What's in it, what isn't, and how we check

1 · What's in

Taxpayer-funded spending by Senate member offices, committees, leadership and officers, from the Report of the Secretary of the Senate (semiannual, 30 reports since April 2011), and by House offices from the Statement of Disbursements (quarterly CSVs since [placeholder]). Campaign and leadership-PAC spending, foreign delegation travel, privately sponsored travel and personal financial disclosures are separate records and come later.

Each parsed row keeps the PDF page index, the printed page label (B-661) and the bounding box of the text it was read from. Money is stored as integer cents with the sign as printed.

2 · What the Senate doesn't itemize Total only

The Senate prints line items for exactly three categories: travel, other contractual services and acquisition of assets. Payroll is printed as an abstract (name, title, dates, amount). Rent, supplies, printing, communications and utilities appear as a single total per office, with nothing beneath it.

So on this site a category stamped Total only has no rows to show, and a blank list under it means not visible, never zero. In the latest report, [placeholder] of Senate-wide spending sits in those three categories with no rows beneath it. The House CSVs itemize everything, which is one reason the two chambers are not directly comparable.

3 · How we check Inspected

No report is published until it reconciles to the totals it prints. For every office block, seven hard checks run:

Two soft checks are recorded and never fail a block: the printed net-payroll line vs. the summary (adjustment timing, per the report's own notes), and a summary showing payroll with no abstract printed (interns, child-care contributions, reserves). Whole-report counts (blocks, lines, items, entries) are also compared to an independent survey of the PDF.

4 · Status by report

A report is published when every failing check has a verdict: a discrepancy printed in the source, a parser defect (fixed before publishing), or unclassified (which blocks publication). The bar is zero unexplained blocks.

5 · Printed discrepancies, latest report

5 of 1,161 blocks have printed discrepancies.

In every case the PDF prints the numbers we read, confirmed by a second script that does not use the parser; these are discrepancies in the report, not in the parse.

6 · Staff names Name withheld

The Senate prints every staffer's name beside their title and pay. We publish titles, service dates and office totals. Names are kept internally for analysis (turnover, moves between offices, later lobbying registrations) and appear only where a confirmed lead depends on one, after the office has been asked for comment. Travel reimbursements to staff show as "Staff" for the same reason. The names are one click away in the original PDF; we don't pretend otherwise, we just don't index them.

7 · Where a model is involved Model · 0.97

Never for a dollar amount. Models suggest finer subcategories for free-text descriptions ("STAFF TRANSPORTATION SEATTLE TO WASHINGTON DC" → airfare), cluster vendor spellings, and propose office-to-member matches. Every such label is stored with the model, prompt version and confidence, shown beside the original text, and never used in a total. Fuzzy vendor merges and name matches stay suggestions until a person reviews them.

Detectors (peer outliers, reversals, trip reconstruction, calendar patterns) run across all offices and report base rates, so no one is singled out by how the search was aimed. A lead is a reason to look, not a finding; only confirmed leads are published, with their rows.

8 · Known issues

9 · Changelog