The AdmitGPT Admissions Dataset
AdmitGPT is built on real data, not vibes. Below is the open dataset that calibrates our probability engine — published so anyone can verify, reuse, or build on it. Most “chance calculators” hide their data. We show ours.
What’s in the dataset
College records are sourced from the federal IPEDS database (admission rates, test scores, cost, graduation rates). The 1,122 applicant profiles are self-reported by real students between 2017 and 2023, and are used to calibrate and validate the engine’s ordinal accuracy (AUC ~0.74).
Most selective schools in the data
Admission rate by school, lowest first (verifiable against each college’s Common Data Set):
| School | Admission rate |
|---|---|
| Atlantic Acting School | 2% |
| California Institute of Technology | 3% |
| Minerva University | 3% |
| Stanford University | 4% |
| Harvard University | 4% |
| Yale University | 4% |
| Columbia University in the City of New York | 4% |
Figures reflect the admission_rate.overall field in the underlying IPEDS extract; always confirm current rates with each school.
Download the data
Both files are published under the MIT license. Use them for research, journalism, or your own tools — attribution appreciated.
- collegesdata.json — 6,273 colleges (admission rates, test scores, cost, outcomes)
- studentsdata.json — 1,122 anonymized applicant profiles
How we use it
The college data populates every school profile in the calculator; the applicant profiles calibrate the additive-logistic engine and validate its ranking accuracy. The full math is on the Transparency page, and the methodology is summarized on About AdmitGPT.