Methodology

Fact layers (GCU)

1. College Scorecard (IPEDS unitid 104717): enrollment, admission rate, SAT average, tuition, net-price inputs, completion, retention, median earnings, demographics, CIP program shares. Each value is stored with a source and as-of date.
2. Wikidata / Commons: founding year, coordinates, official site, campus photo and seal (CC or public domain only).
3. HLC / DAPIP: institutional accreditation status and since-year. Programmatic accreditations only when we have a URL.
4. Editorial file (official/gcu.json): CTA URLs, Common App / ED flags, college list, degree levels, climate one-liner. Missing field → that heading is omitted.
5. Map / climate / safety: OpenStreetMap static map; cited Phoenix climate; link to GCU’s annual security report (the PDF is not scraped).
6. Reddit: OpenAI web search restricted to reddit.com, then a live permalink check. Threads are labeled “Sourced from Reddit.” If search or the check fails, Reviews is omitted.
7. Rankings: outbound profile links only. We do not republish U.S. News, Forbes, Niche, or Princeton Review integers or grades.

Generate / validate

Each prose heading is one model call, then a separate validation call, for at least three rounds. Any number that is not already in the fact JSON fails validation. The last passing draft is kept. Three failures omit the heading. Infobox numbers, logos, CTA hrefs, OSM, and comparison tables never go through the model.

Honest gaps

We do not invent an employment rate, live net-price calculator math, club or Greek lists, meal-plan menus, or a campus “political climate.” GCU’s IRS tax-exempt status, Education Department Title IV classification, and College Scorecard ownership code are different determinations; we do not collapse them into one label. Christian identity can differ between campus and online students.

What we skip

Portal, login, Halo, library, jobs, and logo-download pages belong on gcu.edu. We do not scrape gcu.edu, U.S. News, Forbes, Niche, or Princeton Review at grow time.