NOVEMBER 3, 2026 MIDTERMS/ State-reported returns. Participation is not open.
OUR METHODOLOGY

The context behind the counts.

Good questions deserve clear methods—and honest limits. Here is what this experiment can and cannot tell you.

An experiment, with limits

This is an unofficial public opinion experiment, not an actual election or a scientifically representative poll. Your participation does not affect election results.

Participants are self-selected. Internet access, recruitment, interest in the topics, email verification, address coverage, and nonresponse can all affect who participates. Results are unweighted descriptive counts of participating accounts. No probability sampling, margin of sampling error, or population-level inference is claimed.

What participation counts mean

One final ballot is permitted per verified account per election, with database-enforced uniqueness and transaction-safe writes. Email verification is not one-person-one-identity proof. A person with multiple accounts may evade controls. Shared households and public networks are not treated as one person.

Account and network rate limits, human verification, and review signals reduce abuse. A proxy or VPN signal alone does not prove abuse or determine geographic eligibility. Uncertain cases are routed for review rather than every VPN user being excluded.

Official 2026 midterm sources

This project focuses on the November 3, 2026 U.S. midterm elections. Current read-only snapshots cover California and New York governor and U.S. House races, plus North Carolina U.S. Senate and House races. They reproduce reported votes from the responsible state election authorities and retain an unofficial status.

The source directory also links to election authorities for all states and D.C. A directory entry does not mean that every race in that jurisdiction has been imported. These snapshots are not a complete national tally, and they do not establish national seat totals or declare winners.

Reported returns and certified results are different statuses. A figure is labeled certified only when the responsible authority identifies it that way. Any result published here must retain its source, election date, contest, reporting status, and retrieval time. Unavailable figures stay unavailable.

Participation contests must be manually transcribed or imported from an authoritative state or local election source. The workflow records the source URL, retrieval fingerprint, election date, offices, districts, candidates, parties, and ballot eligibility. A separate administrator reviews the source before publication.

Election boundaries must match the relevant election date. Each hypothetical addition requires explicit approval for that contest. Unverified local contests are unavailable. Linking to an official source or displaying its results does not approve a participation import or open a hypothetical ballot.

Source and boundary entry points include the Federal Election Commission’s election results and voting information and the Census Bureau’s congressional district geography guidance. These are research starting points, not evidence for a populated ballot.

Privacy-aware reporting

The live county map publishes separate batches only when every nonzero candidate cell has at least 50 responses. New votes wait in a new batch. County percentages describe published selections, not all submitted ballots or official turnout.

Consent is necessary but not sufficient for demographic reporting. Arbitrary demographic intersections, household-level views, and individual choices are never public. Demographic cross-tabs remain disabled pending a separate disclosure review. Final weekly totals and geographic participation summaries retain their own publication rules. Live county candidate percentages use the separate batch process, never household locations or arbitrary demographic intersections.

The survey dashboard shows every question. Live results accumulate independent, protected batches for each question, so question samples can update at different times. A batch is withheld when an answer, its complement, or missing responses could expose a group below 50. Multiple selections can total more than 100%. Written answers and background answers in “A little about you” remain private. Optional relationship and family questions can appear as protected totals for each question; individual and household profiles are not published. Published live batches remain the archive after closure. Collections without published live batches can instead receive one final release after a minimum seven-day withdrawal period.

When live reporting is enabled, survey totals and county ballot percentages may be published before collection closes, in protected batches. Optional relationship and family answers, including current status, children, and years married, divorced, or separated, may appear as protected totals for each question. Individual answers and household profiles are not published. Background answers in A little about you, including current age, and all written answers remain excluded from public results. County reporting uses your optional self-reported county or an approved address match, never an inferred ZIP location. Deletion removes pending contributions, but a contribution already included in a published batch cannot be individually subtracted. Published live batches remain the archive after closure. The final-release process and its seven-day withdrawal window apply only to collections without published live batches.

Official and hypothetical results are separate

Official election outcomes must have their own citations and a visually separate presentation. They are not merged into hypothetical counts. The midterm source snapshots and source directory are distinct from this project’s participant reports. Missing research results are not replaced with example counts or figures from another election.

Versioned questions, transparent exclusions

The survey definition is version 2026-11-15.2. Published versions are immutable. Reporting must disclose missing responses, abstentions, rejected duplicate requests, and exclusions without identifying participants. Changes to wording require a new published version.