Calibration trainer

Calibrate your confidence.

Most people are more sure of things than they should be. Suppose measures the gap between how confident you are and how often you're right, in about five minutes a day.

The daily set is free. No ads. No account needed.

Claim 1 of 3

Octopus blood is blue.

chance it's true
50%
Could go either way
FalseTrue

Lock it in to see the answer, the explanation, and your score.

How it works

One claim, one slider, then the answer.

Trivia scores whether you were right. Suppose scores whether your confidence matched how often you're right.

1

Read the claim

A short factual statement, picked so the answer isn't obvious.

2

Say how likely it is

Move the slider anywhere from 1% to 99%. You give a number, not a side.

3

Lock in and find out

You get the answer, a short explanation, and your score right away. Then the next claim.

The scoring

You score best by saying what you believe.

Points come from the Brier score, a proper scoring rule measured against a coin-flip baseline. That means your expected score is highest when the number you give is the number you actually believe. Pushing the slider to 99% and hoping loses points over time.

points = round((0.25 − brier) × 200)   where  brier = (p − outcome)²
99% sure, and rightThe most a single answer can earn.
+50
50% — a coin flipAlways scores zero, so you can't sit a question out.
0
99% sure, and wrongThe worst a single answer can cost you.
−146

The payoff

See where your confidence is off.

If you say 80% on a hundred claims, about 80 of them should turn out true. Your profile plots what you said against what actually happened, next to the line of perfect calibration.

100% 0% 0% 100% you said actually true perfect
Illustrative. Your curve is built entirely from your own answers.

What your profile tracks

All of it computed from your own history, on your device.

  • Lifetime calibration curve against the perfect diagonal
  • How overconfident you are, as a single number
  • Mean Brier score, with a plain-English explainer
  • Accuracy and calibration broken out by category
  • An activity heatmap and your longest streak
  • Estimation range stats: average interval score and hit rate

Ways to train

Two kinds of question, one skill.

True/false claims and numeric estimates train the same thing: knowing how sure you actually are.

Free

Today's set

Ten true/false claims a day, across a mix of categories. Feedback on each one, then a score and a streak at the end. Free, always.

Pro

Rapid fire

The same drill without the daily limit. Pick 10, 20, 50, or unlimited, and pick which categories you get.

Pro

Estimation

Numbers instead of claims. Give a range at the confidence level you choose, then see where the real answer fell. Narrow ranges score better, as long as they contain the answer.

The questions

Every answer is checked.

Each claim has a verified answer and a short explanation of what's true. Difficulty is spread on purpose, so you're not answering easy questions at 95% forever.

  • A roughly even true/false split, so you can't win by always guessing one way
  • A growing bank of numeric estimation questions alongside them
  • Seven categories, with easy and hard questions mixed into every set
  • Free World & geography
  • Free Science & nature
  • Free History
  • Free Sports
  • Arts & culture
  • Technology
  • Business & economics

Found a wrong answer? Tell us.

There's a report button on every question. Corrections are the most useful feedback we get.

Questions

Common questions.

Is this just a trivia app?

It looks like one, but it measures something else. A trivia app scores whether you were right. Suppose scores whether your confidence matched your hit rate, which is the part that shows up in estimates, forecasts, and decisions.

What's a Brier score?

The squared difference between what you said and what happened. Say 0.8 on something that turns out true and your Brier score for that answer is (0.8 − 1)² = 0.04. Lower is better, and because it's a proper scoring rule, the way to lower it over time is to say what you actually believe.

Suppose converts it to points against a coin-flip baseline so higher is better, and shows your raw mean Brier score in your profile.

Does calibration training actually transfer to real decisions?

Partly. The research supports two things: people get better calibrated on the kind of question they practise on, and feedback is what drives the improvement. How far that carries over to messy real-world judgment is less settled, so we won't claim more than that.

Do I need an account?

No. The app works fully signed out, and your history backs up in the background from first launch. Signing in is what lets you reach that history from a new phone. It's with Apple, Google, or a six-digit email code — there are no passwords anywhere.

Where does my data go?

Your answers, confidence ratings, and history live on your device and back up automatically, whether or not you've made an account. We use privacy-respecting analytics to see which features help people improve. We don't sell your data and there are no ads. The full details are in the privacy policy.

How many questions are there, and will they run out?

A checked, growing bank across seven categories, covering both true/false claims and numeric estimates. You'll never see a question you've already answered, and we keep adding to it.

I found a wrong answer.

Please report it. There's a button on every question, or write to us from the support page. Corrections get priority over everything else.

Android?

Planned. The app is built from one codebase that runs on both, but iOS comes first.

Find out how sure you actually are.

Ten claims a day, about five minutes. The daily set is free.

Coming soon to the App Store