tractatus/governance/TRA-OPS-0003-media-inquiry-response-protocol-v1-0.md

TRA-OPS-0003: Media Inquiry Response Protocol v1.0

Document ID: TRA-OPS-0003 Version: 1.0 Classification: OPERATIONAL Status: DRAFT → ACTIVE (upon Phase 2 start) Created: 2025-10-07 Owner: John Stroh Review Cycle: Quarterly Next Review: 2026-01-07 Parent Policy: TRA-OPS-0001 (AI Content Generation Policy)

---

Purpose

This document establishes the protocol for handling media inquiries (press, academic, commercial) using AI-assisted triage while ensuring human oversight for all external communications.

Scope

Applies to all incoming inquiries received via:

• Contact form (/contact)
• Email (contact@agenticgovernance.digital → john.stroh.nz@pm.me)
• Social media (future)
• Conference/event requests

---

Principles

1. Responsiveness

Commitment: Acknowledge all legitimate inquiries within 48 hours (business days).

AI Assistance: AI triages and drafts acknowledgments, but human approves all sent messages.

---

2. Privacy Protection

Commitment: No personal data processed by AI without anonymization.

Implementation:

• Strip PII before sending to Claude API
• Anonymize email addresses (sender@example.com → REDACTED)
• Remove phone numbers, physical addresses
• Audit trail: who accessed inquiry, when

---

3. Human Decision-Making

Commitment: Humans decide whether and how to respond to inquiries.

AI Role: Classification, prioritization, draft suggestions only.

Tractatus Boundary: AI cannot decide to send responses (STRATEGIC quadrant - external communication is values-laden).

---

Inquiry Classification

Categories

Category	Description	Priority	Response SLA
Press	Journalists, media outlets, news organizations	HIGH	4 hours (business)
Academic	Researchers, universities, conferences	MEDIUM	48 hours
Commercial	Companies, startups, integration partners	MEDIUM	7 days
Community	Individual users, hobbyists, students	LOW	14 days
Spam	Unsolicited marketing, irrelevant	IGNORE	-

---

AI Classification Criteria

Input to AI:

Classify this inquiry into: Press, Academic, Commercial, Community, or Spam.

Inquiry Text (anonymized):
[REDACTED_TEXT]

Context:
- Website: agenticgovernance.digital (AI safety framework)
- Audience: Researchers, implementers, advocates
- Mission: Advance AI safety through architectural constraints

Output format:
Category: [Press|Academic|Commercial|Community|Spam]
Confidence: [0.0-1.0]
Priority: [HIGH|MEDIUM|LOW|IGNORE]
Reasoning: [2-sentence explanation]

Human Override: Admin can reclassify if AI is wrong.

---

Priority Scoring

Factors

AI scores inquiries based on:

Factor	Weight	Description
Reach	30%	Audience size (NYT > Local Blog)
Relevance	25%	AI safety focus (direct > tangential)
Urgency	20%	Deadline (24h > 2 weeks)
Alignment	15%	Values alignment (AI safety advocate > adversary)
Opportunity	10%	Partnership, funding, exposure potential

Score Range: 0.0 (lowest) to 1.0 (highest)

Priority Thresholds:

• HIGH (0.7-1.0): Immediate attention (alert John Stroh)
• MEDIUM (0.4-0.7): Standard workflow
• LOW (0.0-0.4): Best-effort response

---

Response Workflow

Step 1: Inquiry Reception

Trigger: Form submission or email received

Automated Actions:

1. Log to database (media_inquiries collection)
2. Strip PII (email → REDACTED)
3. Send to AI for classification
4. Alert admin (email notification)

No AI Decision: System does NOT auto-respond (even acknowledgment).

---

Step 2: AI Classification & Triage

AI Task: Analyze inquiry and generate:

• Category (Press, Academic, Commercial, Community, Spam)
• Priority score (0.0-1.0)
• Suggested response template
• Key points to address
• Deadline (if mentioned)

Output Example:

{
  "category": "Press",
  "confidence": 0.92,
  "priority": "HIGH",
  "priority_score": 0.85,
  "reasoning": "Request from TechCrunch journalist, 48h deadline for AI safety feature article.",
  "suggested_template": "press_high_priority",
  "key_points": [
    "Framework overview",
    "27027 incident prevention",
    "Interview availability"
  ],
  "deadline": "2025-10-10"
}

---

Step 3: Human Review (Triage Dashboard)

Admin Dashboard: /admin/media-triage

UI Elements:

• Inquiry list (sorted by priority score)
• Color-coded priorities (red=HIGH, yellow=MEDIUM, green=LOW)
• AI classification (with confidence %)
• Draft response (AI-generated, editable)
• Action buttons: Approve & Send | Edit | Ignore | Archive

Human Actions:

1. Review AI classification (override if wrong)
2. Review priority score (adjust if needed)
3. Review draft response
4. Decide: Send, Edit, Ignore, or Escalate

---

Step 4: Draft Response Generation

AI Task: Generate draft response based on template + context.

Input to AI:

Generate a response to this [CATEGORY] inquiry.

Context:
- Inquiry: [ANONYMIZED_TEXT]
- Category: [Press/Academic/Commercial/Community]
- Priority: [HIGH/MEDIUM/LOW]
- Template: [Template Name]
- Key Points: [List from classification]

Guidelines:
- Professional, friendly tone
- Concise (2-3 paragraphs max)
- Include relevant links (docs, demos)
- Offer next steps (interview, meeting, resources)
- Sign off: "Best regards, Tractatus Team"

Output: Plain text email (no HTML)

Human Review:

• Fact-check all claims
• Adjust tone (friendlier, more formal, etc.)
• Add/remove details
• Personalize (use requester's name, reference specifics)

---

Step 5: Approval & Sending

Approval:

• Admin reviewer clicks "Approve & Send"
• System logs approval (who, when, what changed from AI draft)
• Email sent from contact@agenticgovernance.digital (ProtonBridge)

Follow-up:

• Set reminder for follow-up (if no response in 7 days)
• Track conversation thread
• Archive when resolved

---

Response Templates

Template: Press (High Priority)

Subject: Re: [Original Subject]

Hi [Name],

Thank you for your inquiry about the Tractatus Framework. We'd be happy to discuss how architectural constraints can advance AI safety.

The Tractatus Framework is the world's first production implementation of AI safety through architectural boundaries (rather than behavioral alignment). Our key innovation is the principle: "What cannot be systematized must not be automated."

Key points for your article:
- [Key Point 1 from AI analysis]
- [Key Point 2 from AI analysis]
- [Key Point 3 from AI analysis]

I'm available for an interview on [Availability]. You can also explore our interactive demonstrations at https://agenticgovernance.digital/demos.

Best regards,
The Tractatus Team
[John Stroh, Founder]

---

Template: Academic (Medium Priority)

Subject: Re: Research Collaboration - Tractatus Framework

Dear [Name],

Thank you for your interest in the Tractatus Framework for your research on [Topic].

We're actively seeking academic partnerships to validate and extend the framework. Our current focus areas include:
- Boundary enforcement mechanisms
- Cross-reference validation for instruction persistence
- Context pressure monitoring for degraded AI operation detection

For your [Conference/Paper], we can provide:
- Technical documentation: https://agenticgovernance.digital/docs
- Code examples: https://github.com/tractatus (future)
- Consultation: [Contact Information]

I'd be happy to discuss collaboration opportunities. Please let me know your timeline and specific research questions.

Best regards,
The Tractatus Team

---

Template: Commercial (Medium Priority)

Subject: Re: Integration Inquiry - Tractatus Framework

Hi [Name],

Thank you for your interest in integrating the Tractatus Framework into [Company/Product].

The framework is currently in Phase 2 development (soft launch). We expect production-ready packages in [Timeframe].

For early adopters, we offer:
- Implementation consultation
- Custom integration support
- Co-development partnerships (aligned organizations)

To get started:
1. Review the implementation guide: https://agenticgovernance.digital/implementer
2. Explore the API reference: https://agenticgovernance.digital/api
3. Schedule a technical discussion: [Calendar Link]

Best regards,
The Tractatus Team

---

Template: Community (Low Priority)

Subject: Re: [Original Subject]

Hi [Name],

Thanks for reaching out! We're glad you're interested in the Tractatus Framework.

For [Question/Topic], I recommend:
- [Relevant documentation link]
- [Demo link]
- [Case study or blog post link]

If you have specific questions after reviewing these resources, feel free to follow up or join our community discussions at [Future: Discord/Forum].

Best regards,
The Tractatus Team

---

Escalation Procedure

When to Escalate to John Stroh

Immediate Escalation (within 1 hour):

• Major media outlet (NY Times, Wired, BBC, etc.)
• Government/regulatory inquiry
• Legal threat or concern
• Security/privacy breach report
• Criticism/controversy requiring official response

Standard Escalation (within 24 hours):

• Partnership opportunities (funding, collaboration)
• Speaking invitations (conferences, podcasts)
• Ambiguous requests (not clear what they want)

Escalation Process:

1. Admin flags inquiry as "Escalation Required"
2. Email sent to John Stroh with:
   • Original inquiry (full text)
   • AI analysis
   • Admin notes
   • Suggested response (if any)
3. John Stroh responds with:
   • Approval to send draft
   • Revised response
   • "I'll handle this personally" (admin archives)

---

Spam & Abuse Handling

Spam Indicators (AI Detection)

• Generic language ("Dear Sir/Madam")
• Unsolicited sales pitches
• Cryptocurrency, SEO, marketing services
• Requests for backlinks, guest posts
• Obvious phishing attempts

Action: Auto-classify as "Spam", flag for human review (in case of false positive).

No Response: Spam inquiries are archived without response.

---

Abuse Handling

Definition: Harassment, threats, hate speech

Immediate Action:

1. Flag inquiry as "Abuse"
2. Do NOT respond
3. Alert John Stroh
4. Document incident
5. Block sender (if persistent)

Legal Threshold: Threats of violence → report to authorities (NZ police).

---

Privacy & Data Retention

PII Handling

Before AI Processing:

• Strip email addresses: sender@example.com → REDACTED_EMAIL
• Strip phone numbers: +64 21 123 4567 → REDACTED_PHONE
• Strip physical addresses
• Keep first name only (for personalization)

After AI Processing:

• Store AI-generated draft (no PII)
• Store final response sent (full email for audit)

Data Retention

Data Type	Retention Period	Reason
Original Inquiry	2 years	Legal/audit
AI Classification	2 years	Training/improvement
Draft Response	2 years	Audit trail
Sent Response	Indefinite	Legal/historical
Spam Inquiries	90 days	Reduce false positives

GDPR Compliance: Inquiry senders can request deletion (email contact@agenticgovernance.digital).

---

Performance Metrics

Response Quality

Metrics:

• Response time: % within SLA (target: 95%)
• Classification accuracy: % AI correct (target: 90%)
• Priority accuracy: % AI scoring matches human (target: 85%)
• Response rate: % inquiries receiving a response (target: 100% non-spam)

Engagement

Metrics:

• Follow-up rate: % inquiries leading to further conversation
• Partnership rate: % commercial inquiries → partnership
• Media coverage: # articles/mentions from press inquiries

---

Revision & Updates

Review Cycle: Quarterly

Update Triggers:

• Classification accuracy <80% (templates need improvement)
• Response SLA missed >10% of time (workflow issue)
• User complaints about response quality

---

Related Documents

• TRA-OPS-0001: AI Content Generation Policy (parent)
• TRA-OPS-0005: Human Oversight Requirements
• Privacy Policy (to be drafted)

---

Approval

Role	Name	Signature	Date
Policy Owner	John Stroh	[Pending]	[TBD]
Technical Reviewer	Claude Code	[Pending]	2025-10-07
Final Approval	John Stroh	[Pending]	[TBD]

---

Status: DRAFT (awaiting John Stroh approval) Effective Date: Upon Phase 2 media inquiry form launch Next Review: 2026-01-07