Methodology

How records move from filing to checkable claim

This page describes process, not performance. It does not publish accuracy rates, influence scores, or undated graph totals.

Claim (all four travel together)

What we show you is what we ingested. We can prove when we ingested it. We can prove we have not altered it since. We cannot prove the filer told the truth.

Coverage dates by source

Coverage dates vary by source. The graph is built from federal filings on an ingestion schedule, not in real time, and some series are months behind their sources. We publish what we ingested and when.

Table measured 2026-08-05 (UTC) against the live graph and on-disk ingest artifacts. This is a fixed snapshot, not a live query. Loading this page does not remeasure the graph. Days-behind figures use that measurement date as the anchor.

Grain tells you what the date means: day is a calendar date on the record; cycle or fiscal year is coarser; write-time is when the node was written, not a federal filing date; none means no usable date field was present. Newest record dates exclude impossible future values (for example, transaction dates after the measurement day).

SourcePopulationNewest recordGrainNewest ingested_atBoundary
FEC contributions1,311,097 edgesElection cycle 2026 only. No day-level contribution date on these records. Cycle year is not a calendar-day freshness claim.cycle (year)2026-08-03Coarse grain. A cycle of 2026 means some 2026-cycle bulk is present, not that coverage runs through any specific day in 2026.
FEC independent expenditures60 nodes2025-01-06 (expenditure date)day2026-03-01576 days behind measurement date on record date. Small population (n=60).
FARA agents7,022 nodes2026-02-12 (registration date)dayNot present on these nodes174 days behind measurement date on registration date.
FARA foreign principals12,482 nodesNo date property on these nodesnoneNot present on these nodesNo usable record date or ingest stamp to publish as a coverage date.
SEC Form 4 (insider trades)7,158 edges2026-02-26 (transaction date)day2026-02-27160 days behind measurement date on record date. Form 4 is a continuous two-business-day filing regime.
SEC enforcement actions1,475 nodes2026-02-12 (node write time, field created)write-time2026-02-12174 days behind measurement date. Date is write-time, not a separate federal filing date field.
Lobbying disclosures325,653 edgesMax non-null year is 2024, on 5 edges only. 325,618 edges have year unset (null).year (sparse)Not present on these edgesToo coarse and too incomplete for a single useful coverage date. Do not read as coverage through 2024-12-31.
Lobbyist registry nodes47,793 nodes2026-02-14 (node write time, field created)write-time2026-02-14172 days behind measurement date. Write-time is not a disclosure filing date.
USAspending federal contracts32,362,226 nodes (as of measurement)2026-03-01 (award start date, dates after measurement day excluded)dayNot present on the measured path157 days behind measurement date on start date. Population count is the series size at measurement, not a promotional total.
USAspending awards (limited subset)15,862 nodes2025-09-30 (award start date)day2026-08-04Subset population only; not a replacement for the full contracts series above. Record start dates lag; ingest stamp is 2026-08-04.
Earmarks / community project funding16,764 nodesFiscal year 2026 (8,527 of 16,764 rows)fiscal year2026-02-14Fiscal-year grain only. FY2026 means appropriations year labels are present, not day-level coverage through the measurement date.
STOCK Act PTR (House)21,267 active edges (House source tag)2026-07-20 (transaction date; three future-garbage dates excluded)day2026-07-2916 days behind measurement date on transaction date.
STOCK Act PTR (Senate)7,960 active edges (Senate source tag)2021-02-16 (transaction date)dayNot present on these edgesYears behind measurement date (newest transaction 2021-02-16).
Congressional travel disclosures1,778 nodesNo start or end date on measured nodesnone on record2026-02-11 (ingest stamp field)175 days behind measurement date on ingest stamp. No usable trip date field in the measured population.
USAID contracts14,446 nodes2026-02-10 (contract start date; far-future end dates excluded)dayNot present on these nodes176 days behind measurement date on start date.
Filed disclosure (single relationship present)1 edgePeriod text only; no day-resolution fieldperiod text2026-04-14113 days behind measurement date on ingest stamp. Population is a single edge.

Snapshot only. Figures above are the census values as measured on 2026-08-05 UTC. They are not rounded for comfort.

Pipeline stages

Ingest
Resolve
Graph
Bind
Verify

Bind means attaching source identifiers, ingest time, and integrity hashes. Verify means a reader can leave Sentinel for the federal primary source when an id exists. No stage is marketed as a prediction score.

Federal source families we draw from

Listing a family here does not assert complete coverage of that family on any date. Measured coverage dates for series we have in the graph are in the table above.

House PTR disclosures

Periodic Transaction Reports filed by House members and covered staff

Senate eFD

Electronic financial disclosures for senators

FEC bulk data

Campaign contribution and expenditure records from the Federal Election Commission

Congressional records

Legislative history, committee assignments, and related public records

LDA lobbying filings

Registered lobbyist activity and client relationships

FARA filings

Foreign Agent Registration Act disclosures

Federal court dockets

Public federal case filings where used as a cited source

DOJ public releases

Department of Justice press releases and enforcement announcements

IRS Form 990

Public nonprofit filings when cited as a source

Federal spending records

Public award and assistance records when cited as a source

What we publish about verification

Primary-source ids

Answers that can print a federal transaction or disclosure id do so, so a reader can verify without trusting Sentinel.

Ingest and integrity

Anchoring and hashing prove a stored artifact existed at a time and has not been altered since. They do not prove the underlying filing is accurate.

Dual computation when used

Some statistics are recomputed on independent paths sharing only the primary government source. When that happens, the write-up says so with the date of the run.

What is not published

Undated node or edge counts, retired classifier AUCs, and influence score leaderboards are not paper of record.

Back to site claim →