How this is built
If you are about to plan a bid around something you read here, you should know where it came from and how often it is wrong. This page is that, including the parts that are not flattering.
Where the records come from
Governments publish their agendas, legislation and contracts through a handful of systems — Legistar, PrimeGov and CivicClerk. This index reads those systems directly, through the same interfaces the governments publish them on. Nothing is retyped, nothing is paraphrased, and no record is edited. Every page here links back to the government's own document, and that document is the authority, not this site.
The index currently covers 1,531 governments and refreshes daily. The homepage shows when it last ran and how many meetings are happening today, so you can check that claim against your own city's calendar rather than take it on trust.
How contract facts are extracted
A record's title, and sometimes the document behind it, is read by a program looking for the counterparty, the value and the term. This is pattern matching, not judgement, and it is right often enough to be useful and wrong often enough that the source link matters.
End dates carry two labels, and both are shown on every record:
- Quoted — the government wrote the end date down and it was read back out.
- Calculated — the government stated a term ("a three-year agreement") and the end was worked out from it and the award date. That is an inference, and it is labelled as one.
Where the text was read from is shown too: the agenda line, the item page, or the signed agreement filed with it. A date read off an executed contract is a stronger claim than one read off a one-line agenda entry, and the page says which you are looking at.
What this cannot tell you
Across the whole index, about 20% of contract records carry an end date. The rest do not, because the government never published one — and no index can publish a date that was never written down. Where that happens, the record says so plainly and points you at the signed agreement instead of guessing.
How complete this is depends almost entirely on the government, not on us: some publish a term in most of their agreements, many in very few. Every government's coverage is published, the bad numbers alongside the good, so you can check the places you actually bid in before relying on any of it.
Coverage is also uneven by design of the source rather than by choice. A government that publishes little appears thin here. The absence of a record is not evidence that nothing happened.
When we get it wrong
Extraction errors are found and corrected, and a wrong end date is the error this project treats most seriously — somebody can calendar it and miss the real rebid. Dates recovered by a method later shown to be unreliable are removed rather than left in place to inflate a number. If you find something wrong, tell us and include the link; it is the fastest way to get it fixed.