Program · Workforce Stability
Workforce Stability Among Post-9/11 Veterans
A founder-operated evidence program, chartered to build a defensible, reproducible picture of veteran workforce stability — and run, start to finish, under Six Sigma DMAIC and military after-action discipline. This is the anchor program of the section, and the one with the full chartered artifact set.
Sponsor
Legends' Return Foundation — founder-operated, disabled veteran-owned LLC
Methodology Base
Six Sigma DMAIC · Military After-Action Review discipline
Charter
v1.0 · chartered 2026-04-16 · Classification: public-interest research
Build
v0.16 — scaffold to current build in twenty-six days, no scope drift
The Program in Brief
Why this program exists.
Post-9/11 veterans separate from active service into a labor market that does not consistently translate military experience into stable civilian employment. The federal data needed to characterize that gap already exists — the American Community Survey microdata, Bureau of Labor Statistics local-area unemployment statistics, Bureau of Economic Analysis regional accounts, and the VA's National Center for Veterans Analysis and Statistics — but it is rarely integrated into a single analytical view that a policymaker, an employer coalition, or a veteran-services organization can act on.
Legends' Return Foundation chartered this program to close that gap: to produce a defensible, reproducible evidence base on workforce stability for the post-9/11 cohort, published under a methodology any qualified analyst can audit. It is presented here as a program — not just an output — because the discipline of how it was run is itself the deliverable. The charter, the phase plan, the risk register, and the after-action discipline below are the actual artifacts the program produced; each is downloadable at the foot of this page.
Commander's Intent
Mission.
Produce an integrated, reproducible analytical evidence base describing post-9/11 veteran workforce stability across geography, disability status, occupation, and time — and publish it under a methodology any qualified analyst can audit.
End state: a public case study, a reusable data pipeline, an illustrative retention model with a transparent model card, and an interactive explorer — delivered to a standard that withstands scrutiny from a federal program office, a regional employer coalition, or an academic reviewer.
Objectives
Five objectives, each with an auditable success criterion.
An objective without a verifiable success criterion is an aspiration. Each row below reads as “we will do X, and we will know we succeeded when Y is observable.”
| # | Objective | Success Criterion | Status |
|---|---|---|---|
| 01 | Integrate four federal data sources at the PUMA / state / sector level | All four sources joined, validated, and version-controlled by v0.10 | Met |
| 02 | Characterize the post-9/11 cohort on geography, disability, and labor-force outcomes | Reproducible cohort funnel with row-count tolerances at every step | Met |
| 03 | Build an illustrative retention model with a transparent ceiling | Model card published with AUC, calibration, limitations, and a section on synthetic-target effects | Met |
| 04 | Publish a case-study page, model card, and interactive explorer | All three artifacts live on PatrickNeilBradley.com with build-log references | Met |
| 05 | Maintain build discipline across the program | Build log appended for every ingest, join, and model refit | Met |
Scope
What this program does — and what it deliberately does not.
The out-of-scope list is the more important of the two. It is what protects the program against scope creep and against being read as something it is not.
In Scope
- ACS PUMS 2019, 2021, 2022, 2023 — 2020 unavailable due to COVID disruption
- BLS LAUS, BEA Regional Accounts, VA NCVAS as joinable context layers
- Post-9/11 cohort filter (MLPA = 1), ages 22–64, U.S. states plus DC
- O*NET Work Context composites for occupation-level features (Phase 6 addition)
- Illustrative logit retention model with O*NET composites folded in
- Public case-study page, model card, and interactive model explorer
Out of Scope
- Pre-9/11 era veterans — described only, never modeled
- Active-duty service members — filtered out via ESR codes
- Causal inference on policy interventions — the program publishes correlation, not causation
- Individual-level identification or PII — public microdata only
- Predictions about any named employer
Phase Plan
From scaffold to Build v0.16 — and the two phases held open.
Six phases delivered in twenty-six days. Phase 7 is intentionally deferred, not cancelled; Phase 8 and the formal close follow public launch. Deferred work is named openly, in the plan, rather than quietly dropped.
| Phase | Milestone | Target | Status |
|---|---|---|---|
| 1 | Scaffold and ACS PUMS ingest live (v0.1) | 2026-04-16 | Complete |
| 2 | Full ACS pull — all years and states (v0.2) | 2026-04-16 | Complete |
| 3 | BLS / BEA / VA joins; cohort validation | 2026-04-17 | Complete |
| 4 | Modeling prep; baseline logit | 2026-04-18 | Complete |
| 5 | Model card v1; case-study page draft | 2026-04-19 | Complete |
| 6 | O*NET Work Context composites (v0.16) | 2026-04-20 | Complete |
| 7 | Tier-2 OCCP → 8-digit SOC crosswalk for clean composite identification | Blocked on data access | Deferred |
| 8 | Public launch and supporting blog post | Pending | Pending |
| — | Formal After-Action Review | Within 14 days of public launch | Pending |
Risk Management
Ten risks, scored and owned at charter.
Each risk is scored on two ordinal dimensions — likelihood and impact, Low / Medium / High as 1 / 2 / 3 — with the composite as their product. The scoring is judgment-based, not actuarial; the value is the disciplined surfacing of risk, not numerical precision. The full register, with triggers, mitigations, and contingencies for every risk, is downloadable below.
| ID | Risk | Category | L × I | Score | Status |
|---|---|---|---|---|---|
| R1 | Synthetic retention outcome misread as a real prediction | Communications | Med × High | 6 · High | Mitigated |
| R2 | Tier-2 OCCP → 8-digit SOC crosswalk blocked outside the allowlist | Data | High × Med | 6 · High | Open |
| R3 | Sandbox API restrictions block reproducibility for outside reviewers | Reproducibility | Med × Med | 4 · Medium | Mitigated |
| R4 | Brand inconsistency across case study, model card, and explorer | Communications | Med × Low | 2 · Low | Open |
| R5 | Reviewer challenges lived-experience legitimacy of the cohort framing | Governance | Low × Med | 2 · Low | Mitigated |
| R6 | Quasi-multicollinearity between O*NET composites and SOC-major dummies | Methodology | High × Low | 3 · Medium | Mitigated |
| R7 | Federal data-source policy change mid-program | Data | Low × High | 3 · Medium | Open |
| R8 | Model card limitations section not read by a casual reader | Communications | Med × Med | 4 · Medium | Mitigated |
| R9 | Build-log loss or version drift between artifacts | Governance | Low × High | 3 · Medium | Open |
| R10 | Reproducibility failure on an external reviewer's machine | Reproducibility | Low × Med | 2 · Low | Open |
The highest-scored risk illustrates the discipline. R1 — a casual reader mistaking an illustrative model for an operational prediction — is mitigated structurally: the model card's limitations section was written before any results section, the case study uses the word “illustrative” in every relevant paragraph, and the executive deck carries an explicit honesty callout on the modeling slide. The contingency, if a misread is still observed in public reference, is a short pinned clarification post. The risk is named, scored, owned, mitigated, and given a fallback — before it activates.
Governance
One governance rhythm.
Daily
A build-log entry for every ingest, join, or model refit. Anything whose absence would make the program unreproducible.
Weekly
Program Manager review of risks, model-fit trajectory, and blockers.
Phase Gate
A documented decision-log entry with rationale — what was decided, why, and what was rejected.
At Close
A formal After-Action Review — sustains, improves, lessons learned. The closing deliverable.
The Output
What the evidence shows.
Program discipline is a means, not an end. Run correctly, it produced an evidence base with four findings a retention program can act on — each detailed, with its caveats, in the linked case study and executive summary.
Where veterans work is concentrated, not uniform. The top five occupation groups cover 44% of the employed cohort, tilted toward operational and protective roles — Protective Service is over-represented 4.14× relative to non-veterans. A hiring team that expects veterans in food-service or personal-care applicant pools is reading the labor market wrong.
The federal–civilian corridor is structural. About 17% of the employed cohort works for the federal government — roughly ten times the non-veteran civilian rate. A private-sector retention program competes for talent already filtered by a federal pipeline.
The disability margin sits at the door. Employment-to-population ratio drops from 85% at the 1–30% severity band to 63% at 61–100%, while earnings and hours are flat across every rating band once inside employment. Disability severity predicts who enters work, not what they earn once there.
The model is illustrative, and says so. The retention target is synthetic — built from cross-sectional panels rather than observed longitudinal transitions. Test AUC of 0.566 is flat by design; the model card discloses that ceiling openly. Feature importances remain interpretable; predicted probabilities are illustrative only. Every descriptive rate above is observed, not modeled.
Program Artifacts
The artifacts, downloadable.
Four program-management artifacts and three analytical deliverables. The claim on this page is carried by the documents, not by the prose — so they are all here to open.
Project Charter · Markdown
Program Charter
Mission, objectives, scope, stakeholders, the eight-phase plan, assumptions, constraints, and the five top risks identified at charter. Version 1.0, dated 2026-04-16.
Download Charter (.md) →Executive Deck · PDF / PPTX
Program Leadership Deck
A fourteen-slide walkthrough of the program — commander's intent, objectives, scope, stakeholders, phase plan, data architecture, results, modeling honesty, risk management, and an after-action preview.
Download Deck (.pdf) → Download Deck (.pptx) →Risk Register · PDF
Risk Register
All ten risks with a heat map, register summary, and a full profile for each — owner, trigger conditions, current mitigation, and the contingency to be executed if the risk materializes. Register v1.0.
Download Risk Register (.pdf) →Executive Summary · DOCX
Executive Summary
A two-page handout for recruiters, HR leaders, and veteran-employment program directors — the thesis, the four findings, the geographic snapshot, and a plain statement of what the data is and is not.
Download Summary (.docx) →The analytical deliverables
The program's published outputs live on this site as their own pages:
Work Together
This is how I run a program.
A chartered mission, an auditable execution loop, risk scored before it activates, and an honest after-action review at close. If your team needs a program run to that standard, let's talk.