Campaign Commonsreceipts, not conclusions
United States Senate seal

US Senate 2024Pennsylvania

Election held ·

Notable candidates

Democratic Party

Ask a plain-English money question about this race. A language model writes a read-only query over the filings graph, and every returned row links to where it was filed. Matching precomputed pages are offered as related links. Methodology →

Start with @graph to draw the answer as a flow diagram.

How the question is read. Questions are answered by a language model that writes one read-only query over the filings graph; the site validates and runs it and shows every returned row with its filing source, and the model's summary is withheld unless every number in it is in those rows. When a precomputed page matches the question it is linked as a related page. Where the graph is not connected, the browser matcher picks a page instead.

On issue pages the groups' stated positions were read from their own sites by a model offline and kept only where the quote appears verbatim on the page; each is labelled verified or unverified.

How the answers were built. Answers are precomputed from this race's ledger, ads, chain and entity files by pipeline/campaign_commons/trails.py; no language model or graph database is involved in building them. Three questions are supported: who is spending for/against a candidate (Schedule E), who paid for the ads about a candidate (ad sponsors, their Schedule E about the candidate, and each sponsor's own funders), and who funds a committee (its receipts hop by hop). Every figure links to the FEC or Google record it was read from. Money edges and targeting edges are separate; an independent expenditure moves no money to a candidate, and a funder is never joined to a particular ad because pooled dollars cannot be allocated that way.