How we read the record.

Public records first

Candidate and committee identity, money raised and spent, cash on hand, and filing dates come from the Federal Election Commission. District boundaries and place lookup data come from the Census Bureau; the district lines are those of the 119th Congress (2025–2026). Roll-call records come from the House Clerk and the United States Senate, and bill summaries, sponsors and progress from Congress.gov. Members' terms and committee seats come from the public congress-legislators project, and lobbying reports that name a bill from Lobbying Disclosure Act filings at lda.gov. Primary and runoff results come from civicAPI. Voter registration deadlines are read from each state's own election pages and laws. Polls come from The New York Times. Priorities, endorsements and committees' statements of purpose are read from their own websites.

Bulk snapshots keep core browsing available during upstream outages. Live APIs enrich those records when available; they never need to succeed for a directory page to render.

Filings are not endorsements

PAC totals and candidate finance figures describe reported money movement. Committee names, connected organizations, transaction types, support and oppose flags, and coverage dates are retained so readers can distinguish direct contributions from independent spending.

Where a campaign's money came from is the FEC's own summary of its receipts by source: individuals (split at the $200 itemization line), PACs and other committees, party committees, and the candidate, whose loans to their own campaign count as their money. The compare sheet's outside spending adds up independent expenditures for and against each candidate from periodic reports, plus the 24- and 48-hour notices a committee has filed since the last period it reported, since a notice is repeated by the next periodic report and counting both would count it twice.

A PAC or committee profile lists, with its FEC filings, the audits and administrative fines the Federal Election Commission has published against it, matched on the committee's own FEC ID. An audit is a periodic examination that reports findings; an administrative fine is a penalty for filing a specific report late or not at all. They are separate instruments and are never combined into a compliance score. Most committees have neither, and an absent record is stated as such rather than read as a clean one.

Enforcement matters, the Commission's investigations of alleged violations, are not shown. They are published against respondent names with no committee identifier, and matching one to a committee by name risks attributing an enforcement action to the wrong organization.

A PAC's party-support chart adds its direct contributions to candidates' committees and its independent expenditures marked as support, grouped by the party of the candidate they went to. A candidate is found by FEC candidate ID, then committee ID, then name; any not found fall under "Other or unresolved". Spending against a candidate is never counted as support and is shown as its own figure. Individual, earmarked and conduit transactions are left out, and a chart built only from a PAC's latest transactions says it is a sample. A PAC giving to both parties is common and is not an endorsement.

The lobbying shown for a PAC's sponsor is matched by name, because the FEC and the Senate share no organization identifier. The committee name is cut down to its sponsor ("Lockheed Martin Corporation Employees' Political Action Committee" becomes "Lockheed Martin") and lda.gov is searched for filings whose client name contains it. A name too generic to search on is skipped, and the profile shows the name it matched so a reader can judge it. Separately, the organizations listed on a bill's page are those whose filings cite the bill by number, so that list is a floor, not a full count.

In their own words, in their own order

A candidate's priorities are what their own campaign site publishes: its issues, priorities or platform pages. Every heading and statement shown is copied exactly from that page and links back to it, and under each heading everything the candidate wrote there is shown, in the page's order. Nothing is summarized, paraphrased, scored or placed on a scale, and a member's official .house.gov or .senate.gov site is never read for campaign positions.

Code reads each page into the candidate's own headings with the sentences beneath them. An AI model decides only which headings are topics, which group topics, and which are navigation or biography, and which lines under each heading are its text rather than buttons, sign-up pitches or the next topic; it returns labels, never text, and each label is remembered so the same page gives the same result on every run. A panel of three AI models then checks each heading, and one is left out only when all three agree it is a slogan or a page header rather than a priority; when two of the three do, a person decides. A heading whose every line is already shown under another, such as a homepage card and the topic page it links to, is shown once.

There is no list of issues we sort candidates into. A fixed list would be our idea of what a voter should care about, so candidates are shown on the topics they chose, in the order they chose. A candidate who publishes no list is shown as having none; one is never written for them. An editor's correction always overrides the models. Party affiliation is never used as a substitute for what a candidate says or how they vote.

Who is still on the ballot

Primary and runoff dates, and the results of contests already held, come from civicAPI, which aggregates them from Secretaries of State and equivalent election bodies; the November general election date is set by federal law. An FEC filing does not lapse when a candidate is beaten, so a candidacy is only described as ended on the strength of a reported result.

A defeat is shown only when the race has been called and the candidate is matched to a losing line in it. Absence from a published field is never read as a defeat, and a race reporting no winner leaves every candidate in it undecided. Each result links back to the published race so the count can be checked.

A death is recorded for members of Congress only, from Wikidata's date of death, joined to the member by Bioguide ID and never by name. A date is kept only when it is exact to the day and Wikidata gives one date for that member. The FEC keeps listing the dead as running, so the death leads the candidate's status, with the ballot as the detail: who replaced them, or the contest their name still appears in.

Race calls on this data are made by civicAPI's own decision desk rather than by a state election authority. Where a call and an official certification could differ, the published result and its source are shown so readers can judge for themselves.

Voter registration deadlines for each method (online, by mail, in person and same-day) are read from each state's election office pages and statutes, and every one links to its source. Where a state's own sources disagree, the page says so. The sources are checked daily for changes, and a change is re-read by a person before anything on the site moves; a state's election office is the final word. Each state page also links to the state's own polling-place lookup, registration check and mail ballot pages, since polling places are assigned by street address and only the state can say where yours is.

How votes are read

Every vote shown is a roll call recorded by the House Clerk or the Senate, with each member's position as recorded. Bill titles, sponsors and progress come from Congress.gov.

The three focus votes on the votes page are picked by a fixed score, not by an editor. They are drawn from the measures on either chamber's 30 newest roll calls, and nominations are left out. A measure earns points for being a final vote in its chamber, for the policy area Congress.gov files it under (most for the budget, taxes and defense, then foreign affairs and finance, fewer for health, energy, immigration, education and similar areas), and for having several recorded votes; a roll call with no measure behind it scores far lower. The top three show, the newest first when scores tie, and each card lists the reasons it scored.

A member who voted is tied to their 2026 candidate profile by identifier, not by name. House roll calls carry a Bioguide ID and Senate roll calls a Senate ID, and the congress-legislators roster lists the FEC candidate IDs each member has filed under, so a bill's sponsor links to their profile the same way. Names are compared only for a member the roster does not place, and then within the same chamber and state (and district, for the House), with the same surname and a matching first name.

Polls as published, not a forecast

House and Senate polls come from the poll data sets The New York Times publishes under the CC BY 4.0 license, and cover the current cycle only. Each poll is listed as published: pollster, field dates, sample, sponsor and every figure, with a link to the release. Polls a party, a campaign or another partisan group paid for are left out of every list, chart and figure, and each race's poll list says how many were left out.

A race's lead is the average gap between the two leading candidates across its five most recent nonpartisan polls, from the third poll on; partisan-sponsored polls are left out. A poll whose gap is more than 8 points from the median gap is left out, with a note in the poll list giving its gap and the median, so one far-off poll cannot set the lead. A gap compares more fairly than shares across polls that offered different choices, since undecided answers lower both shares alike. Beside the lead, the home page shows the newest poll's own numbers, as its pollster published them. The line on a poll chart is each candidate's median share across the same polls. With fewer than three nonpartisan polls, the newest poll is shown on its own and labelled as one poll. It is arithmetic on the published figures that anyone can repeat from the list beneath it. No win probability is computed, and no district is given a poll it does not have: most House districts are never publicly polled.

Before the primaries, pollsters test pairings that never reach the ballot. The chart follows the matchup the most recent polls tested among candidates who can be on the November ballot; the other pairings are listed separately and not charted.

Each poll also shows how it was conducted, which voters were asked (a primary poll of one party's voters says so), which round of a ranked-choice count it reports, and when it was added. Where a public poll is missing from the Times' data, an editor may enter it from the pollster's own release; it is marked “Added by our editors”, links to that release, is checked before it appears, and gives way to the Times' copy once the Times adds the same poll.

Once a primary, runoff or general election is called, its result is set beside the final polls: each pollster's last poll ending in the three weeks before the vote, read for the two leading finishers. The comparison is of margins (a poll's lead against the actual lead), because primary polls often leave a fifth or more of voters undecided. The polls are measured against a result only when at least 95% of the vote has been reported; results are unofficial counts from civicAPI, and where its count stopped short the page says so rather than reading a counting gap as a polling miss.

Broad records, honest gaps

Every current-cycle House and Senate filing receives a profile. Finance depth varies with filing activity, and published priorities vary with what a campaign's own site says. Each page labels unavailable, imported, live, automated, and reviewed information distinctly.

Ballot status depends on a filing being matchable to a name on a published ballot, so a candidate who never appears in a reported field keeps their filing status alone.

Glossary

Cash on hand

Money a campaign or committee had in the bank at the close of its latest report, not today.

Itemized

A donation listed by donor. Once one person has given more than $200 in a cycle, the filing names them; smaller gifts are reported only as a total.

Independent expenditure

Money spent for or against a candidate without coordinating with the campaign, such as ads or mailers.

24- and 48-hour notice

A quick report an outside group files after spending near an election, before its next regular report. The regular report repeats it, so we count it once.

Super PAC

A committee that may raise unlimited sums but may not give to candidates. It can only spend independently.

Hybrid PAC

A committee with two accounts: one gives to candidates within limits, the other raises unlimited sums and spends independently.

Conduit

A service that passes donors' money on to campaigns. We treat only ActBlue and WinRed as conduits; the FEC files no flag for one.

Cloture

A Senate vote to end debate on a measure. Legislation usually needs 60 of the 100 senators.

Nominee

The candidate a party's primary chose for the November ballot. If the party later names someone else, the first nominee is shown as replaced.

The poll lead

The average gap between the two leading candidates in the latest nonpartisan polls, with far-off polls set aside. It is not a forecast.