CapitolMarkets
Methodology

Editorial standards

How Capitol Markets is built, what role AI plays in it, and the rules every claim on this site is held to.

Source-of-truth rule

Every empirical claim on Capitol Markets — every dollar amount, every alpha number, every committee assignment, every trade date — is traceable to one of these primary sources:

  • The U.S. House Clerk financial-disclosure portal at disclosures-clerk.house.gov
  • The U.S. Senate Ethics Committee EFD portal at efdsearch.senate.gov
  • The Office of Government Ethics 278e filings at oge.gov
  • SEC EDGAR filings at sec.gov/edgar
  • Daily-close price data from public market sources (Yahoo Finance public endpoints)
  • Committee membership rosters from clerk.house.gov and senate.gov

Every politician profile links the underlying PDF or HTML filing for each disclosed trade. If you can’t click through to a primary source from any number on this site, that’s a bug — please email it to corrections@capitolmarkets.org.

How AI is used here

Capitol Markets uses computational tools, including large language models (Anthropic Claude, OpenAI GPT), in three specific roles:

  • Ingest tooling. Parsing PDFs from House Form A filings, extracting structured rows from Senate EFD HTML tables, and resolving fuzzy politician name matches across sources (e.g., joining “A. Mitchell McConnell, Jr.” on a Senate filing to “Mitch McConnell” in the roster). This is data-extraction work — the model doesn’t generate facts, it normalizes them.
  • Editorial drafting. Long-form deep-dives on /news are AI-drafted from structured database queries and human-reviewed before publish. Every empirical claim in those pieces is sourced to the underlying filing or to named published reporting; we link the source inline. AI is not used to generate opinions, speculation, predictions, or characterizations of any individual’s motives.
  • Pipeline automation. AI tooling drafts alert copy and the structured-data scaffolds for social-media posts. The format is deterministic; the numbers come straight from the database; the distribution is human-supervised.

Where AI is not used

  • We do not use AI to generate opinions about specific politicians beyond what their public filings state. Every characterization of a trade — “Smart sell,” “Bad buy,” and so on — is a deterministic function of the disclosed dollar range and the post-trade price move, not a model’s judgment.
  • We do not use AI to predict future trades. STOCK Act data is structurally late (30 to 45 day filing windows), so any “signal” in it is already priced into the market. We report on disclosures; we do not forecast.
  • We do not use AI to fabricate quotes or paraphrase named sources. Every quoted statement on this site is a direct quote pulled from a named source we link.
  • We do not use AI for image generation of any identifiable person. Politician portraits come from official bioguide.congress.gov photos and the politician’s own office. Editorial illustrations for long-form pieces are symbolic, are flagged as such, and never depict any specific individual.

Human review

Every long-form deep-dive published on capitolmarkets.org/news is read end-to-end by a human editor before publish. The editorial-auditor checklist includes:

  • Every numerical claim cross-checked against the linked primary source
  • Every characterization of a politician’s motives, intent, or knowledge structurally framed (“the structural question is”) rather than asserted
  • Every legal-doctrine reference (e.g., Dirks v. SEC) cited to the actual case or statute
  • Disclaimer paragraph confirming that no claim on the page constitutes an accusation of illegal conduct

Short-form posts (alerts, tweets, replies) are generated from the same database under the same source-of-truth rule — every number is DB-sourced — but are not individually editor-reviewed. The deterministic format (politician + role + committee + trade band + ticker) keeps the surface area for editorial error narrow.

What this is not

Capitol Markets is not investment advice. The data on this site, including computed alpha numbers and conflict flags, is presented for journalistic and educational purposes. It does not constitute a recommendation to buy, sell, or hold any security, and the editors take no position on whether the politicians profiled have committed any illegal act. STOCK Act disclosed trades are legal under existing law. The reform debate is whether the law should change — that’s a policy question this site reports on, not a verdict it issues.

Funding and independence

Capitol Markets is independently funded. We do not accept payment, advertising, sponsored placement, or paid content from any:

  • Member of Congress, congressional campaign, or political party
  • Lobbying firm or registered lobbyist
  • Publicly traded company that appears in our database
  • Trade association representing any sector covered by our conflict-flagger

Revenue sources are limited to: paid subscription tier for individual readers, structured-data licensing to newsrooms and research organizations, and (when launched) co-publishing arrangements with credentialed journalism outlets.

Corrections

If you find an error — a misparsed filing, a wrong dollar amount, a wrong committee assignment, a wrong attribution — email corrections@capitolmarkets.org. We correct quickly, log every change on the corrections page, and (for material errors) issue a dated note on the affected article.

A note on this page

This editorial-standards page is itself the result of a 2026-05-12 decision: rather than wait to be asked how we handle AI, document it openly first. The Poynter Institute and the Columbia Journalism Review both recommend AI-using publications disclose their methodology in public; this page is our compliance with that recommendation. If our methodology changes, this page is the canonical place we’ll document the change, with a dated revision history below.

Revision history

  • 2026-05-12 — Initial publication.