The appeal and the trap
Everyone wants one number per county. It sorts, it maps, it fits in a column, and it lets you look at 3,221 counties without reading 3,221 briefings.
The trap is that a composite score hides its own inputs. Two counties can score the same for completely different reasons — one because a single large employer is wobbling, another because its whole labour force is slowly ageing out. The same number, two entirely different responses.
So the score is the start of the question, not the answer to it.
What ours covers
3,221 counties. 52 state codes — the fifty states plus DC and Puerto Rico.
That coverage number matters more than it looks. There are roughly 3,144 counties and county equivalents in the United States; a dataset that quietly covers 2,800 of them and leaves the rest null will look complete in a dashboard and be wrong in exactly the places you were worried about. Coverage should be stated, not implied.
Three things a score cannot do
It cannot tell you about a single employer. A county-level indicator is an aggregate. If your actual question is "is the plant on the north side closing", the score is the wrong instrument — that is what WARN filings and court records are for.
It cannot see the future by itself. A score built on reported data is a description of conditions that have already been measured. Reported county unemployment lags. Treat a score as a statement about the recent past that is suggestive of the near future, and you will be right more often than if you treat it as a forecast.
It cannot be compared across a methodology change. This is the one that catches people. If the inputs or weights change, scores before and after the boundary are different quantities wearing the same name. Rank correlation across such a boundary can be near zero — meaning the "top ten counties" lists on either side may share almost nothing. Any serious vendor should tell you where those boundaries are.
How to actually use one
The honest workflow is narrow-then-verify:
1. Sort by score to get from 3,221 counties to a shortlist of twenty. 2. Look at the components for those twenty. Which input is driving it? 3. Go find the specific thing. WARN filings, court dockets, local reporting, the actual employers.
The score does step one, which is genuinely valuable — it is the step that is impossible by hand. It does not do steps two and three, and a vendor implying otherwise is selling you a feeling.
Seasonality, and why some counties carry a flag
Some counties swing hard by season — agriculture, tourism, seasonal construction. A normal seasonal trough can push a county up a risk ranking for reasons that will reverse on schedule and mean nothing.
Counties with high seasonal amplitude are flagged rather than silently adjusted, because the flag is information: it tells you to read the trend direction with extra caution rather than pretending the swing was not there.
Try it before paying
Connect an MCP-capable assistant to `product.kcalbinmcp.uk/mcp` and query county signals. No account needed. County names stay masked until you subscribe — you can confirm the coverage and see the severity and forecast before paying for the detail.