Methodology & sources

The dataset

Every figure comes from the U.S. Department of Education College Scorecard (Most-Recent-Cohorts-Institution, 2026-06-10 release), one record per institution. It flows through a transparent bronze → silver → gold pipeline into the Atlas Commons lakehouse; the page reads the gold serving surface. Compute is never local — the data is materialized from the published file, not hand-edited.

Four-status epistemics

Every number on a school page is labeled by how much weight it can bear:

• RECORDED — published Dept. of Education values: net cost, median debt, median earnings (with the 25th–75th-percentile band), completion rate, and the share earning above a typical high-school-graduate wage.
• DERIVED — two disclosed in-house ratios: debt ÷ a year's earnings (the manageability signal) and the earnings premium vs. the $28,000 HS benchmark. The formula and inputs are shown so anyone can reproduce them.
• INTERPRETATION — the boundary note. The lowest-warrant claim: framing, not a measurement.

What this is NOT

• • Not a prediction of your outcome. Every figure is an institution-wide median. A computer-science and a fine-arts graduate of the same school can earn wildly different amounts.
• • Not a field-of-study figure. For earnings and debt by major, use the official Scorecard's program-level data.
• • Not causal. Selective schools admit students who would likely earn more anyway — selection, not proof the school caused the outcome.
• • Not a "best college" ranking. There is no score and no ordering of quality.
• • Scope: earnings and debt cover federally-aided students, measured 10 years after entry (not graduation).

Agent-native

Every school page is also available as research objects (JSON-LD), plain Markdown, and JSON, with each figure carrying its epistemic status and a falsifier. The resolving vocabulary is at /vocab/agent-native; orientation for agents at /llms.txt.