An admissions predictor that grades itself in public
The first accuracy report publishes after decisions arrive, once 50 admit or deny outcomes are in. None are in yet: most Early Decision and Early Action results arrive in December. Below is exactly how it will be graded.
No grade is published yet, and none will be invented. When 50 admit or deny outcomes are in, this page shows the full calibration table and the Brier score, and it updates every hour after that.
How the grading works
Published in full before the first result, so the method cannot move to fit the outcome.
Each report carries the estimate the student had before the decision arrived. It is not recomputed afterward.
Admitted counts as a yes and denied as a no. Deferred and waitlisted wait until they resolve.
Estimates are grouped 0 to 9%, 10 to 19% and up to 90 to 100%. In each range with 10 or more outcomes we print the average estimate next to the share admitted.
The Brier score is the average of (estimate minus outcome) squared, with the estimate as a fraction. Lower is better.
Reports are opt-in, checked against each college's verified release date, and unusual ones are held back. No group under 5 students is shown anywhere.
Students choose to report, so the sample is self-selected. The grade describes the estimates for those students, not every applicant.
Log your decisions in My Schools and turn on the anonymous tracker. Score bands only, never your name or exact numbers.
Share my decisionsQuestions
How accurate is CollegeCalcAI?
We will say once it can be measured. The first accuracy report publishes after decisions arrive, once 50 admit or deny outcomes are in. None are in yet: most Early Decision and Early Action results arrive in December. Below is exactly how it will be graded.
How is the calculator graded?
Students who use My Schools can share each decision anonymously, with the estimate we gave them before it. Admit and deny outcomes are graded; deferrals and waitlists are not final. Estimates are grouped by range (0 to 9%, 10 to 19% and so on), and in each range we compare the average estimate with the share actually admitted. A range prints once it holds 10 outcomes. We also publish the Brier score, the average squared gap between estimate and outcome.
Why can the reports be skewed?
Reporters choose to share, so they are not a random sample of applicants, and a student who heard good news may be more likely to report. The grade measures the estimates for the students who reported, and we say so next to every number.