How Spurna calculates its scores

Spurna places politicians and candidates on a left-to-right policy scale. The scale runs from 50 points left, through the center, to 50 points right. A score tells you the direction of a person's estimated policy position, not whether that position is good or bad. We show an overall position and, where there is evidence, positions in 14 policy areas such as health care, immigration, taxes, and foreign policy.

For members of Congress: actions and words

We collect recorded House and Senate votes, the bills and amendments a member sponsors or cosponsors, and public policy statements. We read each proposal for the policy change it would make. For example, an amendment that cuts a particular program is scored for that cut, even if it is attached to a larger bill about something else. Votes about procedure, or measures without a clear left-to-right policy direction, are kept apart from policy-position votes.

AI organizes the text. It identifies the issue a proposal concerns, the direction of the proposed change, and whether a member's statement supports it, opposes it, or does not take a clear position. This makes it possible to compare a vote, a bill sponsorship, and a campaign statement on the same policy question.

A statistical model combines the evidence. It starts with a broad estimate of each lawmaker's position and updates it as their record adds information. Related issues share information, so a thinly documented issue does not swing a person's score as much as a well-documented one. The model estimates positions on 54 narrow questions, averages them into the 14 areas, and combines those areas into the overall score. Every lawmaker's overall score uses the same area weights, making the definition consistent across lawmakers.

For candidates: what they have actually said

A sitting member of Congress who is running again uses their congressional record. A challenger or governor candidate may not have federal votes or bills, so we instead read campaign pages and public posts. AI identifies concrete policy claims and attaches the supporting quotation when one is available. It estimates an issue position from those statements; issues with no placeable claim remain unscored. When the available claims support an overall placement, the candidate can also receive an estimate based on the issues they discussed.

Candidate estimates appear on the same left-to-right ruler as lawmaker scores. A hollow mark means an estimate from statements; a solid mark means a congressional record. The issue areas are aligned for display, so you can compare the direction and size of two positions while still seeing which one comes from a voting record.

Reading the rest of a profile

“What they care about” shows which subjects receive more of a lawmaker's visible activity: bills, statements, and posts. A large bubble means more attention, not a more extreme position. Its color shows how often that lawmaker voted against the majority of their party on votes where the parties took different sides. A candidate's cited topics show what their campaign has discussed; they are based on a different kind of evidence.

The “fit to reputation” number compares a lawmaker's record position with the rhetoric or reputation estimate on issues where both exist. It measures how closely two estimates line up; it is not a count of promises kept.

Money panels summarize reported campaign receipts, individual donations, PAC money, and, when available, donor patterns. Federal races use FEC filings; governor races can use state filings. Election odds reflect prediction-market prices for a candidate or party to win. A chamber-control figure shows what the balance would be if current market favorites won their races, rather than a separate election forecast. The quiz match percentage is a measure of similarity between your stated priorities and a politician's positions. None of these numbers is an ideology score, and money does not establish why anyone voted a certain way.

From evidence to the website

The scores are calculated before pages are published. The website receives a dated snapshot containing each person's scores, supporting details, and other profile data. New votes or posts enter the site only after they have been collected, processed, and included in a new export. A blank issue means there was not enough placeable evidence for that issue; it does not mean the person is a centrist.

We also check the lawmaker model by asking it to predict votes on bills withheld from its training data. That test helps us judge whether the estimated positions capture more than party labels.