How Vote Impact works, in detail

Vote Impact is an independent project built to make the public legislative record readable.

It is not affiliated with, endorsed by, or funded by any government agency, political party, campaign, or advocacy organization.

Bill text, sponsors, actions and votes are public records. Vote Impact does not originate any of them — it retrieves them from official sources, links back to those sources on every page, and adds a plain-language reading on top. Where the reading and the record disagree, the record is correct. Tell us and we will fix it.

The U.S. Congress and the legislatures of all 50 states, plus the District of Columbia and Puerto Rico. Federal coverage is limited to the current 119th Congress; state coverage follows each legislature's published sessions.

Not every bill carries an analysis. A bill appears as soon as it exists in the official record; the plain-language summary and scores are generated separately and can lag, fail, or be marked unavailable. Bills in that state say so rather than showing a score we cannot stand behind. See data sources and attribution for the full provider list.

Every score is produced by a language model reading the bill's own text against a fixed rubric. The rubric is the same for every bill in every state. It never receives the sponsor's name, party, or chamber majority as an input, and it is instructed not to infer legislative effects that the source text does not support.

Four numbers come out, and they measure genuinely different things. The most common misreading is treating them as one axis:

Civic-fit score

0–100. How well the bill fits the public interest as written: 0 is severe downside or poor fit, 50 is mixed, uncertain or balanced, 100 is strong fit with limited downside. It is not popularity and not the probability the bill passes.

Risk level

Low, Moderate or High — the severity of the credible downside, not the inverse of civic fit. High is reserved for material downside the source actually supports: constitutional or due-process concerns, civil liberties, privacy and surveillance, public safety, large fiscal exposure, regressive effects, environmental harm, enforcement overreach, accountability gaps, or irreversible effects. A bill can carry real benefits and still be High risk.

Impact level and score

Low, Moderate or High, plus 0–100 — the magnitude of the real-world effect, independent of whether that effect is good or bad. High means a broad population, a major budget or regulatory change, a material shift in rights, benefits or enforcement, or a significant change in institutional power. Low means procedural, commemorative, technical or narrow.

Sector outlooks

Environment, economy and social, each 0–100. These are outlook scores, not volume: 0 is clearly harmful to that sector, 50 is neutral or balanced, 100 is clearly beneficial. A bill that simply does not touch a sector scores 50 there — it is not penalised for being out of scope.

Risk and impact are deliberately separate judgments. A bill can be High impact and Low risk (a large, well-bounded program), or High risk and Low impact (a narrow change with a serious flaw). Any product that collapses those into one number is hiding the more useful half.

Score quality depends entirely on what source was available, so every analysed bill states which one it used and carries a confidence band. There is no way to get a high-confidence badge from thin evidence:

Full bill text — high

The complete introduced or enrolled text was retrieved and analysed. This is the only source that earns a high confidence band.

Official summary — medium

The legislature's or provider's own summary was analysed, but not the full text. Operational detail in the bill body is invisible to the analysis.

Metadata only — low

Only the title, sponsors, subjects and action history were available. Treat the reading as an orientation, not a finding.

Alongside the band, each analysis publishes its own limitations and warnings — what it could not see, and the single most important concrete risk it did see. The rubric explicitly forbids downgrading a bill to Low risk merely because the evidence was thin.

They are not legal advice, and they are not a prediction of whether a bill will pass. Vote-count projections shown on a bill page are model estimates, not whip counts.

They are a machine reading of a document, produced at scale. Machine readings are wrong sometimes — they miss cross-references to existing statute, misjudge how a provision interacts with state law, and can over-read an ambiguous clause. That is exactly why the official text is one click away on every bill page, and why the source and confidence band travel with every score instead of being buried. For anything consequential, read the bill.

If a summary misstates a bill, a score looks indefensible, or a link points at the wrong record, report it. Corrections to the plain-language layer are made directly; the underlying official record is not ours to change, and we will point you at the legislature's own correction process instead.