Editorial standards

Fact and simulation stay separate

Every page keeps three layers distinct. Verified facts: offices, states, election dates and race structure, maintained in the site's own data and checkable against official sources. Sourced material: headlines and background, always attributed and linked to their publishers. Simulation: AI-generated results and analysis, always labeled as AI simulation and never presented as polling, results or fact. If you cannot tell which layer you are reading, that is a bug; report it.

Sourcing

Race background is drawn from Wikipedia; current coverage is aggregated from established outlets via Google News, with every headline linking to its original publisher. The site does not write news, does not paraphrase reporting into new claims of fact, and does not cite sources that do not exist. Election structure comes from public records of which seats are on the 2026 ballot.

Neutrality

Electorate HQ endorses no candidate and no party. Simulation rules are identical for every ticket. Issue guides describe how issues behave electorally without taking positions on them. Advertising, when present, is visually separated, labeled, and has no influence on simulations or editorial content.

No fabrication

The site does not invent polls, endorsements, statistics, quotes from real people, testimonials, or user counts. Simulated voices in the simulator are explicitly fictional characters inside a labeled simulation, never presented as real individuals. Where verified information is unavailable, pages say so rather than filling the gap.

How simulations are constrained

Left unconstrained, a language model asked to simulate an election will reproduce the familiar electoral map, because that map is overwhelmingly represented in what it has read. That is the single largest failure mode of this kind of product, and the site fights it in three places. The prompt forbids reasoning from how a state "usually" votes and requires each state to be decided issue by issue. The output schema makes the model record which issues it weighed and which way each one cut, so the reasoning is inspectable rather than implied. And server-side rules enforce arithmetic the model cannot override: shares that total 100, a ceiling on third-party share, and a cap on the margin in states whose own issue scoring came out even.

Two rules exist because of specific failures found in testing. Candidate names are treated as untrusted input, because names padded with praise were used to talk the model into flattering conclusions about people with disqualifying records. And party letters are treated as ballot slots rather than parties, because a real Democrat placed in the Republican slot was being scored by the party they belong to in real life, which produced results that contradicted their own explanations.

What we will not publish

No invented polling numbers, no invented endorsements, no quotes attributed to real people that they did not say, no fabricated citations, and no statistics that cannot be traced to a real source or computed from verifiable data. Where the site states a factual number, it is either drawn from public records or calculated from published boundary and apportionment data, and the method is described on the page or in the methodology.

The site also carries no testimonials, no user counts, no "trusted by" claims and no artificial urgency, because none of those could be substantiated. If you ever see something on Electorate HQ that reads like a marketing claim, treat it as a bug and report it.

Corrections

Factual errors are corrected promptly once verified, and pages carry updated dates when their content materially changes. To report an error, use the contact page and include the page URL and what you believe is wrong. Simulation outputs that violate the site's own rules, for example a result that contradicts its own reasoning, are treated as bugs and fixed in the product.