VISIBLEGOV

How this works

This page explains, in plain terms, how a fact gets from an official government document onto this site — and where to look when something is disputed or missing.

From official record to this page

Every fact on this site starts as a document published by an official source: LegiScan's bulk data service for Missouri House and Senate roll calls, cross-checked wherever a matching record exists against the Missouri Senate's own journals and the house.mo.gov XML data service (which serves only the current General Assembly), and Congress.gov, the Clerk of the U.S. House, and the U.S. Senate's Legislative Information System for federal bills and votes. Nothing on this site is entered by hand from a news story, a press release, or anyone's recollection of what happened.

Before an individual legislator's recorded vote ("cast") on a Missouri roll call is stored, the count of casts for each position — yea, nay, not voting, absent — must match the tally the source itself printed for that roll call. If the counts don't match, nothing is quietly corrected or averaged to make them fit: the mismatched roll call is held back, or, when two independent sources genuinely disagree about the same roll call, it is flagged and kept open until it is resolved as described under "Reconciliation flags and quarantined data" below. A cast is never invented to make a tally add up. Federal roll calls are stored cast by cast as the Clerk of the U.S. House's and the U.S. Senate's own roll-call files list them (a cast whose member cannot be matched to this site's roster is counted as unresolved and left out, never guessed); this site does not currently compare those casts against the totals printed in the same file, so that check does not apply to federal votes.

A bill's plain-language summary is authored only from that bill's own official text — never from a model's general knowledge of the subject, and never from news coverage or commentary about the bill. Before a summary publishes, a second, independent model reads the same official text and tries to find anything the summary claims that the text does not actually support — an adversarial check, not a second opinion that simply agrees. Summaries have been produced on two lanes, and the stored record of every summary names the models used: on one, claude-opus-5 authors and claude-sonnet-5 verifies; on the other — budget-capped API runs, today a small daily batch under a fixed weekly spending cap — claude-haiku-4-5 authors and claude-sonnet-5 verifies. The checks and publication rules below are the same on both.

A summary publishes automatically only after it clears mechanical completeness checks — every required field present, the stated purpose verified as a verbatim excerpt of the bill's own text, and every point carrying a citation that matches the section the verifier actually checked — and the adversarial verification itself comes back clean: a second model reading each point against its cited section and trying to refute it, never simply agreeing. Neither model's own report is trusted directly; the completeness checks and the "clean" result are both re-derived mechanically in code from what the models actually produced, not read off a claim either model makes about itself. (Ten summaries published in July 2026, before the completeness check existed, were verified on the adversarial support check only and have not been re-checked as of 2026-08-23.) Anything the verification flags is never published as it stands: the flagged draft is held back, unpublished, and the bill may be re-authored with the verifier's specific objections as instructions, publishing only if a fresh adversarial check on the new draft comes back clean; the draft it replaces is then kept in the record as a rejected attempt, never deleted. A bill that has already had two drafts rejected, or whose text breaks the checks structurally, is not re-authored automatically: its draft is held, unpublished, for this project's maintainer to review. A bill page with no published summary says so plainly; a draft that has not cleared verification is never shown.

Every published summary point and every glossary entry carries its own receipt: the quoted excerpt it was drawn from, and a link to the official source wherever one is on file. For a summary, that excerpt is the passage the checking model relied on; for a glossary entry, it is the passage the authoring model quoted, mechanically checked to be a verbatim substring of the source text before it can be stored. The link goes to the official source so you can confirm the wording yourself, which is the check that actually settles it. Every glossary entry has that link. Bills and votes link to the official document wherever the source supplied one; as of 2026-08-23 a few hundred Missouri roll-call rows from the house.mo.gov feed and a few dozen bills with a published summary have no source link on file, and those pages show no link rather than a guessed one. Nothing here asks you to trust a rating or a label in place of the source.

Reconciliation flags and quarantined data

A reconciliation flag means two of this project's independent data sources could not be confirmed against each other for a roll call — for example, LegiScan's tally and the Missouri Senate's own printed journal disagree about how a specific legislator voted, or about the final count; or one source's own cast list is too incomplete to compare; or LegiScan records a roll call for which the Missouri source has no record whose tallies correspond to it. When that happens, the disagreement is not resolved by picking whichever number looks right, or by averaging the two. It is recorded as an open flag. A flag is closed by a person reading both sources and deciding, or — in one mechanical case only, when the missing Missouri record later arrives and matches LegiScan's tallies exactly — by recording that match inside the flag itself; a flag is never deleted. (As of 2026-08-23 no flag has yet been closed mechanically, and the Missouri Senate journal days affected by a since-fixed parsing fault are being re-read; flags raised while that fault stood stay open until they are closed as above.) Any vote shown on this site carrying a "⚑ reconciliation flag" badge links back to this page.

While a flag is open, this site does not blend or split the difference between the disagreeing sources, and it does not hold either record back on the strength of the disagreement alone: both records are stored and shown, each exactly as its source printed it — the Missouri source's record carries the badge and the LegiScan record is listed beside it unchanged; nothing is merged. That holds whether the two sources print contradictory yea or nay totals, differ only on who was absent or not voting, or cannot be paired at all. A LegiScan record is hidden in one case only: when it has been paired with a Missouri record as the same roll call — mechanically, when a later pass of the same check finds their tallies agree, or by a person closing a flag and recording the pairing. Then the Missouri source's own record is the one displayed and the LegiScan copy is kept but hidden from public pages, marked as its duplicate so the pairing stays traceable; neither row is deleted. This is primary-source-first display: the badge is the reader's signal that a second source could not be confirmed against the first, so you can go compare both original documents yourself, rather than trust a number this site invented to reconcile them.

Missouri Senate roll-call detail — which legislator cast which vote — comes from the Senate's journal PDFs, whose text has to be mechanically extracted and reassembled before it can be checked against LegiScan's own tallies. When a given journal day's extraction cannot be automatically confirmed against the printed tally (a parsing failure, an unreadable page, a mismatch that cannot be resolved without a person looking at the source), that day is quarantined: held out of the published record rather than guessed at, and counted honestly as a day "pending human review," never silently dropped and never silently published as if it had been confirmed. The data-freshness table below shows the current quarantined-day count for every source that has one.

The same discipline runs in both directions: a fact is never invented to fill a gap, and a fact is never suppressed just because it is inconvenient. A dispute or a gap is shown as a dispute or a gap, with a path back to the primary documents so it can be checked directly.

What the vote-aggregate numbers mean

The charts on a legislator's page (position distribution, party comparison, party-line cohesion) are numbers only — no grades, no adjectives, no ranked lists of politicians. The exact definition of each figure, in the same wording the charts themselves are built from, is rendered below.

Data freshness

What is ingested, from where, and through when — the same table a legislator's profile page draws its freshness box from, shown here in full for every source this project tracks.

What's in the database right now

These five counts are fetched live from the same database the site reads — not a cached snapshot, not a rounded estimate.

What is not here, so a gap is not mistaken for a record: Missouri roll-call coverage begins with the 2015 session (Missouri Senate roll calls with 2016) and federal coverage with the 119th Congress (January 2025); earlier years are not on this site. The Missouri Senate's own journals are being read for the current session and, among earlier years, were sampled only for 2018, so other years' Senate roll calls rest on LegiScan's record alone. Missouri House roll calls from 2024 are only partly here: the source prints no date for any of them, and this site does not store a roll call under an invented date. Where the same source's own record of the bill prints that vote's exact yea-nay count beside a dated action, and only one such action exists, the vote is stored under that date and carries the action it was taken from as its receipt; that covers about two-thirds of them (185 of 275, as of 2026-08-23). The rest — where no dated action prints that count, or two votes on the same bill share one — are not shown at all, rather than shown under a guess. A term shown as ending in “present” means the source recorded a start date and no end date. That is what the record says, not a separate confirmation that the person still holds the seat: of the 731 open-ended terms on this site, 729 belong to someone who cast a vote in 2026, and 2 do not. Where a term’s start is missing, it says so plainly instead — an unknown start is not an open end, and is never shown as current. Not every bill has a summary — as of 2026-08-23 roughly one in four bills with a recorded passage vote has a published one; a bill page without a summary says so and shows the record it does have.

Attribution

This site republishes and summarizes records published by the sources below. The license notices are stated because their terms require it, and because anything on this site should be traceable back to where it actually came from.