2026 Midterms · Methodology
Documented positions, not inferences.
Every claim on this site traces to something a candidate published under their own name. This page explains where the data comes from, what the vocabulary means, what we deliberately leave out, and how fresh any given page is.
Where positions come from
For each settled race we capture two kinds of sources per nominee: their own campaign website and their Ballotpedia record. Positions, key points, and quotes are extracted from those captures — and only those captures. Quotes are verbatim, and each one is attributed to its source right where it appears. Both source links are published on every race page, so you can check our reading against the original in one click.
The numbers today: 497 of 506 tracked races are settled; 496 have full issue guides covering 4,594 issue matchups and 9,026 verbatim quotes.
What we refuse to show
The extraction pipeline also produces strategist material — likely attack lines, coverage gaps a campaign could exploit. None of it renders here, by rule. This site shows voters what candidates say, not how to fight them.
When a candidate hasn’t published a position on an issue, we say “no documented position in our sources” — we never paper over silence with a guess. Absence of a documented position is not evidence the candidate has no view; it means our sources don’t state one.
The ratings vocabulary
Competitiveness tiers come from the major nonpartisan handicappers — Cook Political Report, Inside Elections, Sabato’s Crystal Ball, and Decision Desk HQ — as of the data date. Where they disagree we take the most competitive rating on offer, not the average: if one outlet calls a seat Lean and three call it Solid, we show Lean. That errs toward surfacing a race rather than hiding one, and it means a few battlegrounds rest on a single forecaster’s call. Tiers describe how contested a race is expected to be. They are navigation, not prediction: nothing on this site forecasts a winner.
A tier names a degree, so we print the side with it — Lean R, not a bare “Lean”. The letter is whichever party the outlets’ ratings point to; a Toss-up has no side by definition, and on the rare race where the outlets point opposite ways the chip stays bare.
- Toss-up
- Either nominee can win it; the closest races in the cycle.
- Tilt
- The barest edge to one side — closer than Lean, not a coin flip.
- Lean
- A modest but real advantage for one side.
- Likely
- A clear favorite; an upset would be a surprise.
- Safe
- Not considered competitive this cycle.
We call Toss-up, Tilt, and Lean races battlegrounds; those are the races featured on the home page. Issue salience (high · medium · low) records how prominently an issue figures in that specific race — in the candidates’ own emphasis and the race’s coverage — so the issues that are actually being fought over read first.
Settlement and freshness
A race is settled when both major-party nominees are determined — by a primary, a runoff, or a nominee running unopposed. The dataset grows as primaries resolve: unsettled races show their primary date instead of a matchup, and issue guides appear once the matchup exists.
Every race page carries a “positions verified” date — the day its sources were captured and extracted. Campaign sites change; a position published after our capture date won’t appear until the next refresh. Data last gathered September 10, 2026.
2026 Midterms · Settlement
When the matchups arrive.
Every tracked race, bucketed by its primary date. Ink is settled; hollow is still waiting. Issue guides appear as each cluster resolves.
Who makes this
2026election.com is a Pluribus AI publication, sister to 2024autopsy.com. The pipeline is automated; the rules above — verbatim quotes, named sources, honest absences, no strategist framing — are not optional steps in it. They are the product.