Method
How a public data point becomes a signal you can act on - and how to read our confidence in it.
01 - Source
Every signal traces back to a specific row in the 162-source library: a named publisher, an access method, and a cadence. We link the exact source and, where possible, the exact query - so you can verify in under a minute.
02 - Detect
The detection engine pulls each source on its own schedule (daily for drought and filings, weekly for grants and agendas, monthly-to-annual for census and infrastructure data) and runs it against threshold and anomaly rules - a magnitude/novelty/geography/audience/actionable/confidence rubric scores each candidate 0–100. A score of 60 or higher clears the bar to become a lead worth writing up.
03 - Write
A cleared signal becomes a brief: what moved, why it matters, who should care, and how to replicate the check yourself. Every brief carries a claim label so you know exactly how much weight the evidence can bear.
Claim-label key
Verified directly against the primary source's live data - a number we pulled and checked ourselves, with the exact API call shown.
A pattern worth tracking, but not yet confirmed across enough data points to call it a trend. We say so explicitly.
A worked example showing the kind of signal this source can produce, built for demonstration rather than reported as a live, dated finding.
Marks a source (not a signal) whose access method and data freshness were manually checked. A check status means it's queued for that verification pass.
What we don't do
We don't editorialize beyond what the data supports, we don't hide the source, and we don't sell what a number means before showing our work on what the number is. If a signal turns out to be wrong or the source migrates and breaks a link, we correct the page and note the correction.