Every number here has a paper trail.

Source hierarchy

Five rungs — and the bottom two don’t count.

Every claim is graded by where its evidence sits on a fixed ladder. The higher the rung, the more a claim is allowed to do.

1. Official school pages

   Financial aid, scholarship, honors, and department pages on the school’s own domain. The standard everything else is measured against.

2. Official state, federal, and provider pages

   State grant agencies, federal aid programs, and scholarship sponsors publishing their own rules.

3. Credible secondary sources

   Reputable reporting that points back to official policy.

4. Forums and parent groupsSignals only

   Reddit threads and parent groups are leads to chase down, not evidence. Nothing publishes on a forum post.

5. AI summariesNever final evidence

   AI helps us search and compare. A claim never ships on an AI summary alone — no exceptions.

The evidence standard

Every claim carries receipts.

Before a claim can back a number on this site, its ledger entry has to carry nine fields:

• Source URL
• Source type
• School or provider
• Aid year
• Date accessed
• Last verified
• Confidence label
• Verbatim excerpt
• Conflict status

The excerpt field is the school’s own words, captured verbatim — never our paraphrase. 3,566 award entries on our college pages carry one.

Create vs verify

The author never grades its own work.

A single reviewer — human or AI — shares its own blind spots. So verification here is structural, not a promise:

1. An author extracts

   One system reads the school’s published policy and drafts the claim — amount, conditions, source, verbatim excerpt.

2. An independent verifier re-checks

   A separate system, built on a different model family, re-reads the source and grades the claim. The system that extracts a claim is structurally barred from approving it.

3. A code gate decides

   The publish decision is made by deterministic code checking the verifier’s grade, the grounding check, and the conflict status — never by the model that wrote the claim.

4. High-stakes claims add a human

   Dollar amounts, stat cutoffs, stacking rules, and renewal conditions additionally require human sign-off before they publish.

Proportionate rigor

Rigor scales with what a number can cost you.

Not every fact deserves the same machinery. The bar rises with the stakes:

The living ledger

12,692 revisions, every one on the record.

Merit policy moves — schools revise cutoffs, retire awards, and quietly edit stacking pages mid-year. The ledger is built to notice. Every change to a claim is journaled, 12,692 revision records so far as of June 10, 2026.

• Changes are recorded, not overwritten. A revised amount or cutoff leaves a journal entry behind it.
• Stale and conflicting sources get flagged for re-verification — not silently kept because the old number was convenient.
• Every college page shows its last-verified date, so you can see how fresh the answer is before you act on it.

The same ledger powers our published research: The State of Merit-Aid Transparency, 2026 reads and classifies the outside-scholarship policies colleges actually publish — and shows what they leave out.

Coverage & limits

What we cover — and where the edges are.

of public four-year colleges — 308 of 782

39.4%

of private nonprofits — 443 of 1,536

28.8%

of all 2,318 public + nonprofit four-year colleges

32.4%

Denominators are the 782 public and 1,536 private nonprofit four-year degree-granting institutions in NCES Digest Table 317.10 (AY 2022–23). Measured against the list that actually matters — the roughly 880 schools where merit aid is a meaningful lever, per our internal coverage tracker — coverage is 749 schools, about 85%.

For-profit colleges are excluded on purpose: they are 310 of the 2,628 four-year institutions (11.8%) but enroll only 5.6% of four-year undergraduates (NCES Table 303.70, Fall 2023) — and they are not where merit strategy happens.

What this cannot promise

• Policies change mid-year. A school can revise a cutoff in March that we verified in January. The ledger flags staleness; it cannot prevent it.
• Conflicts are shown, not smoothed over. When two sources disagree, you see the disagreement instead of the friendlier number.
• Unclear rules stay labeled.When the evidence runs out, we hand you the exact question for the aid office — we don’t fill the gap with a guess.
• None of this guarantees aid.It guarantees you’re deciding from sourced numbers, labeled honestly.