Trust
How we research, refresh, and verify.
One discipline runs through every state page: publish only what we can source, and say how fresh it is.
Primary sources first
Every fact starts at the source. The statute, Section 71401 of Public Law 119-21. The CMS program documents and award announcements. Each state's published plan, its lead agency pages, and its procurement portals. We read the original, then we record it with a link, so you can read it too.
Stability tiers
Not every fact ages at the same speed. A statutory dollar figure is settled. A named program officer can change next quarter. An open deadline moves. So each fact is handled according to how fast it changes.
Award amounts, statute, sub-initiative budgets. Settled facts, shown without a date stamp.
Named officials, gate organizations, at-risk hospital lists. Shown with the quarter we last confirmed them.
Open funding opportunities, listening sessions, deadlines. Shown with a live timestamp and refreshed most often.
Citation discipline
Every claim resolves to a citation: source, link, access date, confidence. You can see how this works on the provenance page. When two sources disagree, we surface the more authoritative one and note the difference rather than averaging them away.
Freshness is earned, source by source
A national refresh does not stamp every citation with today's date. Each source has to respond, and any stored quote has to match what the source currently says. A successful fetch can advance the access date. A missing page, blocked request, binary document we cannot quote-check, or quote mismatch stays at its prior verified date and enters the public verification queue.
On August 9, 2026, this process checked 812 citations across 602 unique source URLs for all 50 states. The unresolved blocked and transient sources are published on the errata and verification page, separate from corrections to facts that were previously shown.
We publish only what we can source
When a state has not released a detail, we leave it marked as open rather than fill it with a guess. When we infer a figure from a comparable state, we label it as an inference. The goal is a map you can stake a bid on, which means the gaps are honest too.
When we get something wrong, we fix it in the open. Every correction since launch lives on the errata page.