Methodology
Perch doesn’t score legislators. It computes specific, cited measures from the public record — and shows you the receipts for every one.
What Perch reads
Four sources, all public.
Committee hearings. Every statement a member makes in a hearing, matched to the person who said it and tagged for topic and sentiment. Matching a spoken remark to a resolved legislator identity is the hard part — and it’s the part Perch is built to do.
Floor votes. The recorded roll-call votes: how each member actually voted, not how they were expected to.
The public record beyond the hearing room. Official press releases in a member’s own words, and verbatim quotes attributed to them in news coverage — carried separately from hearing testimony, and cited the same way.
Bill authorship. Who filed and carried which bills, pinned to the right session.
Where these agree, that’s signal. Where they diverge — warm words, opposing votes — Perch shows the gap instead of smoothing it over.
What Perch measures
Named techniques from political-science and NLP research — not house metrics. The overall read follows Leadership Trait Analysis (Hermann), the framework for profiling leaders from their own speech.
Questioning style. Is a member asking to understand, to challenge, or to steer the room? Perch reads the function of their questions across every hearing — and shows the questions behind the label.
Basis: rhetorical role of questions — Zhang et al., EMNLP 2017; message politics in committee hearings — Park, J. of Politics 2021.
Consistency. Does what a member says match how they vote? Perch pairs spoken positions against recorded votes on the same subjects, and flags where they part ways.
Advocacy depth. Mentioning a topic once isn’t the same as carrying it hearing after hearing. Perch measures how sustained a member’s engagement really is.
Alignment. When a group or agency testifies, does a member move with them or against them? Perch reads alignment across tens of thousands of member-witness moments in the record.
Voice. Every member has a way of speaking. Perch learns it from the language itself — which is how it surfaces the members who approach an issue most alike.
Basis: word-embedding text scaling — Rheault & Cochrane, Political Analysis 2020. Embeddings: OpenAI text-embedding-3-small (1,536-d).
What they champion. Straight from the filing record: the bills a member actually chooses to author, and the subjects where their bills become law. The most concrete measure of priorities there is.
How Perch keeps it honest
Ranges, not false precision. Perch reports directions — “consistently aligned,” “high probing” — not decimals that imply more than the record supports. Thin sample, and it says so.
Every measure is cited. Nothing appears without the statements or votes it came from, one click away.
Neutral by design. “Challenger” isn’t an insult; “facilitator” isn’t praise. Perch describes how a member works, not whether they’re good at it.
Role-adjusted. A chair speaks far more than a freshman by design. Perch accounts for role so activity isn’t mistaken for conviction.
Speech and votes stay separate. Perch never treats a statement as a vote or the reverse. When they line up, that’s the consistency measure — computed independently.
Silence is silence. No attributed statements on a topic? Perch says exactly that. It never reads silence as opposition.
Perch turns the public record into measures you can check. It won’t tell you whether a legislator is good or bad. It’ll tell you how they question, how their words track their votes, what they sustain, who they align with, how they sound, and what they carry — with the receipts for all of it.
Sources: Texas committee hearing transcripts, legislative floor votes, official press releases and attributed news quotes, and bill filing records. Hearing transcripts are drawn from the official Texas hearing video record.
See the measures on a member.
Request access to Perch