Last reviewed August 17, 2026
What is the Common Data Set
A 47-page PDF on a random institutional-research URL is not a data system. It is a print form that three guidebook publishers asked schools to fill in once.
The CDS Initiative
In the late 1990s the College Board, Peterson's, and U.S. News sat down with college institutional-research offices and agreed on one template. The point was to stop asking every school for the same enrollment, admissions, and aid numbers in fifteen slightly different shapes. The Common Data Set Initiative still publishes that template. It is currently a 47-page workbook with 1,105 fields and an Answer Sheet tab.
Participation is voluntary. There is no federal mandate, no central filing cabinet, and no API. Each school posts a file on its own website, or does not.
Sections A–J, in English
| Section | What it covers |
|---|---|
| A · General Information | Name, calendar, degrees, respondent contact. |
| B · Enrollment and Persistence | Headcount, race/ethnicity, retention, graduation rates. |
| C · First-time, first-year admission | Applicants, admits, enrolled, tests, wait list, ED/EA. |
| D · Transfer admission | Transfer funnel and requirements. |
| E · Academic offerings | Special programs and policies. |
| F · Student life | Housing, activities, percent in-state. |
| G · Annual expenses | Tuition, fees, food and housing. |
| H · Financial aid | Need and non-need packages, H2A merit counts. |
| I · Faculty and class size | Student-faculty ratio, class-size bands. |
| J · Degrees conferred | Disciplinary areas of degrees awarded. |
A fillable PDF is the easy case
Harvey Mudd College still publishes the 2025-26 Common Data Set as an unflattened fillable PDF. Named AcroForm fields carry the canonical tags from the template. AP_RECD_1ST_MEN_N is 3,452. That is a 200-millisecond pypdf.get_fields() extract, not a model guess. See the Harvey Mudd archive.
Point the wrong extractor at that file and C1 shifts
The same kind of table, after a layout parser has flattened it, is why this archive exists as software and not as a folder of PDFs. Docling on the Harvey Mudd fillable file collapsed C1 so men-applied became “3452 1761” and women-applied became “4”. Kerned year numerals read as “202 5 -202 6”. The running header printed “Common Data Set 2025-2026” as a heading five times. That is documented in docs/known-issues/harvey-mudd-2025-26.md. The live extract is the AcroForm path. The Docling notes are what happens if you pick the wrong tool.
Virginia Tech publishes XLSX and hides the index
Virginia Tech's 2025-26 Common Data Set is an Excel workbook — the theoretically ideal format. The school's own page at aie.vt.edu explains what a CDS is, then says files are available via request to aiesupport@vt.edu. The PDFs and workbooks exist on a DAM path. The landing page is a dead end. The Virginia Tech archive is the public HTML that actually hands over the extract.
The publishing mess
Schools do not publish one way. In this archive we already have, as real files:
- Unflattened fillable PDFs (Harvey Mudd 2025-26).
- Flattened PDFs where the form structure is gone.
- Image-only scans with almost no extractable text.
- XLSX workbooks (Virginia Tech 2025-26).
- DOCX uploads of the Word template.
- HTML pages, some of them JS-only.
- Box, SharePoint, Google Drive, and Digital Commons item pages.
- URLs whose path year is a CMS upload month, not the academic year.
- Section-only files, blank templates, and “test” uploads that are the real file.
None of that is a data system. It is a distributed print workflow that happens to use a shared template. Stanford's 2017-18 year page is already a used URL: historical files convert when the school no longer lists them.
What “extracted” means here
Extracted means: canonical field IDs such as C.101 and H.2A, values with provenance back to the archived bytes, and a spreadsheet download of those fields. It does not mean we are the publisher. The numbers are the school's. The year page links the archived original and, when we have a usable HTML URL, the school's own CDS page.
For a counselor: send the year page, not a 47-page PDF, and keep the official link in the same paragraph so you are not vouching for a scrape.
What we are not
Not the publisher. Not IPEDS. Not College Scorecard. Those are different systems with different mandates, lags, and field definitions. This page is the school-authored form, archived and extracted.