Methodology
Every figure on GovSpend traces to a public source and a documented computation. Nothing here is modeled behind closed doors.
Data sources
Award records, recipient totals, quarterly agency series. No auth required. Pulled Oct 9, 2026.
Reference for the historical series behind forecasts and win-rate denominators (FY2022 onward).
230k+ awarded hourly labor-rate records. No key required. Pulled Oct 9, 2026.
Checked at build time. The public extracts are multi-gigabyte and the API is key-gated, so v1 uses USASpending recipient records for vendor identity. SAM resolution is a v2 item.
What was computed
- Leaderboard: recipients ranked by obligated dollars, Oct 2025 to Sep 2026, via the spending_by_category endpoint, contracts only (award types A, B, C, D).
- Award counts: the 4,000 largest awards of the period were sampled and counted per recipient. This undercounts small awards and is labeled as an approximation everywhere it appears.
- Win rates: approximated as awards per vendor relative to the sample. Bid-level data is not public, so true win rates cannot be computed. The approximation method is stated next to every figure.
- Forecasts: ordinary least squares linear trend on 8 fiscal quarters of obligated dollars per agency, projected 2 quarters ahead.
- Recompetes: contracts awarded or modified in the trailing 12 months whose period of performance ends between Oct 9, 2026 and Oct 9, 2027, ranked by award amount.
- Labor benchmarks: percentiles of CALC+ ceiling rates per labor category and experience bucket (0-2, 3-5, 6-10, 10+ years).
What was left out
- Assistance awards (grants, loans): the leaderboard covers contracts only.
- Classified and unreported spending: DOD notes national-security reporting limits on its USASpending profile.
- Subawards: only prime awards are counted.
Reproduce it
The ingest script lives at scripts/ingest.py in the repo. It needs only Python 3 and network access. The committed snapshot in data/snapshot.json is exactly what the script wrote on Oct 9, 2026.