Overview
AgrStak, powered by NumeriXGPT LLC, transforms USDA agricultural research into AI-powered calculators for farmers, researchers, and agricultural professionals. As the platform scales toward enterprise contracts with USDA research institutions and extension services, formal data governance is essential to maintaining research integrity, user trust, and institutional credibility.
Our governance framework covers three interconnected domains:
Research Data
How USDA datasets and scientific models are sourced, versioned, validated, and maintained across all 431 calculators.
User & Platform Data
How farmer inputs, calculation outputs, analytics, and account information are classified, stored, and protected.
AI & Cloud Data
How AI model interactions, cloud infrastructure, and third-party APIs are governed to protect partner and user data.
AgrStak's methodology for translating USDA research into AI-powered calculators is patent pending. Data governance records — including model sourcing logs and versioning history — serve as supporting documentation for IP protection.
Governance Structure & Roles
AgrStak operates as a lean startup. Governance roles are assigned with clear accountability and will expand as the organization grows.
| Role | Responsibility | Current Owner |
|---|---|---|
| Data Governor | Policy authority and final decisions on all data use across domains | CEO / Founder |
| Research Data Steward | USDA dataset sourcing, versioning, citation accuracy for all 431 models | CEO / Founder |
| Platform Data Steward | User data, privacy compliance, analytics hygiene | CEO / Founder |
| AI/Cloud Data Steward | AI model inputs/outputs, cloud security, API contracts | CEO / Founder |
| External Auditor | Annual third-party compliance review | TBD — Year 2 |
Research Data Governance
AgrStak's core value proposition is built on USDA agricultural research. How that research is sourced, maintained, and cited directly affects credibility with institutional partners and the accuracy of every calculator output.
Sourcing Standards
- All agricultural models derived from peer-reviewed USDA-ARS publications or official USDA technical reports
- Source documents archived internally at the time of each model's creation
- No proprietary third-party datasets incorporated without a written licensing agreement
- Model inputs and coefficients must be traceable to the original USDA publication
- No proprietary third-party data incorporated without written licensing agreement
Model Versioning
Each calculator model is tied to a specific version of the underlying USDA research. When USDA updates its findings, AgrStak updates the corresponding model. Every model in the registry carries the following metadata:
| Field | Description | Required |
|---|---|---|
| Model ID | Unique identifier per calculator (e.g., AGRSTAK-SB-001) | Yes |
| Source Citation | Full APA citation of USDA source publication | Yes |
| Version Number | Semantic version of the model (e.g., v2.1.0) | Yes |
| Last Validated | Date model was reviewed against current source data | Yes |
| Research Contact | USDA researcher who authored the source publication | Recommended |
| Deprecation Date | Date model will be retired if not updated with current research | Yes |
All models undergo internal validation before deployment. Material discrepancies greater than 5% deviation from USDA source data trigger a model review before launch. All 431 models are reviewed annually against the latest USDA publications.
User & Platform Data Governance
AgrStak collects data from farmers, researchers, and agricultural professionals. This section governs what is collected, how it is stored, and how it is protected.
Data Classification
| Class | Examples | Protection |
|---|---|---|
| Public | Calculator descriptions, USDA model summaries, published citations | Standard |
| Internal | Business logic, pricing, model coefficients, internal analytics | Restricted |
| Confidential | Partner contracts, LOIs, unpublished USDA research, DSA data | Highly Restricted |
| User PII | Names, emails, farm location, subscription records | Maximum Protection |
| Calculator Inputs | Soil data, crop parameters, field measurements | Ephemeral — session only, never stored |
Data Retention
AI & Cloud Data Governance
AgrStak's calculators are powered by AI models, including the Anthropic Claude API for the AgrStak Chat Widget (ACW). Every AI interaction carries governance implications — particularly when pursuing federal institutional contracts that have strict data handling requirements.
Decision Support Only
AI outputs are decision support tools. All AI-assisted results carry a disclaimer recommending validation with a qualified agronomist or extension agent.
No Training on Partner Data
Institutional partner data and non-public USDA research is never used to train or fine-tune AI models without explicit written consent.
No PII in AI Calls
All prompts sent to the Anthropic API are reviewed to ensure no personally identifiable information is transmitted. API data processing agreements prohibit training use.
Hallucination Monitoring
AI outputs referencing non-existent USDA studies are flagged for review. Users can report inaccuracies via the feedback mechanism on every calculator page.
Version Transparency
AI model version changes are documented. Calculator outputs are re-validated against USDA source data before any new AI model version is deployed.
Cloud SOC 2 Standards
AgrStak's hosting infrastructure (Vercel) maintains SOC 2 Type II certification. All data is encrypted in transit (TLS 1.2+) and at rest (AES-256).
Data Incident Response
A data incident is any event that compromises the confidentiality, integrity, or availability of AgrStak data. Our response protocol follows a six-step process with defined timelines.
| Step | Phase | Action | Timeline |
|---|---|---|---|
| 1 | Detect | Identify incident via monitoring alerts, user reports, or internal discovery | Immediate |
| 2 | Contain | Isolate affected systems, revoke compromised credentials, disable affected APIs | Within 1 hour |
| 3 | Assess | Determine scope, data types affected, and number of users impacted | Within 4 hours |
| 4 | Notify | Notify affected users, institutional partners, and regulators as required by law | Within 72 hours |
| 5 | Remediate | Fix root cause, implement additional controls, restore services | Within 7 days |
| 6 | Review | Post-incident report documenting cause, impact, and prevention measures | Within 30 days |
In the event of a data incident involving Shared Data governed by a Data Sharing Agreement (DSA), the affected party will notify the partner institution within 72 hours of discovery per the DSA breach notification clause.
Governance Review Calendar
Data governance is not a one-time activity. AgrStak maintains a structured review cadence to keep the framework current and effective.
Analytics Review
GA4 data quality, event tracking accuracy, anomaly detection across all storefronts.
Security Audit
API key rotation, GitHub secret scanning, vendor access review, and cloud configuration check.
AI Output Review
Flagged AI outputs reviewed, hallucination rate assessed, user feedback patterns analyzed.
Full Framework Review
All policies updated, USDA model validation completed, privacy policy refreshed.
Vendor Review
All third-party Data Processing Agreements reviewed. New vendors assessed before integration.
DSA Execution
A custom Data Sharing Agreement executed with each new institutional partner prior to data exchange.
Governance Roadmap
As AgrStak scales, governance infrastructure grows with it. The following milestones define our path from startup to enterprise-grade data governance.
Foundation
Publish this Data Governance Framework publicly. Complete the internal model registry for all 431 calculators. Execute first DSA with a USDA-ARS institutional partner.
Section 508 Compliance Audit
Complete WCAG 2.1 AA accessibility audit across all storefronts. Remediate any barriers to ensure the platform serves the full spectrum of agricultural professionals.
Team & Role Separation
Hire a dedicated Platform Data Steward as the team grows. Separate governance roles across research, user, and AI/cloud domains.
SOC 2 Type I Certification
Pursue SOC 2 Type I to meet the security requirements of enterprise and federal institutional customers.
SOC 2 Type II & External Audit
Achieve SOC 2 Type II certification. Establish a formal annual external audit cycle with a qualified third-party auditor.
Automated Data Lineage
Implement automated data lineage tracking across all 14 storefronts — tracing every data point from USDA source to calculator output.