US county unemployment intelligence (BLS LAUS)

US county unemployment intelligence from the BLS Local Area Unemployment Statistics county bulk files (keyless, U.S. public domain): 3,225 counties and county-equivalents (incl. DC and Puerto Rico municipios), monthly January 1990 to the latest published month, NOT seasonally adjusted. Core measures: unemployment rate, unemployed persons, and labor force (unadjusted). Method: the closed county slice is resolved from la.series (fail loud on any missing or duplicate series); the ~336 MB data file is streamed one series at a time; M13 annual averages are dropped; '-' cells are kept as honest NULLs; preliminary (footnote P) months are flagged. Derived signals per county: month-over-month and year-over-year change of the rate (YoY is the seasonal proxy for unadjusted data), trailing-12-month z-score of the rate, labor-force YoY percent change, a documented 0-100 labor_stress_score (fixed weights: 50% rate level vs the county's own history, 30% clipped YoY direction, 20% clipped z-score spike), within-month national quintile tier and rank of the score, a distress flag (YoY >= 2 pp) and a tight-labor flag (rate < 4% and not rising). Use as prediction features: join to customers/orders/leads by county_fips + date for demand, churn, and lead-scoring models — local labor-market stress leads changes in discretionary spending, subscription churn, and B2B close rates. Caveats: county estimates are model-based with larger error for small counties; the latest ~14 months are revised each January; always use YoY (not MoM) signals because the data are not seasonally adjusted; the stress score is relative to each county's own history, not an absolute cross-county level comparison.