This page is the technical companion to the About page. Those who are interested in becoming partners on this project should write to hello@hiddenlevers.org.
Every figure on the site derives from a public filing or an openly licensed dataset. Each source, the licence it is published under, and the attribution that licence requires are listed in full on the sources and attributions page.
The primary source is Federal Election Commission bulk data: candidate, committee, and contribution filings together with the itemized transactions behind them. An amended report is a full restatement of the original, and both arrive in the bulk data. We keep the original rather than deleting it, and filings that a later amendment has superseded are marked as such, so that a figure is drawn from the latest filing of a report rather than from both.
Race ratings are third-party editorial calls by the Cook Political Report, shown with their date. We carry one on every 2026 Senate race Cook rated Toss Up, Lean D or Lean R on its public ratings page dated Aug 20, 2026 (accessed Sept 2, 2026), plus the three-way Montana Senate race (Cook: Solid R).
Hidden Levers is independent. It is not affiliated with, endorsed by or acting for the Federal Election Commission, or any other government agency whose records it republishes.
Campaign-finance filings identify people by whatever string the filer typed. The same donor appears as several different names across several filings, and two different people frequently share one. Resolution is the step that reconciles those strings into a single canonical record per person, committee and vendor. A predictive model scores each proposed match, on name forms, employer and occupation fields, and address and committee co-occurrence, and the decisions taken on those proposals are recorded, so the same filings and the same recorded decisions produce the same database. It is also the step where a mistake becomes a claim about a named individual. That said, our models have produced an F1 score of 99.45% in closed testing. Continuous analysis seeks to eliminate that remaining 0.55% of possible errors.
Every source string stays attached to the entity it was resolved into, so any merge can be re-examined rather than having to be taken on trust. Matches below our confidence threshold are left as separate records rather than merged. That choice is deliberate and asymmetric: a missed link understates somebody's giving, which is a gap in the data, while a wrong link attributes a contribution to a person who did not make it, which is a false statement about them. We would rather have the gap. Our data should be verified independently by the user before any conclusions or publication is made.
Only disclosed money is in the database. Spending that federal law does not require to be itemized — small-dollar aggregates below the reporting threshold, and giving routed through organizations that do not disclose their donors — is outside the record and therefore outside this database. Where a page shows a total, it is a total of what was disclosed, which is a floor rather than the full picture.
2026 candidate and committee totals run through mid-October. Itemized donor gifts are as processed by the FEC, largely through mid-2026.
There are two different kinds of error here, and they have different remedies.
The filing itself is wrong. We publish what was disclosed. If a committee misreported a contribution, the authoritative fix is an amended filing made by that committee with the Federal Election Commission; once the amendment is filed and a data refresh has loaded it, it supersedes the original here. We do not edit filing data to match a correction we have been told about but cannot verify against the record.
Our processing is wrong. If a contribution has been attributed to the wrong person or group, two entities have been merged that are not the same actor, or a derived figure is miscomputed, that is our error and we fix it. Corrections are made as they're identified and the affected records are recomputed, so the fix holds through the next refresh instead of being undone by it.
To report either kind, email corrections@hiddenlevers.org (it is linked in the footer below). A report that identifies the specific page and record is the fastest route to a fix.
Contribution records published here come from federal disclosure filings and are public records. Being named in one does not make a person a Hidden Levers account holder, and the account-data rights described in the privacy policy do not extend to those records. Corrections to a filing are made with the FEC by the filer; corrections to our handling of a filing are made by us, through the same address: corrections@hiddenlevers.org.