Five universities — Harvard, Yale, Princeton, Stanford, and MIT — sit at the end of a funnel that grew narrower every year of the last decade. We took their Common Data Sets, the dry filings each school submits annually, and read them as a story.
Seven cycles ago, the Class of 2023 applied to Harvard from a pool of 43,330 students. In 2024, the Class of 2028 numbered 54,008. At Yale and Stanford, the totals now top 57,000. The pandemic-era surge — when test-optional policies opened the door — widened applicant pools that had already been growing for years.
MIT, the smallest of the five, climbed from roughly 21,000 to nearly 30,000. The trajectory is uneven: Harvard's Class of 2026 marked the all-time peak before a partial retreat. But the long arc is unambiguous. Each spring, more students compete for the same roughly two thousand seats per school.
More applicants. Same number of seats. The math takes care of the rest.
In 2019, MIT admitted 6.6% of applicants and Yale 5.9%. Today both schools sit below five percent, and the entire group has converged into a narrow band between 3.6% and 4.6%.
The Class of 2024 looks like an aberration — every rate ticks up — because COVID gap years flushed the pipeline and inflated the numerator. Strip that anomaly, and the line moves only in one direction. Every dot below an earlier dot.
When the front door closes this far, the side door begins to matter more.
Common Data Sets break out the early round, and the gap is striking. In the 2023-24 cycle, applying early at Harvard meant a 7.6% chance of admission; applying regular meant 2.6%. Yale's split was 10.0% versus 3.5%. Stanford's was 9.2% against 3.0%.
These are not different students applying. They are mostly the same kind of students filing into two different lines. The early line moves faster. Whether that's because committed applicants are stronger, or because admissions offices use the round to lock in yield, depends on whom you ask.
Most of the schools cluster around a 3x advantage. One does not.
Divide each school's early acceptance rate by its regular rate and a clean ratio appears. Harvard, Yale, and Stanford cluster near 3x. MIT, whose Early Action is non-restrictive, offers only a 1.5x bump — the early pool is less self-selected, more of a try-your-luck round.
Then there is Princeton. 15.8% early. 2.9% regular. A multiplier of 5.5x, nearly double the rest. The school has long been more transparent than its peers about valuing demonstrated interest, and the multiplier shows it. For applicants, this number is not abstract: it changes where you should apply early.
Once admitted, students still have to be admitted on paper. That paper looks remarkably uniform.
Almost every CDS reports the middle 50 percent of submitted SAT scores in the same place: 1500 to 1580. The five schools are statistically indistinguishable from each other on this metric. The 25th percentile sits within ten points across the entire group.
The lone differentiator is MIT's math floor — 780 at the 25th percentile, compared with 760 elsewhere. For a school where a third of admits will major in Course 6 or 18, the implicit message is unsubtle.
Submission rates vary widely. MIT requires a test (83% of admits submitted). Stanford, the most permissive, sees only 47%. Read the SAT band at Stanford as a ceiling for those who chose to show their work; there is no public median for the silent half.
If the test scores barely separate the group, the GPAs separate them even less.
At Stanford, 75.2% of enrolled students arrive with an unweighted 4.0. Harvard, 74.0%. Princeton, 66.7%. Three quarters of admits already maxed the scale before the application even arrived.
The shape is what matters. Rather than a normal distribution centered somewhere short of perfect, the GPA distribution at HYPSM piles up against the ceiling and then trails off. Below 3.5, the entire enrolled cohort fits in a single percent or two — generally recruited athletes, specific institutional priorities, or transfer cases.
Test scores plateau. GPAs plateau. What's left is the part of the file that's harder to graph.
Once admitted, where do students go? Yield is the share of admits who actually enroll, and it is the metric that separates the five schools more cleanly than acceptance rates ever do.
Harvard tops the chart at 84%. Stanford follows at 82%. MIT sits near 82% as well, helped by its STEM specialization — students who chose MIT rarely had a parallel offer they preferred. Yale lags at roughly 70%, a curiosity in the data: cross-admits from Yale go somewhere, and that somewhere is usually Harvard or Princeton.
Princeton's exact yield is harder to pin down — its CDS reports enrollment under expanded class sizes — but recent estimates put it near 75%. The upshot: an admit letter from Harvard functions like a near-certainty of attendance. A Yale letter functions like a strong invitation.
After the dust settles, the Class of 2029 is where the cycle ends — and our visualization, too.
The most recent CDS files — for the cycle that ended this past spring — give us the clearest snapshot of the funnel as it stands. Over a quarter million applications. Just under ten thousand admits. Roughly 7,400 students will spend the next four years on these five campuses.
Stanford received the most applications (57,326) and posted the lowest rate (3.6%). MIT remained the most selective per-applicant, but its STEM focus keeps the absolute pool smaller. Yale's applications nudged ahead of Harvard's for the second straight year, even as its rate dipped to 3.9%.
For the simulation engine on this site, these are the calibration anchors: 3.5% to 5.0% for the HYPSM cluster, with Princeton's early multiplier as the outlier and Yale's lower yield as a known asymmetry. Read the chart as a map of the year that just finished.
data_hypsm_cds.md (each school's section). Some Princeton points approximated.data_hypsm_cds.md. Princeton Class of 2027 marked Not Released.data_hypsm_cds.md.data_hypsm_cds.md.data_hypsm_cds.md.data_hypsm_cds.md. Yale and MIT do not publish full GPA distributions.data_hypsm_cds.md.