Huntington Analytics STEWARDSHIP
DATA STEWARDSHIP // FOR THE GUMPTION PROJECT

Sensitive data deserves
stewardship, not just a privacy policy.

Coalfield Development has spent a decade refining a curriculum with people facing real barriers to employment. This is the governance architecture for how that curriculum — in conversation with Marshall University and Huntington Analytics — becomes a technology platform without participants losing custody of their own data.

Governance Model
3-LAYER
Vendor / Steward / Research
Federal Frameworks
4 CFR
42 CFR 2 · HIPAA · FERPA · Common Rule
Coalfield Curriculum
10+ YR
Refined with people facing real barriers.
// Section 01

The Three-Layer Governance Model

Each layer has a distinct role, distinct legal exposure, and distinct accountability. The architecture prevents any single actor from holding unilateral control over participant data.

Layer 01
Platform Vendor
Huntington Analytics

Builds the interface, the AI coaching, and the employer-facing dashboards. Holds zero PII outside of session contexts — data is processed in transit and never persisted to HA infrastructure beyond operational necessity. HA pays for the AWS infrastructure, earns license revenue from employer partners, and is contractually bound to the data steward. HA does not own participant data and cannot transfer it, sell it, or retain it after contract termination. The relationship is vendor-to-steward, not data controller.

Layer 02
Data Steward
Coalfield Development

Holds the keys. Holds participant consent. Holds the legal trustee role for all participant data. Can revoke vendor access at any time and for any reason. In this collaboration, the steward is Coalfield Development — a Huntington, WV 501(c)(3) that has spent over a decade refining a curriculum with adults facing the highest barriers to employment, and whose community accountability structure is what gives the model legitimacy with regulators, IRBs, and participants themselves. As the work scales, the steward may expand into a formal consortium that includes Coalfield, Marshall University, and partner employers — but the steward role belongs to Coalfield, not to the platform.

Layer 03
Research Partner
Marshall University

Generates aggregate insights — publications, white papers, program evaluation reports — under full Institutional Review Board oversight. Touches individual-level data only under 45 CFR 46 protocols and only with participant consent already captured by Coalfield as steward. De-identified outputs are the deliverable; identifiable records never leave steward custody. In this collaboration, that research partner is Marshall University — specifically the Brad D. Smith Schools of Business and the AI Steering Committee co-chaired by David Wiley, Ph.D. Marshall's Skills Exchange platform and the Durable Skills microcredentials curriculum provide the natural channel for credentialing participants as outcomes accrue, turning a decade of Coalfield practice into peer-reviewed evidence.

Participants consent once. The steward holds the keys. The vendor delivers the experience. The research partner generates the insights. Data never leaves community control — even when employers license the platform.

// Section 02

The Frameworks We Build On

The Gumption project is not building its governance model from first principles. It is applying an established playbook — a decade of data collaborative practice, responsible research infrastructure design, and workforce data portability work — that long predates this conversation.

The GovLab at NYU

Stefaan Verhulst's group at NYU GovLab; the canonical taxonomy of data collaboratives and responsible data infrastructure. Defines the typology of data-sharing arrangements between institutions.

WE USE: Typology framework for classifying our three-layer arrangement; language for partner agreements.
Coleridge Initiative — ADRF

Julia Lane / NYU; the Applied Data Research Framework. Secure enclave model for state administrative and workforce data. Practical architecture for researcher access without raw-data transfer.

WE USE: Enclave access pattern for Marshall IRB research; model for state workforce agency data agreements.
Recidiviz

Open-source nonprofit building responsible data platforms for state corrections agencies. Sets the standard for reentry-context data governance: agency-controlled, transparent methodology, no surveillance creep.

WE USE: Governance contract templates; design pattern for populations who have been systematically harmed by data misuse.
WorkRise — Urban Institute + Markle

Workforce mobility research network focused specifically on workers facing structural barriers to employment. Applied research on what data governance looks like when the workers themselves are the constituency.

WE USE: Measurement frameworks for workforce outcomes; definitions of barrier populations we adopt directly.
Open Referral Initiative — HSDS

Human Services Data Specification — the 211 interoperability standard. The field-tested schema for community resource data, built to survive across vendor transitions because the community owns the spec.

WE USE: Data schema for the services layer in Pathway Fynder; interoperability model for partner systems.
Future of Privacy Forum

Cross-sector privacy advocacy and policy organization. Produces sector-specific privacy frameworks for education, workforce, and health contexts — the guidance that sits between the federal floor and actual practice.

WE USE: Ed-tech and workforce privacy guidelines; consent design standards for sensitive-context data collection.
Markle Foundation — Rework America Alliance

Workforce data sharing initiative built on the premise that workers should benefit from — not merely be subject to — data collected about them. Practical policy frameworks for skills-based hiring and portable credentials.

WE USE: Skills taxonomy alignment; data portability principles that inform our participant data-export design.
// Section 03

The Regulatory Floor

Coalfield's 4-Month Crew Member Selection Rubric explicitly awards 30 of 100 points for Need for Opportunity — and that weighting is designed to include people in SUD recovery, people re-entering from incarceration, and people from households in generational poverty. Each of these regulations was written precisely because these populations have been harmed by data misuse before. Our architecture satisfies all four simultaneously, not sequentially.

42 CFR Part 2
Federal SUD-Treatment Record Protection

Stricter than HIPAA for a reason: substance use disorder treatment records carry a history of being used against the people they document. Part 2 bars disclosure without explicit written consent, even to other treating providers. No general authorization, no "public health" carve-out, no law enforcement exception without a court order.

Applies: any participant flagged as in SUD recovery
HIPAA
Health Insurance Portability and Accountability Act

The baseline for health and mental-health data. Applies wherever wellness scoring, clinical referrals, or treatment-context data touch the platform. Creates the minimum de-identification standards we apply to any aggregate employer reporting that could reveal health status — even indirectly.

Applies: wellness scoring, behavioral health context
FERPA
Family Educational Rights and Privacy Act

Educational records protection. The moment Marshall University or any accredited institution is in the data flow — as a research partner, a credential issuer, or a referral recipient — FERPA applies to any student or former-student records. Our data architecture isolates educational record handling from workforce-outcome data to prevent commingling that would expand FERPA's scope unnecessarily.

Applies: any higher-ed partner in the data flow
45 CFR 46 + IRB Review
The Common Rule — Human Subjects Research

Federal baseline for any human-subjects research. Subpart C adds explicit protections for prisoners and justice-involved populations, who are designated a vulnerable population requiring heightened consent and review processes. Marshall's IRB review covers all research use of participant data. No research use proceeds without a current IRB protocol on file with the steward.

Applies: all research; Subpart C for justice-involved participants
// Section 04

Operational Patterns

These patterns are not abstract — each maps to a concrete implementation decision for the Gumption platform. Differential privacy means an employer licensing the platform never sees an individual Crew Member's score. Federated learning means Coalfield's participants are not co-mingled with participants from a different employer deployment. Each mechanism makes the governance model enforceable, not merely promised.

// Section 05

The Gumption Project, In This Architecture

Here is how Coalfield's existing curriculum components map into the three-layer model. Nothing is invented. Nothing is replaced. Everything currently being done by Coalfield's team stays in Coalfield's custody. Technology only changes how it is expressed — and who can respond when a participant needs something.

Layer 02 — Steward Holds the Keys To

Coalfield Development

  • The Strengths and Needs Survey — the intake instrument that sets each participant's initial goals and Stepping Stones
  • The 4-Month Crew Member Selection Rubric, including the 30-point Need-for-Opportunity weighting, which never leaves community control
  • Milestones, Certifications Received, and the 12 Personal Themes self-evaluation responses
  • Participant consent — granular, revocable, scoped by purpose of use
Layer 01 — Vendor Delivers

Huntington Analytics

  • A Claude-coached, conversational version of the Personal Development Journal — currently a static PDF; finally able to respond to where each participant actually is
  • A modern interface for the Weekly Engagement Scorecard, the 13 Professional Themes monthly eval, and Milestone tracking — replacing forms with dialogues
  • Role-specific dashboards: Crew Chief view, Workforce Development Coordinator view, Program Manager view, employer aggregate view — each with strict record-level isolation
  • Migration tooling from Apricot (Bonterra / Social Solutions) — clean transition, nothing left behind
Layer 03 — Research Partner Generates

Marshall University

  • Aggregate program outcome reports for Coalfield's funders and grant reporting
  • Durable Skills microcredentials issued through Marshall's Skills Exchange as participants meet documented learning outcomes
  • IRB-approved evaluation studies suitable for publication — turning a decade of practice into peer-reviewed evidence
  • De-identified pattern analysis to inform the next generation of the curriculum without exposing individuals

Brandon Dennison's framing question was whether a technology platform — built off everything Coalfield has learned about human development — could become a product other employers benefit from as they develop their own workforces. The answer is yes, and the three-layer model is how. Employers license the experience. Coalfield holds the curriculum and the data. Marshall measures whether it works. Participants stay in control. What scales is the platform; what stays rooted is the trust.

// Section 06

Why Huntington Analytics

Building AI platforms with sensitive populations is not just a technical problem. It is a dignity problem first. The populations this work serves — people in reentry, people in recovery, people who have been failed by workforce systems that were not designed with them in mind — have every reason to distrust institutions that collect their data. That distrust is earned. Our governance architecture is designed to be worthy of overriding it, not to paper over it with consent forms.

Huntington Analytics is a West Virginia company. Our roots are Appalachian. The people whose data we are talking about are our neighbors. That is not a marketing line. It changes the calculus on every design decision: what we build here will be used here, by people we know and in communities we are part of. We do not have the luxury of treating governance as a compliance exercise. The standard we set becomes the standard this region's data infrastructure runs on.

Pathway Fynder — our workforce coaching platform — is the proof of architecture, not a prototype. It is FERPA-compliant and role-isolated at the record level. AI coaching runs without persisting individually identifiable session data. Employer dashboards receive aggregate cohort views only. The same architectural pattern scales to any partner context. What we are proposing for this collaboration is not aspirational — it is an extension of what we already ship. The mapping for the Gumption platform is direct: Pathway Fynder's educator dashboard becomes the Crew Chief view; the student becomes the Trainee or Crew Member; the industry partner becomes the participating employer. The architectural pattern is shipping. The instantiation is the work.

The framing that guides everything here is one we have held from the beginning: "ed-tech, but also concepts in transhumanism and preserving human and Appalachian dignities." Those are not in tension unless you let them be. The governance model is how you hold them together.

Proof Point // Pathway Fynder
Same Architecture, in Production
  • FERPA-compliant by design, not by addendum
  • Role-isolated at the record level — employers see cohort data, never individuals
  • AI coaching pipeline holds zero persistent PII between sessions
  • Consent captured at enrollment, scoped by data use purpose
  • Audit log maintained for all data-access events
  • Portable: participant data exports on request, in standard schema
  • Same architectural pattern proposed for the Gumption platform — no second system to invent