Twenty-five colleges sit just outside the Ivy halo — Brown to Boston College, Caltech to UVA. Their Common Data Set submissions tell a coherent story about who gets in, who doesn't, and what these schools quietly decline to say.
The Common Data Set is the closest thing American higher education has to a standardized ledger. Once a year, each college fills out the same form: how many applied, how many were admitted, how many showed up. From those numbers, the simplest possible metric falls out — the overall acceptance rate.
Across our 25 schools the rate ranges from 2.6% at Caltech to 20% at the University of Virginia. That is not a smooth gradient. It is a long, stretched curve with a cliff at the low end, a flat plateau in the middle, and a public-flagship tail that finally lets in roughly one in five.
Notice that no school here cracks 25%. The very phrase "selective college" used to describe anything below 50%. In the 2023–24 CDS cycle, "selective" now means rejecting at least four out of every five applicants — even at the most permissive school in the set.
The flat shape in the middle hides three distinct populations of school. Look closer.
Sort the 25 by acceptance rate and three groups appear naturally. The first six — Caltech, Columbia, Brown, Vanderbilt, UPenn, Duke — sit below 6%. These are the schools where binding Early Decision and a small handful of recruited categories overwhelm the regular pool.
A nine-school Very Selective band runs from Dartmouth (6.2%) through Rice (9.5%). Several are Ivies. Several are not. UCLA shows up here despite being a public flagship — the UC system's test-free policy puts it in a different optimization regime entirely.
The third band — Selective at 10–20% — is where most applicants will actually find their odds. Middlebury, Carnegie Mellon, Emory, Tufts, WashU, Georgetown, Notre Dame, Boston College, Michigan, UVA. Reach schools for the median high-achieving senior, but not the lottery the front of the list has become.
A flat acceptance rate hides a non-flat applicant pool. Which brings us to scores.
Every CDS reports the 25th and 75th percentile SAT scores of enrolled students. Plot them across the 25 schools and a striking pattern emerges: the bands are tighter at the top than at the bottom, and they almost touch the ceiling.
At Caltech the middle 50% runs from 1510 to 1570 — a 60-point window pressed against the 1600 cap. At Columbia, Brown, Cornell, Vanderbilt, and UChicago, the 75th percentile is already 1560. There is barely any room left for "above the band" to mean something.
The bands stretch as you scroll down the list. Michigan's 25th percentile sits at 1340 — the in-state, out-of-state, and recruited-athlete pools widen the floor. UVA, Notre Dame, and Georgetown all show 1400-floor bands. These are scores submitted by the strongest test-takers in their classes; the actual enrolled distribution is wider.
Those bands assume scores were submitted at all. Most weren't.
The 2023–24 CDS sits in the middle of one of the largest policy reversals in American admissions. In 2021, nearly every school in this set went test-optional. By 2025, half of them had reversed course. The other half doubled down.
This matters for how to read every SAT band on the previous chart. Submitter pools are self-selected upward: at most schools, only 40–60% of enrolled students reported scores. The published 25th percentile is the 25th percentile of the people who decided their scores would help.
Three schools sit outside the wave. Georgetown never went test-optional, so its 1400–1540 band reflects its full admitted class. UCLA went test-blind permanently — its CDS reports no SAT data at all. MIT reinstated tests early, in 2022.
Tests are one screen. The other is the high-school transcript — and that one is even fuzzier.
CDS Section C11 asks colleges to report the GPA distribution of enrolled students. Eight of our 25 schools — Caltech, Brown, Dartmouth, Duke, Cornell, Columbia, Boston College, Tufts, Middlebury, Williams, Amherst, Rice, Notre Dame, and Georgetown among them — leave it blank.
The schools that do report tell a remarkably consistent story. At UPenn, 58% of enrolled freshmen arrive with a 4.0 unweighted GPA. At UChicago, it is 63%. At Johns Hopkins, 63%. At WashU and UVA — which use weighted scales — 65% and 91% have a 4.0+. Below 3.5 effectively does not exist.
The top of the GPA scale has been compressed flat by high-school grade inflation, weighted scales, and self-selection. A 3.9 unweighted student is now in the bottom half of an enrolled class at almost every school in this group. That's why test scores, essays, and extracurriculars do most of the discriminating work.
If almost everyone is qualified, almost everyone is also waitlisted.
CDS Section C2 asks colleges to report waitlist activity. The numbers are spectacular and only sometimes meaningful.
Carnegie Mellon offered 12,000 waitlist spots last cycle and admitted 32 — a 0.3% pull rate. Dartmouth waitlisted 2,200 and admitted 29. Johns Hopkins admitted 30 from a waitlist of 2,478. For most of these students, the waitlist is a polite no in slow motion.
But pull rates vary wildly. Tufts admitted 354 from a 1,500-person waitlist (24% pull rate of those who accepted a spot). UCLA pulled 1,211. Brown, 230. The list is a real budget tool at some schools, theatre at others — and the CDS doesn't make you tell the difference.
When the front door is this narrow, where do students actually come from?
Here is the sentence the CDS does not contain: which high schools sent us the most students. None of our 25 colleges publishes feeder-school matriculation in the CDS. Most do not publish it anywhere.
What they do publish is a public/private split — usually a single line in the admitted-class summary. Princeton's Class of 2028: roughly 60% public. Yale's Class of 2027: roughly 60% public. Beyond that, you're on your own.
The best feeder data the public has comes from investigative journalism: the Harvard Crimson's 2024 reconstruction identified 21 schools collectively responsible for roughly one in eleven Harvard admits. State data portals (Michigan, the UC system) publish high-school-to-college pipelines for residents. Crowdsourced sites (IvyLeagueFeeders.com) scrape what they can. Niche.com and Naviance fill in the gaps.
Feeders are missing. So are hooks, ED rates, and yield breakdowns. Time to map the holes.
The Common Data Set is a triumph of voluntary standardization, and it has hard limits. Some questions are reliably answered for every school in this group: overall accept rate, application volume, in-state/out-of-state split, first-generation share, Pell share, SAT bands.
Some are reported only by some schools, only some years: ED and EA acceptance rates, waitlist conversion, round-by-round yield. These are the fields where serious applicant strategy actually lives — and the CDS lets colleges decide whether to fill them in.
And some questions are simply absent from the form: legacy admit rates (estimated at 30–40% from court records), recruited-athlete rates (roughly 2–4× base rate per Harvard trial discovery), donor case rates, essay rubrics, geographic quotas. The CDS tells you what the door looks like. It does not tell you who is holding it open.