About
Public data, read correctly
Every federal contract award is public record. What the government paid, to whom, for what work, in what place — all of it is published, and anyone can download it. Almost nobody can read it.
The records confuse obligations with award values, report ceilings that will never be spent, and scatter one piece of work across code systems that disagree with each other. Getting a defensible answer to the question a bidder actually has — what has work like mine gone for? — takes weeks of cleanup, or a subscription that starts around $15,000 a year.
BidBenchmark exists to close that gap. We ingest the government's own award records, classify them into categories a bidder would recognize, and publish the distribution of what comparable work was actually awarded for. The benchmarks are free to read. Paid plans cover the work that runs on your behalf — alerts, exports, and solicitation analysis.
What we hold ourselves to
- No number without a source
- Every published figure carries its sample size, its as-of date, and the definition of what was measured. Every award behind it links to the government record it was read from.
- No model writes a statistic
- Machine learning helps classify awards into categories. The statistics themselves — medians, percentiles, exclusions — are computed by versioned code, deterministically, so the same inputs always produce the same published number.
- Thin data is said, not smoothed
- When too few comparable awards exist to support a figure, we withhold it and say why, rather than publishing a confident-looking number built on a handful of data points.
Go deeper
The full method — inclusion rules, outlier handling, sample thresholds — is published on the methodology page. Where each record comes from and when it was last refreshed is on the sources page. And if something here doesn't hold up to your reading, tell us — the fastest way to keep the numbers honest is to be told when they aren't.