Bolt Series · Human Performance Calibration · Article 25
There may be no such thing as simply “a well-calibrated person”
A student can know exactly how good they are at algebra.
Ask them to predict their score on a set of algebra questions and they are usually close.
They know which answers are secure.
They know which errors they tend to make.
They know when they need help.
Then give the same student an essay.
They think the argument is excellent.
The teacher finds the reasoning thin and the evidence poorly connected.
Same learner.
Very different calibration.
This matters because we often describe confidence and self-awareness as if they were global personality traits.
“She knows herself well.”
“He is overconfident.”
“She lacks confidence.”
Those descriptions may be too coarse.
Metacognition contains both shared and domain-specific components
Recent research gives us a more careful picture.
A 2025 study examined how people updated beliefs about their own ability across memory, visual and general-knowledge tasks. Confidence showed some broad commonality, but the way people updated their self-beliefs after new evidence was strongly domain-specific.
In other words, new evidence in one domain did not simply rewrite one global confidence setting.
People adjusted beliefs according to performance in the particular domain.
A 2026 study across multiple decision tasks found only weak evidence for a completely domain-general metacognitive ability, with stronger clustering within some kinds of tasks.
The research is still developing, and we should not turn either result into a universal law.
But it gives education an important safeguard:
Do not assume that accurate self-judgement in one skill automatically transfers to every other skill.
Why would calibration differ across skills?
Because different tasks expose different cues.
In Mathematics, a learner may have relatively clear receipts.
The answer is correct or incorrect.
A method works or it does not.
Repeated question types make patterns visible.
In writing, the evidence can be less immediate.
A sentence can be grammatically correct but rhetorically weak.
An argument can feel convincing to the writer because the intended meaning is already inside their head.
Judgement depends on criteria that may take longer to internalise.
Different domains therefore offer different feedback environments.
And calibration improves only when the learner has access to useful cues and informative receipts.
Even within one subject, calibration can split apart
A student can be well calibrated about routine algebra and badly calibrated about unfamiliar problem solving.
They know exactly when a procedural answer is safe.
But when a problem requires selecting the method, they mistake familiarity with the topic for understanding the structure of the problem.
Or the reverse can happen.
A student may understand concepts deeply but systematically underestimate their speed under timed conditions.
This gives us a much more useful picture than:
“This student is confident.”
We can ask instead:
- Calibrated about which skill?
- Under which conditions?
- At what level of difficulty?
- With what kind of evidence?
- Across how many attempts?
Sport gives us the same warning
An elite athlete can be exquisitely calibrated about one part of performance and less accurate about another.
A sprinter may know their maximum-velocity state extremely well but rely on a coach or video for technical details in the start.
A runner may judge effort accurately during familiar training but misjudge readiness after injury or illness because the usual cues have changed.
Expertise is therefore often local.
So is the self-knowledge built from expertise.
This is one reason Bolt repeatedly returned to a specific weakness rather than describing himself globally as simply fast or slow.
The start.
The first 30 to 40 metres.
Acceleration.
Those are higher-resolution objects than “my sprinting.”
Global labels can hide useful differences
A child says:
“I’m bad at Science.”
That sentence contains almost no diagnostic information.
Perhaps the learner recalls facts very well but struggles to interpret experiments.
Perhaps they understand mechanisms but write vague explanations.
Perhaps they perform well in untimed discussion and poorly under examination load.
Perhaps their self-judgement is accurate in one component and badly wrong in another.
The more global the label, the more likely we are to lose the mechanism.
Calibration should therefore be built as a profile
Instead of asking:
“How confident are you in Mathematics?”
ask questions with resolution.
- Can you retrieve the core method without notes?
- Can you recognise when that method applies?
- Can you execute it accurately?
- Can you transfer it to an unfamiliar form?
- Can you do it under time pressure?
- Can you tell when your answer is probably wrong?
Then compare each prediction with performance.
The learner may discover:
“I judge my procedural accuracy well, but I systematically overestimate transfer.”
That is a much more useful self-model than “I’m good at Maths.”
Do not let one calibrated domain create false authority elsewhere
There is a broader human lesson here too.
Success in one domain can make people feel generally expert.
But the receipts that calibrated us in one world may not exist in another.
A brilliant athlete does not automatically become a brilliant investor.
A strong teacher does not automatically become an expert clinician.
A student who knows their Mathematics precisely may still misjudge their writing.
Calibration must remain tied to the domain in which it has actually been tested.
Why this matters for education
Education should not aim to create children who simply “feel confident.”
It should help them build increasingly accurate, skill-specific maps.
Where am I strong?
Where am I uncertain?
Which estimate has repeatedly matched performance?
Which part of my self-model still needs better evidence?
You can know yourself accurately in one part of the map and still be wrong in another. Good calibration respects the boundaries of the evidence that built it.
That makes self-knowledge less flattering.
It also makes it much more useful.
