How deep the data actually goes.

Coverage claims are easy to inflate, so we compute them from the database and show them here. The numbers below are live counts of what we hold, and — more importantly — exactly which integrity detectors can run on it. A detector is only ever counted as “live” where its input data is actually present.

These are three different measures, not one. 141,480 is every award we hold; the repeat-winner and award-vs-estimate detectors run across all of it. The single-bidder detector needs a disclosed bidder count, which only some sources publish — today 3,708awards carry one (chiefly Prozorro's OCDS records; Colombia SECOP II is being ingested to grow this). We show the small number plainly rather than imply the flagship signal runs on the whole corpus.

What each detector can run on

By source

A jurisdiction with notices but no award history can run window and threshold detectors, but not repeat-winner or single-bidder — coverage is a capability question, not just a country count. We’d rather show a source as lightly populated than pretend depth we don’t have.

Connector reachability

We run many source connectors, but not all of them return usable data on every cycle. This is the last probe result for each, straight from the monitor — so you can see exactly which sources are ingesting today and which are merely reachable. We’d rather show a connector as idle than imply depth we aren’t pulling.

How fresh the data is

A detector running on a stale snapshot is a weaker claim than one running on this week’s records, so we show the age of each source’s most recent ingest — computed live from the database, not asserted.