Voters get plenty of adjectives about politicians. They get very few numbers computed the same way for everyone. This project scores officials on two independent axes — did they do what they said (Honesty), and did the place get better on their watch (Performance) — under rules published before they are applied and changed only in public. The name is the whole idea: every official gets a grade from the same public data — not cheerleading, and not from any campaign.
We publish the methodology, the data sources, and the exact inputs behind every number. If you think a score is wrong, you don’t have to trust us — you can recompute it. If it’s still wrong, dispute it, and the ruling gets published too.
One rulebook. It binds us too.
- No promise rating that can damage a reputation — an Abandoned or Betrayed — will publish without an evidence trail and two agreeing human reviews. Performance scores are different: they are computed from public government data under preregistered rules, and publish with the full math attached so anyone can check them.
- Every rule must read the same with the party labels swapped.
- Uncertainty is badged, never hidden — provisional scores, thin data, and coverage gaps are labeled on the page.
- Corrections are logged publicly. No silent edits.
Who runs this, and who pays for it
Political Grades is an independent project, funded entirely by its founder. It takes no money from any party, campaign, PAC, advocacy organization, or donor with a pending rating — no advertising, no sponsorships, no grants. Nobody scored on this site has any financial relationship with it.
Political Grades is operated by Green Wave Ventures LLC and answers at a monitored address and a public mailing address: 712 H Street NE, Suite 1555, Washington, DC 20002 — the same address every notice this site mails carries by law. Every official this site scores is offered right of reply there and at challenge@politicalgrades.com, and corrections follow the published policy below — the site holds itself to the answerability it asks of the people it grades.
How this site is built
The code behind this site is developed with AI assistance, under the founder’s direction and review. What that does not mean: no sentence about any official is composed by a model at publish time. Every sentence you read about a person is assembled by a fixed template from exported data — the template can choose which verified fact leads, but it cannot supply a fact, an adjective, or a causal verb the data doesn’t license — and that rule is enforced by automated tests that re-read the built pages against the published numbers on every release, not by anyone’s promise. The reason to say this plainly is the same reason the rest of this page exists: a method you have to discover looks like a method being hidden. If a sentence on this site is wrong, it is wrong in the data or in a template — both fixable, both changelogged — and the dispute address above reaches a human either way. Two things on this site are written by a model, and both are labeled where they appear. The “Ask about this bill” box generates an answer at a reader’s request, grounded only in that page’s verified record and marked as not the site’s editorial voice — if it ever disagrees with the page, the page is right. And on bill pages, the plain-English description beside each dollar figure, and the extraction notes beneath them, are drafted by a model reading the enacted text: every dollar is then mechanically re-located in that text as an exact string, every description is checked against it for anything the law does not say, and both are re-verified by a second adversarial pass before publication. The dollars and the citations are the record; the wording around them is drafted and then checked. The editing pipeline itself — what an edit is, what the publish gates check, and why no dashboard changes the live site directly — is documented at How This Site Is Edited.
Data sources
State outcome data comes from 12 published source systems — BEA, BLS LAUS, CDC BRFSS, CDC WONDER, Census STC ÷ BEA personal income, EIA, EPA AirData, EPA SDWIS/ECHO, K-12 School Shooting Database (CHDS), NAEP (Nation's Report Card), US DOL ETA, USDA FNS — behind the 17 sourced metrics on the state card. On each state page, the Show the Work table lists each metric scored for that state beside its source — scored metrics only, so a state with nothing scored yet shows no rows there. Separately, every one of the 17 sourced rows on the state card opens on the “?” to that metric’s source, the year its latest number describes, and a link to the source itself. The remaining card rows carry no number yet, and say which status they hold instead. The Congress pages run on different sources: Center for Effective Lawmaking effectiveness scores, attendance and party-line figures computed from the roll calls the House Clerk and the Senate publish themselves, bill dollar figures read from the enacted text at GovInfo, and federal budget figures from OMB Historical Tables — the annual totals, the breakdown of those totals into the nineteen budget functions the government appropriates against, and the composite outlay deflator OMB deflates its own tables with, which is what lets two fiscal years or two laws from different years be compared at all.
The governor roster — names, parties, dates of office — was independently verified on July 12, 2026, with mismatches re-checked, and the sources used are listed under Roster verification sources on every state page. Honesty about those sources: for 29 of the 50 states the listed links are Wikipedia articles (used as a compilation source and linked so you can check them); the rest add official announcements or news reports. Roster facts are never the measured series — the scored series come only from the government statistical agencies named on each metric — but dates of office do decide who is graded and over which years, which is exactly why they are verified and mismatch-checked at all.
Four files ship. The governor scores are at /governors.json and the Congress member cards and grades at /congress.json. The cities split in two: /cities.json is the verified mayor roster and carries no grades, and the city grades, their inputs and their series are at /city_scores.json.
Disputes & corrections
Officials and campaign staff named on this site may file a dispute against any published rating, and any reader may flag an error: write to challenge@politicalgrades.com.
What happens next, and when. Every dispute is acknowledged within 3 business days — that one is a commitment, not a target. The ruling itself is reviewed against the published methodology and published within 30 days; if a case needs longer, the delay and the reason are posted rather than left silent. This site is run by one person, so the acknowledgement is what you can rely on.
Corrections are appended, never overwritten — the original stays visible alongside what replaced it — and every one is recorded in the public changelog & corrections log, dated. A dispute that turns out to be right is the most useful mail this site receives.