Methodology

What we cover

Democratic Atlas tracks Democratic U.S. House candidates named on official November 3, 2026 general-election candidate lists, plus primary winners awaiting certification. We call a candidate certified only when the official document uses certification language. In California and Washington, every Democrat who advanced to the top-two general ballot is included. Louisiana’s November ballot is an open primary: every qualifying Democrat is eligible for inclusion, not a single party nominee.

State election authorities take precedence over FEC registration records for ballot status. Uncertified listings must be marked. Primary losers are retained historically but excluded from active totals. Loaded records do not by themselves establish national completeness.

No endorsements, candidate rankings or left-to-right scores. A party, group or caucus is never used to infer a candidate’s position.

How a position gets published

Find → Extract → Classify → Check against the source → Review → Publish → Recheck.

A position requires evidence linked to an original source and date. AI summaries are never evidence. AI may help find sources, extract passages and suggest classifications.

Who reviews. Every published position, affiliation and correction is reviewed by an independent AI fact-checker, which checks it against the saved copy of its source before it goes live. No human sign-off is claimed. “Reviewed” on this site means that AI fact-check.

Position and research status are separate

Support / Oppose

Evidence backs or rejects the specific policy.

Mixed or qualified

The candidate backs part of it or attaches conditions.

Unclear

The candidate addressed it, but not clearly enough to classify.

No documented position

Research found no evidence either way. This is not opposition.

Not yet researched

The candidate has not been reviewed on that issue.

Data unavailable

Records could not be loaded or are not available for that view.

Evidence strength

• A — Explicit candidate statement or published policy.
• B — Vote, bill sponsorship or formal legislative action.
• C — Explicit platform or pledge commitment.
• D — Reliable secondary reporting when no primary source is available.

Source hierarchy

1. Campaign website.
2. Official congressional record.
3. Sponsored legislation.
4. Official candidate statements.
5. Debates and interviews.
6. Organization records and endorsements.
7. Reputable reporting.
8. Other verified sources.

Conflicting sources should remain visible. A changed position keeps its earlier evidence and records the revision; it does not erase history.

How we count

Every percentage identifies its numerator, denominator, update time and query version.

• Candidates researched: active candidates with at least one reviewed position divided by tracked active candidates.
• Research coverage: candidates with a documented position on the selected issue divided by all tracked active candidates.
• Among documented: a specific position count divided by candidates with support, oppose, mixed or unclear evidence on that policy.
• Among tracked: the same count divided by all tracked active candidates.

Missing evidence never counts as opposition. Zero denominators produce unavailable percentages. Child Tax Credit records shared between Taxes and Families are deduplicated by candidate and policy for cross-issue totals.

Organizations

Membership, endorsement, support, ratings, leadership and former membership are distinct dated relationships. DSA endorsement and membership counts are shown separately, alongside their deduplicated union. Policy agreement never establishes affiliation. Constituency caucuses do not define ideology.

District boundaries and ZIPs

A state’s district geometry is displayed only after vintage review and, for replacements, an exact approved plan-ID match. Hatched state outlines show pending geometry without old district lines. ZIP crosswalks require their own verification; a new shape file does not automatically validate an old ZIP match. ZIPs can span multiple districts.

Media analysis

Planned source: the GDELT DOC API, limited to published English-language U.S. news query rules. Query terms, retrieval time, article URLs, deduplication and corpus limits must pass review before metrics are presented as findings.

The reviewed corpus, query terms and update cadence are not yet supplied in this build. No media counts are asserted here.

Count relevant Democratic House coverage, not unrelated uses of “socialism.” Repeated and syndicated stories count once; news, opinion and transcripts are distinguished where possible. Results describe the collected corpus, not all U.S. media.

DSA Scare-O-Meter

Candidate footprint uses candidates as its denominator. Media attention uses relevant articles. The attention-to-footprint ratio divides those percentages; it is unavailable when footprint is zero. Both underlying counts remain visible. The ratio does not establish bias, unfairness or causation.

Updates and corrections

Update cadence will be set at launch review. Each data page reports its source update time. Anyone, including a candidate or campaign, can submit a correction or more evidence from candidate pages. Candidates cannot edit their own pages.

A submission changes nothing by itself. The AI fact-checker rules on it against a saved source first. When a published record changes, the candidate page shows a dated line (for example “Corrected Oct 12: …”) and the earlier evidence stays visible. When a campaign disputes a position and supplies a source, its statement is shown beside ours, labelled as the campaign’s own; it never replaces our evidence.

Fixture mode is explicitly marked as fictional. Fixture receipts are temporary and are not durable database submissions.