Hidden Levers is a nonpartisan record of money in American politics, enhanced with public data sources and advanced analytics, to provide a public database that links donors, committees and candidates by the dollars they contribute and spend.
Our source material is already public. Campaign-finance disclosure laws make it easy for any American to access this information themselves. Yet despite being out in the open, the information is difficult to consume and understand.
Thus was the inspiration for the database. The goal was to take the divided pieces and map them out. Every entity we know about — a person, a company, a PAC, a candidate — became a node. Every disclosed transaction between them an edge. To our knowledge, no one has structured this data in this manner, and through the power of proprietary AI models, cleaned that information into a foundation with endless possibilities for future expansion.
Hidden Levers is not a tool for personal or political attacks. While it uncovers associations between benefactor and beneficiary, Hidden Levers does not make causal claims based on these connections. The database is built to serve journalists, researchers, and voters across the political spectrum. We believe that civic engagement in all its forms, including political spending, is a valid way to participate in democracy. That said, transparency and accountability are necessary to avoid corruption and deception.
Hidden Levers is not a government site. It is not affiliated with, endorsed by or acting for the Federal Election Commission, or any other government agency whose records it republishes.
The site is a serving platform for the product of a five-stage pipeline, outlined below. Each stage is repeatable and reviewable: resolution decisions are recorded, so the same filings and the same recorded decisions produce the same database.
Bulk disclosure data is pulled from the FEC — candidate, committee, and contribution filings, plus the itemized transactions behind them. Filings are versioned on the way in, so an amended report supersedes its original without erasing that it existed.
Filings are distilled into a single canonical record for each person, committee and vendor based on a fine-tuned entity resolution predictive model. Each proposed match carries a confidence score; matches below our threshold are left as separate records rather than merged until greater confidence can be ascribed to the decision.
Resolved entities are loaded into the web-like data structure as data points called nodes. These nodes are then connected to separately created politician and committee nodes via the dollars circulating between them and their public affiliations with one another. These connections are called edges. This structure allows us to integrate hundreds of unique data sources while maintaining a schema that reflects the world as most of us understand it through associations and relationships, rather than a spreadsheet tab.
Analytical data and visualizations such as funding-mix breakdowns, percentile ranks and similarity are developed and added to the database post-hoc. Each is recomputed when the underlying data is republished, keeping the picture of influence up to date.
Finished results land in a read-only serving layer the public site reads from. Refreshes replace a published set wholesale rather than patching it in place, so every page reads from one internally consistent published snapshot.
While we believe that our research reflects the cleanest and most comprehensive data on political spending ever made publicly available, it's important to understand that data manipulation at this scale and depth will result in errors on occasion. If you find an issue or a mistake, please email corrections@hiddenlevers.org so we can fix it. We always aim for continuous improvement.