VRSI 2026.1 — A County-Grain Composite
of Veteran Regional Outcomes

The problem: state-level reporting hides where veterans actually live

Public-facing dashboards publish veteran outcomes in state-sized slabs. The county is where employment programs, housing assistance, medical access, and benefits infrastructure are actually administered.

Federal agencies already publish most of the data needed for a county-grain view. They just publish it in eight different places, on eight different cadences, under eight different schema conventions. Nothing connects them. The consequence is that the veteran-focused analyst who wants to ask a simple county-level question — is the benefits pipeline in this county as healthy as the labor market? — has to build the integration themselves every time, from first principles.

VRSI is that integration. The composite does not invent new data; it disciplines the existing public record. Every input is a public federal series. Every transformation is logged, versioned, and reproducible. The value added is the composite structure, the fragility disclosure, and the partial-publish contract.

Methodology: eight sources, twenty KPIs, five domains, one composite

Each KPI is pulled, min-max scaled against the live universe, and equally weighted within its domain. Domains then combine into the composite at equal 20% weights each.

The five domains

Composite results — partial-publish discipline, not vaporware

The live index covers half of its twenty KPIs at full fidelity today. The coverage ladder is the release plan, and it is visible.

The partial-publish contract

A composite that waits until all twenty KPIs are perfect never ships. One that publishes fabricated values to hit a coverage number is worse — it launders data quality into a false headline. VRSI splits the difference. Each KPI either resolves cleanly from its federal source, or it emits a disclosed proxy and a fragility flag. The release manifest carries the live coverage ratio, a per-KPI status vector, and a human-readable rationale for every proxy in play. A downstream consumer can choose to include or exclude proxy-flagged KPIs from their own view.

Per-domain signal — where the variance actually lives

Each domain carries its own geography and its own fragility envelope. The composite is only as useful as the domains that feed it, so they are reported individually before they are combined.

Economic Stability

The four economic KPIs track demand-side (unemployment, LFP) and supply-side (RPP, per-capita income) conditions. Seed-universe variance is dominated by the urban-rural gradient: metropolitan counties carry higher RPP and higher per-capita income but similar LFP, while non-metro counties carry lower unemployment rates with substantially lower income — a common sign of labor-force detachment rather than labor-market tightness. A veteran-specific LFP series (ACS S2101) separates genuine tightness from attrition.

Health & Clinical Access

Disability prevalence correlates only loosely with economic strength. What drives the Health subscore in the seed universe is clinical access — VA facility density and HPSA scores — not the underlying disability rate. The Day 1.9c reroute of VA facility data from the 404-ing Socrata CSV to the authenticated VA Facilities API (api.va.gov/services/va_facilities/v1/facilities, paginated JSON envelope, lowercase apikey header) restored full-fidelity county coverage on this KPI without relaxing the test suite.

Education & Human Capital

GI Bill density and approved-institution density carry the heaviest educational signal because attainment alone is sensitive to the underlying cohort’s age distribution. Counties near large approved institutions show predictably higher GI Bill beneficiary density; the signal is geographic more than it is behavioral, and the composite treats it that way.

Benefits Access

This is the single most fragile domain. Three of the four KPIs depend on VA NCVAS county-level tables, which update on a once-a-year cadence and are historically the most prone to format drift on the source side. Day 2.1 widened _BENEFITS_OPTIONAL_COLUMNS to 19/19/15 Socrata-style matchers per KPI and added state-column detection. Expected next-run outcome: two Benefits KPIs move off proxy. DOL VETS is the only non-VA series here and drives a stabilizing effect on the domain subscore.

Community Infrastructure

The Community domain is carried by HUD: CHAS cost-burden shares, Fair Market Rent ratios, and USPS vacancy rates. IRS Exempt Organizations contributes the veteran-non-profit-density KPI, NTEE-filtered. The HUD series are unusually clean at the county grain, which is part of why Community ended up with the highest live coverage of any domain at Day 2.0.

Cross-domain insights — what the composite is telling us

Patterns visible once the five domains are aligned to a single geographic grain.

Economic strength and community infrastructure move together cleanly — counties with lower veteran unemployment and higher per-capita income also tend to have lower cost-burden shares and lower vacancy. That is not surprising, but it is rare to see it at the county grain.

Economic strength does not predict clinical access. The HPSA and VA-facility signals respond to federal infrastructure decisions, not to local labor markets. A wealthy county can still be a mental-health desert, and the composite does not blur that distinction.

PCA on the normalized KPI matrix places the first principal component at roughly 40% of explained variance, loading most heavily on the Economic and Community domains. The second component separates counties on the Health / Benefits axis. That two-axis structure shows up again in k-means clustering with k = 4: Urban Benefited (high on both axes), Rural Resilient (moderate-economic, strong-community), Transition Hubs (high-education, mixed-benefits), and At-Risk (low across the board). These are the four archetypes that emerged in Day 2.1 unsupervised analysis.

Modeling insights Unsupervised — descriptive

PCA and k-means run on the twenty-KPI matrix after imputation of proxy-flagged cells. The intent is exploratory, not predictive.

The PCA biplot separates the seed counties along two meaningful axes. PC1 (~40% variance) is a general “regional prosperity” axis — labor market + cost of living + community infrastructure load together. PC2 (~18% variance) captures the health-and-benefits axis — counties with strong VA facility access and active compensation pipelines push in one direction, counties with high HPSA scores and thin benefits throughput push in the other.

K-means at k = 4 produces four archetypes with face-valid county membership in the seed universe:

Risks and fragility — disclosed up front

Every composite is fragile somewhere. VRSI publishes where.

The Benefits Access domain is the most fragile, and it is fragile in a specific, reproducible way: three of its four KPIs depend on VA NCVAS county-level releases that historically ship with inconsistent column headers between years. The Day 2.1 fix widened the Socrata-style column matchers to 19/19/15 candidate needles per KPI and added a state-column fallback that lets the loader detect county-level data even when the column naming changes. That reduces the failure mode from “pipeline breaks” to “KPI drops to proxy flag.”

External API fragility is the second risk class. The Day 1.9c VA Facilities reroute caught a real failure: the prior Socrata CSV endpoint started returning 404, which would have silently zeroed the facility-density KPI. The fallback is the authenticated VA Facilities JSON API with key rotation, paginated envelopes, and a format-dispatch pattern that lets the loader accept either Socrata-style or API-style responses without code duplication.

HRSA HPSA scores and IRS Exempt Organizations refreshes sit on quarterly update cadences; the pipeline respects their cadence and does not attempt to synthesize intra-quarter values. USPS vacancy and HUD FMR are the most reliable series in the stack — they have not required any schema repair across the build log.

Recommendations — three tiers

Policy readers, data-investment readers, and pipeline-maturation readers each get a distinct actionable ask.

Tier 1 — Policy

County-grain visibility is the lever that unlocks targeted action. VA and DOL VETS pipelines administered at the state level mask within-state heterogeneity by design. The At-Risk archetype identified in Chapter Six is the operational target for resource reallocation — these counties show low subscores across all five domains simultaneously, which is the signature that state-level averages smooth away.

Tier 2 — Data investment

The single highest-leverage investment is the VA NCVAS county-level release cadence and schema stability. Every proxy flag in the Benefits domain resolves the moment that release is consistent year-over-year. A secondary investment is HRSA HPSA primary-care and mental-health score integration with the veteran-specific population denominator rather than the total-population denominator currently exposed.

Tier 3 — Pipeline maturation

The Day 2.1 schema-widening pattern generalizes. Any federal source that ships semi-structured CSVs with unstable column headers (NCVAS, some BLS LAUS partitions, parts of the HUD CHAS release) benefits from a Socrata-style matcher with disclosed candidate-column lists. The same pattern is a candidate for open-sourcing as a standalone utility package.

Roadmap

Day 1.9c is done. Day 2.2 is next. Day 2.3 through Day 2.8 are queued.

Download and further reading