it: Worth repeating often

When to use it

You are about to submit an essay, paper, or project and need a written grade prediction before it leaves your hands.
You keep either over-polishing work that was already good enough or submitting work you thought was fine.
You have a few past predictions and results and want the error direction, not just another gut feeling.

What to ask AI

Before submitting: “My prediction: [grade], losing marks mainly on [where]. Don't tell me if that's right. Record it, and ask me two questions that might make me revise the prediction.”

After results: paste your last several predictions versus actuals, including where you expected to lose marks versus where you did.

Ask what should change about how you decide a piece of work is finished, given that pattern.

How it helps

Item-level confidence ratings help revision. Whole-assessment calibration helps planning, and planning failures are the expensive ones.

Every decision about effort is a prediction: this essay needs another day; I've done enough for that exam. If those predictions are systematically wrong, you burn weeks or submit blind.

There is a documented asymmetry: people who are struggling are often the most overconfident, and strong performers often underestimate themselves. If you are struggling and feel fine, that combination is useful information.

Track the total and where you expected to lose marks. Often the total is close and the composition is completely wrong, you knew how good you were, not why.

How to run it

Predict before submitting, a number or band, written down, before it leaves your hands.
Predict the components too: where you expect to lose marks.
Record date, assessment, predicted, and actual in one place you will find again.
After the result, compare overall number and predicted loss points.
Wait for four or five data points before drawing conclusions about direction and size.
Change one planning behaviour: overconfident → add a day and a check; underconfident → stop the polish pass that never moves the mark.

Example exchange

You

I'm about to submit my lab report. I think I'll get about 68%, mostly losing marks on the uncertainty discussion and figure captions.

I'll record that prediction. I won't tell you if it's right. First question only: which rubric row have you not checked against your draft in the last hour?

You

I haven't looked at the limitations row.

Second question: what would a harsh marker say is unfinished?

You

That I asserted precision I didn't justify.

Update your percentage only if those answers change it, then submit. When the mark comes back, we compare the breakdown, not the vibe.

Copyable prompt

I'm about to submit this. My prediction: [grade], losing marks mainly on
[where]. Don't tell me if that's right. Just record it, and ask me two questions
that might make me revise the prediction.

The Tell

If your predictions have been equally wrong for a year, either you are not recording them before you submit, or you are not using them to change anything.

Calibration without a planning change is a scoreboard, not a method.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

Strong research tradition on calibration, predict, then compare, as a core metacognitive skill. Closing the gap between expected and actual performance is what turns effort decisions into decisions rather than guesses.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

AI can coach the recording habit and ask questions that unsettle a rosy prediction. It cannot know your true mark in advance. Treat its prompts as planning aids; the learning comes from your written prediction versus reality over months.