Published before the pilot
Methodology
The badge
“Verified constituent · TX HD-NN · government-ID + residence check” — one badge, one meaning. It says exactly what was checked: a government ID (document + selfie, via Stripe Identity) and a residence check placing the member in the district. It does not check voter registration: Voters Lobby verifies residence, so a verified constituent may or may not be a registered voter, and no count on this site says otherwise. The badge is earned only through verification — there is no other way to get it. Most members verify automatically; a small share resolve a district mismatch by showing a utility bill or lease to the founder (basis: manual review). The mix of verification bases is disclosed on every Certified Brief.
Verification lapses after 12 months without re-attestation; lapsed members drop out of new rollup counts until they re-attest. Frozen historical tallies are unchanged. District maps are versioned; redistricting forces re-resolution.
How we count
Every tally is a count of verified, district-locked members — nothing is weighted, extrapolated, or modeled. Percentages appear only at 25 or more verified positions; from 5 to 24 we show raw counts. Under 5 verified positions we publish the verified n and withhold the support/oppose split and the per-channel counts entirely — at that size a split describes a handful of identifiable people rather than a district, and carries nothing a reader can use. Every tally we do publish carries its recruitment-channel mix (organic / partner-organization / founder) because how people arrived shapes what a count means. We never say the district as a whole supports or opposes anything: our numbers describe verified members only, and we publish exactly that.
When a bill is amended, positions already recorded carry forward and keep counting. The alternative — voiding them and asking everyone to start again — empties the tally in the week before the floor vote, which is exactly when it needs to be worth citing.
But an amended bill is, in a real sense, a different bill. So every tally also publishes how many of the positions in it were recorded against the text as it now stands, as a count beside the total, never as a rate. A reader who thinks only that narrower number is meaningful can use it; it is printed rather than argued about. Members are told when the text under one of their positions has moved, and recording the position again moves it into that count.
Alignment scores
- Universe: pre-declared record votes (Texas Const. Art. III §12(b) — final passage, concurrence, conference reports) on curated bills only. Committee voice votes are tracked but never scored.
- Score: how often the legislator voted the same way as the majority of verified members in the district who had a position at the time of the vote. It is an alignment between a legislator and the verified constituents we heard from, never a statement about what the district thinks. Positions taken after a vote occurred are excluded from that vote’s count. Events with a tied position split are excluded from the denominator (shown, labeled tied).
- Vote day: LegiScan reports a roll call’s date, not the hour it was taken, so we draw the line at the start of the vote date in Central time. A position recorded on the vote date or later is treated as made after the vote, even when it was in fact recorded that morning, hours before the gavel. The alternative would count positions recorded in response to a result already known. We would rather leave out a position that came before the vote than count one that came after it.
- Absences are dropped from the denominator and displayed prominently — an absence is neither agreement nor disagreement, and hiding absences flatters absentees.
- Percentages at 25 or more verified in-district positions; raw counts from 5 to 24; under 5 the split and the per-channel counts are withheld. The recruitment-channel mix of the position sample is published with each scored event and each score.
- A scored event's position sample follows the same floor: under 5 verified positions we publish the count and withhold the support and oppose split, and we do not label the event tied — at that size “tied” beside the count is the split. This is a publication rule only. The event still counts toward every alignment score exactly as it did, and every verdict drawn from it stands.
- An alignment percentage publishes only when every event behind it carried at least 25 verified positions — the thinnest event governs, which is the conservative reading. Below that the percentage is withheld and the matched, broke-with and absent counts publish instead, alongside every roll call the score is drawn from.
- The retrospective 89R (2025) set is scored identically but labeled retrospective everywhere — those positions were taken after the votes occurred, and we say so.
- Provenance: every scored vote links its journal page, roll call ID, and hand-verification date. A vote is never scored until a human has checked the ingested tally against the official journal.
- Every published score carries its methodology version. Scores are opinions on disclosed facts: the underlying roll calls are always shown.
Which votes count — the rule, published in advance
A site that picks which votes count after seeing how they went is not keeping a record, it is writing one. So the selection rule is mechanical, it is published before the session it applies to, and the parts of it that can be enforced by the database are.
- Only these votes are eligible. Texas Const. Art. III §12(b) record votes on a curated bill: final passage, concurrence in Senate amendments, adoption of conference committee report, in either chamber. Committee voice votes, procedural motions and amendment votes are recorded where we have them and are never scored.
- Every event is declared before the vote happens. The declaration timestamp is written by the server at declaration time and cannot be supplied by hand, and the database refuses to store an event whose vote occurred before it was declared. A vote nobody declared in advance can never be scored afterwards — which is the whole point: no vote can join a legislator’s record because of how it turned out, and none can leave it for the same reason.
- Curation is bounded and dated. The pilot scores three to five bills, chosen before the session opens from pre-filed bills, each published with a brief and the strongest case on both sides, and each carrying the date it was curated. Once a bill is curated, its qualifying votes are declared — including the ones we would rather not publish.
- A declared vote that never happens stays visible. An event that is declared and never reaches a record vote is not deleted; it stands as declared and unvoted. A bill that dies without a vote is a fact about the session, not an absence of one.
- Nothing is scored on our reading alone. An event cannot be marked scored until a person has checked the ingested tally against the official journal — the database enforces that too — and the journal page is published beside the result.
- A tally one organisation assembled does not publish. If 70% or more of the verified positions behind a tally were recruited through a single channel — one partner organisation, or our own founder outreach, which is not exempt — the tally is withheld and so is every verdict drawn from it, and the reason is printed where the numbers would have been. The test does not run below 10 positions, where no rate is published anyway and the channel counts are already shown in full. It runs on every published tally: the current count for a district on a bill, and the vote-time sample behind a scored event alike. When it fires, the support and oppose counts and the recruitment-channel counts are both withheld; the verified count itself still publishes, because it is a count of positions rather than a claim about what anyone thinks.
- That is a proxy, and we would rather say so. The test we would prefer to run is whether our verified members reflect the district’s own partisan balance. Texas has no party registration, and this site deliberately holds no party affiliation for any member — so that number does not exist here, and obtaining it would mean keeping the political record we promise not to keep. Where a district’s recent statewide vote has been transcribed from the official return, it is published beside the count as context: how the district voted, never what it thinks of this bill.
- What this rule does not cover. It governs which votes are scored, not which bills exist. We do not claim the curated set is representative of the session, and a bill missing from this site is missing because nobody curated it — not because it was judged unimportant. Every change to this rule is dated in Corrections.
What a legislative office gets
The same site you are reading. Every figure, brief and roll call a legislative office sees is published at a public address, free, at the same moment: nothing an office receives from us is unavailable to the public, and nothing is held back from the public to be shown privately.
Concretely, that means no private dashboard, no early access to a tally before it is published, no figure broken out for one office and not for the district it represents, and no report of who in a district holds which position. A Certified Brief handed to an office is a public document with a public link and a published hash — anyone can open the same one and check it.
This is a rule, not a description of what we happen to do today, and it is the reason there is no office login on this site. If that ever changes, it can only become an account that signs in to see what is already public — and this page will say so, dated, in Corrections.
Pilot gates — published in advance
We hold ourselves to pre-committed targets, published here before the results exist. If we fail these by the end of the 2027 regular session, that is a public signal to stop or re-scope:
- At least 200 verified in-district members in the seed House district.
- At least one documented legislative-office interaction with a Certified Brief (a meeting, a citation, or a written acknowledgment).
- Month-3 retention of at least 40%: the share of verified members active (any authenticated visit or notification click) in the 30-day window starting 90 days after their own verification date.
- Verification completion of at least 50% from interstitial start.
Covenants
These are limits on us, adopted 19 August 2026 and published in advance. They are absolute on purpose: a hedged promise tells a reader exactly which hedge to expect. If any of them changes, the change is dated in Corrections — this page is never edited quietly.
No money, and nothing else of value, ever reaches a sitting legislator
Not a payment, not a gift, not a favour, not a thing of value of any kind, direct or through anyone acting for us — to a member of the Texas Legislature, to their office, to their staff, or to their campaign. This is not a setting that could be changed later without changing what this is. A verified count is the only thing we ever hand a legislator, and we hand the identical thing to the public at the same moment.
Nothing about a vote is ever said to an office in private
Every number we would show a legislator is already published at a public address that anyone can read for free. There is no private briefing, no early look, no figure prepared for one office and withheld from the district it represents, and no conversation in which a consequence is described to a member out of public view.
A place on this site cannot be obtained by giving us anything
Bills are curated under the rule published above, before the session. No organisation, campaign, or individual can cause a bill to be curated, a tally to be featured, or a position to be counted by giving us money, work, access, or anything else. If that ever appears to have happened, it is a correction, and we will publish it as one.
Member identity is never sold, rented, brokered, or handed over
Not to a campaign, a party, an advocacy organisation, a data vendor, or a researcher. We verify identity and then delete it: no ID images, no address, and no biometric data are kept here. Our verification vendor holds its own session records for roughly twelve months before destruction, which is their retention and not ours, and we say so rather than claiming an immediacy we do not control.
A subpoena to us returns pseudonyms and hashes — and we publish that it arrived
What we hold about a verified member is a district, an attestation record, and a one-way fingerprint of a document number that cannot be reversed. That is all a legal demand can reach, because that is all there is. We publish the fact of every such demand we receive, to the extent the law allows us to.
We never state a rate we cannot honestly state
No percentage below 25 verified positions, no confidence interval, no weighting, no model, and never a sentence claiming that a district as a whole supports or opposes anything. Our counts describe verified members and say so on the same line. Where a number rests on a small or lopsided group, the composition is published beside it rather than left for a reader to discover.
The counts that embarrass us are published exactly like the others
Which votes count is fixed before the session and cannot be changed after a vote. A tally is published whichever way it comes out, on whichever member it lands, of whichever party. The first result that hurts our own case is the one that decides whether any of this was worth building.
Errors are corrected in public, dated, and never quietly
A correction is published within 48 hours of discovery, carries the date it was made, stays on the record permanently, and appears on the page that carried the error. Changes to these covenants are published the same way. We do not edit this page silently.
Anyone may audit us adversarially, and we will answer
Every score is an opinion on facts we disclose: the roll calls, the journal pages, the method, and the version of the method are published with it. If you can show a number here is wrong, we would rather hear it from you than defend it, and the correction will name what you found.
Corrections
Corrections policy: errors are fixed in place with a dated correction notice on the affected page, and the underlying data is republished. Every score and brief carries enough provenance (roll call IDs, journal links, document hashes) for anyone to check our work. Every issued notice is logged permanently on the public corrections log, so the record of what we got wrong outlives the page it appeared on. To report an error, use the contact address on the security page.
Data freshness
Bill data is ingested from the Texas Legislature Online record and the LegiScan API on a schedule, never scraped live. Every bill page shows when each source was last captured. If ingestion stalls, the timestamps say so — we show stale honestly rather than guessing.