Methods and data
How numbers are computed and where records come from.
Comparing a vote with most voting Democrats and Republicans
For each recorded vote, we count Democrats who voted Yea and Nay on that exact roll call. “Most voting Democrats” is whichever side had more. We do the same for Republicans, and show both, because the party that controls the House differs by Congress. If the two sides are equal there is no comparison, and if no member of that party voted Yea or Nay there is nothing to compare. Present and Not Voting are counted separately and are never treated as agreement or disagreement. Voting against most members of a party is a fact about one vote, not a judgment that the member did something wrong.
Reactions and straw polls are not polls
Thumbs up and down reactions come from members of this community who chose to respond to an explicit question. Each member has one active reaction per item. They are not a scientific sample and are labeled non-scientific wherever shown. Straw polls carry the same label. Scientific polls appear only with full disclosure of pollster, sponsor, dates, population, sample, mode, wording and margin of error.
Petition counts and campaign progress
A petition count is the number of signature records on file. One member can sign once. Campaign progress is calculated from an append-only ledger of recorded contributions, refunds and disputes, not typed in by anyone. Money is released to vendors only after a second person approves.
Data sources and coverage
- House roll calls: official XML files from the Clerk of the U.S. House. Coverage is the roll calls we have imported, shown with their dates on the home page. Senate roll calls are not imported yet because the Senate site blocks automated access.
- Members of Congress, presidents and vice presidents, committee seats and official social accounts: the public-domain unitedstates/congress-legislators dataset (CC0).
- Executive orders: the Federal Register public API.
- House financial disclosures: the filing index of the House Clerk, linked per filing. Net-worth ranges and stock trades are read from the filings by an editor and confirmed by a second editor.
- Lobbying registrations: Lobbying Disclosure Act filings from the LDA public API. A match to an official shows only after an editor confirms it.
- Bills sponsored and cosponsored: the Congress.gov API, read when an official's page is opened (kept for an hour). Without an API key the page says it is not available.
- Party committee finances: totals reported to the Federal Election Commission through its public API, shown with coverage dates.
- Leadership events and statements: each carries a source link and date, and a statement needs two different reviewers to be confirmed. Secret-ballot votes are never inferred.
- News headlines (Watch Desk and Newswire): headlines and links, never article text, from the public RSS feeds of news outlets and fact-checking sites of different political leans. Editors use them to spot claims worth checking. The Newswire account reposts them word for word, credited to the outlet and linked; we have not checked them.
- AI research: an AI service (Anthropic) with web search drafts fact checks for editors and writes the labeled AI news briefs. AI is never a source: a draft goes through the two-person check, and a brief posts only after our code opens its sources.
Import status, the time of the last successful import and any errors are tracked per source. Missing data is shown as unknown.
Source rights
We import records only from sources whose terms allow it: official government records and public-domain datasets. We keep the source link and fetch date for each import. From news outlets we read only headlines and links from their public feeds, and we link to the outlet rather than copy its articles. We do not use commercial datasets. Corrections are welcome; see the about page for the principles we follow.
Frequently asked questions
How do you make sure your news and fact checks are true?
Every key claim needs at least 5 independent sources with different owners on different websites, tagged right at the claim so you can open them yourself. A person who did not write the piece opens every source before it is published, and the person who publishes is never the writer. AI may help research but is never a source and never publishes a story or a fact check. The one exception is AI news briefs in the Brays feed: they post on their own only after our code confirms at least 3 independent sources, and each is labeled as written by AI and not checked by a person. Corrections are dated at the top of the story (or of the Record section on an official's page) and listed in the public corrections log.
How is “with most voting Democrats” calculated?
For each recorded vote, we count the Democrats who voted Yea and the Democrats who voted Nay on that exact roll call. “Most voting Democrats” is whichever side had more, and we show the same comparison for Republicans, because the party that controls the House differs by Congress. If the two sides are equal there is no comparison, and if no member of that party voted Yea or Nay there is nothing to compare. Present and Not Voting are counted separately and are never treated as agreement or disagreement. Voting against most members of a party is a fact about one vote, not a judgment that the member did something wrong.
Are straw polls and member reactions scientific?
No. Thumbs up or down reactions and straw polls come from members of this community who chose to respond, so they are not a scientific sample of voters and are labeled non-scientific wherever they appear. Scientific polls are shown only with full disclosure of pollster, sponsor, dates, population, sample, mode, wording and margin of error.
Where does the vote data come from?
House roll calls come from the official XML files published by the Clerk of the U.S. House. Members of Congress come from the public-domain unitedstates/congress-legislators dataset. Party committee finance totals come from reports filed with the Federal Election Commission. Each record keeps its source link and fetch date, and missing data is shown as unknown rather than zero.
How are campaign funds handled?
Campaign progress is computed from an append-only ledger of recorded contributions, refunds and disputes. It is never typed in by hand. Payments run only after payment-provider keys are configured and an administrator turns payments on. Money is released to a vendor only after a second person, different from the one who requested it, approves.
How are leadership statements and races verified?
Each leadership event and statement carries a source link and date. Reported and confirmed claims are labeled differently, a statement needs two different reviewers to be confirmed, and secret-ballot votes are never inferred.
Is Project Blue USA affiliated with the DNC or any political party?
Project Blue USA is independent. We are not affiliated with or endorsed by any political party (including the DNC), candidate, campaign, PAC, government or company. We take no money from parties, candidates, campaigns, PACs or governments, and no one outside our editors can edit or approve what we publish. Our news stories, fact checks and the record entries on officials' pages meet our five-source, two-person rule, with sources tagged so you can check them yourself. Official records link to the original. Automated feed posts (outlets' headlines and AI news briefs) are labeled and are not checked by a person.