Methodology

We ingest transaction-level spending from official government open-data portals — primarily Socrata (SODA) “open checkbook” datasets, plus OpenGov and published CSVs. We prefer official APIs over scraping and respect each portal’s terms and rate limits. Every record links back to its official source with the timestamp we retrieved it.

Source schemas vary widely, so we map each portal’s columns to a common set of fields and coerce values consistently: amounts (including refunds shown as negatives), dates, fiscal years, departments, and categories. We keep the original row for audit.

The same company often appears under many spellings (“ABC Inc”, “A.B.C. Incorporated”, “ABC INC.”). We group these into a single vendor entity using conservative matching that favors precision — we would rather leave two records unmerged than wrongly combine two different companies. Where a match is uncertain it is held for human review, and each vendor shows a match-confidence figure.

Figures labeled “analysis” are calculated by Govpurse and are not official designations:

• Vendor concentration — a vendor’s share of spending within a scope, and the Herfindahl–Hirschman Index. High concentration is a prompt to look closer, not a conclusion.
• Spending spike — a month or category whose spend is anomalously high versus its own trailing baseline (it must clear both a statistical and an absolute bar).
• Repeat sole-source — vendors repeatedly paid via sole-source / no-bid procurement, only where the source data marks procurement method.

We never assert wrongdoing. Every analytic links to the underlying transactions so you can verify the math and read the records yourself.

Found an error or a mismatched vendor? We want to fix it. Use the contact path in the footer and include the jurisdiction and record. Corrections to source data should also be raised with the government that published it; we mirror the official record.