Estimate before revealing
A bag contains 3 coral tiles and 2 blue tiles. One tile is drawn at random. What is the probability that it is coral?
Forecast Calibration Game for Probability Judgment
Forecast calibration improves when you state a probability before seeing the answer, measure the distance from the known result, and examine whether the error came from arithmetic or confidence.
Forecast 1 of 8No answers yet
Within 5 pointsWell calibrated for this problem
Within 15 pointsUseful direction with room to improve
More than 15 pointsReview the outcome space before adding confidence
What calibration means in this game
Calibration is the match between stated confidence and what the evidence supports. These rounds use transparent probability problems with exact or clearly rounded answers, so you can measure the distance between your estimate and the known result.
The game reports average absolute error in percentage points. An estimate of 70 percent for an answer of 60 percent is 10 points away. This is a learning measure for the session, not a claim about your forecasting ability in every domain.
Separate arithmetic mistakes from confidence mistakes
An arithmetic mistake means the outcome space or calculation was wrong. A confidence mistake means the reasoning was uncertain but the estimate was pushed toward an extreme anyway. The feedback explains the exact outcome count so you can distinguish the two.
In real decisions, exact probabilities are rare. The same discipline still helps: define the event, choose a relevant reference class, state a range, and avoid precision that the evidence cannot support.
Use reference classes for real forecasts
A reference class is a group of similar past cases. Instead of asking whether this project feels likely to succeed, define success and examine comparable projects with similar scope, team, dependencies, and deadlines.
The comparison does not need to be perfect. Its purpose is to provide a starting rate before details about the current case move the estimate. Without that anchor, confidence can be driven by enthusiasm, fear, or the persuasiveness of one story.
A practical forecasting record
- Event: state exactly what will or will not happen.
- Deadline: name when the event can be judged.
- Probability: record one estimate or an honest range before the outcome.
- Evidence: list the reference class and the strongest case-specific adjustment.
- Review: compare the outcome with the original record without rewriting the prediction.
Review the process as well as the outcome
A good forecast can miss and a weak forecast can get lucky. Check whether the event was clear, the reference class was relevant, and the adjustment was recorded before judging the method from one result.
Continue with expected value for two options when outcomes have defensible probabilities, or use the decision journal and outcome review to learn from forecasts over time.