Editorial policy

We publish the official record — legislators, bills, and roll-call votes — sourced from public-domain or CC0 government data, with every figure linked back to where it came from. See Methodology & sourcesfor exactly where the data comes from and how it’s kept accurate.

No vote curation.There is no “important votes” list reflecting our priorities or anyone else’s. Votes are sorted by neutral, disclosed rules (like date) — never by how newsworthy, controversial, or consequential we judge them to be.

No causal claims about money and votes.When campaign-finance data ships, we will show disclosed receipts and disclosed votes side by side. We will never assert or imply that money caused a vote, changed a mind, or bought an outcome — the official record doesn’t establish causation, and neither will we.

No partisan framing.No editorializing, no loaded language, no benefit/harm framing of any bill, vote, or member. If the numbers embarrass a side, that’s the data, not us.

Every user-facing record links to its official source, so anyone can check our work. When we find — or someone reports — something wrong on the site, we fix it and log what changed. Every correction is classified by materiality:

• Minor— typos, formatting, broken links, and other fixes that don’t change the substance of what was shown.
• Substantive— anything that changed a fact a reader relied on (a vote tally, a party label, a bill summary’s claim about what a bill does, etc.).

Substantive corrections are published on the public corrections record: the affected page, what it said, what it says now, why, and when it was discovered and fixed. That record is distinct from (though a correction may reference) our internal ingestion audit trail, which logs every detected upstream data change for review — the corrections record is only for changes that reached the public site incorrectly.

Spot something wrong? Email corrections@civicroll.org.

Input source:a summary is generated only from the bill’s official Congress.gov summary text already in our database — never the full bill text, and never any other source. The exact source text is hashed and stored with the summary.

Validation:if the bill’s official summary text changes after a summary was generated, the stored source hash no longer matches and the existing summary is treated as stale — it cannot be approved and must be regenerated against the current text before it can go live.

Human rubric: every summary is held as pending_review until a human reviewer checks it against the full official source text for: accuracy, completeness on major provisions, zero benefit/harm framing, and that it matches the correct bill version. Only summaries that pass go live; the rest stay pending or are rejected.

Provenance:every summary row stores which model generated it, a hash of the exact prompt template used, the source text hash, and the bill version it was generated from — so any summary’s origin can be reconstructed and re-checked. The site only ever renders the latest live row for a bill; if none exists yet, the summary section is simply absent, never a placeholder.

CivicRoll is informational: it shows you the official record so you can draw your own conclusions. It is not, and is not intended to be, voting advice, legal advice, or a recommendation to act any particular way. This framing is provisional, pending review by counsel— the wording here may change as that review completes, but the underlying principle (we show, we never tell) will not.