ADR-014: Attribution-First Development
Status: Proposed
Date: August 17, 2025
Decision Makers: PM, Chief Architect, Chief of Staff
Context
Piper Morgan’s development has benefited from numerous external sources: frameworks, research papers, open-source patterns, and community knowledge. As we formalize our Agent Charter and operational principles, we recognize that intellectual integrity requires systematic attribution of these influences.
Recent discoveries include:
- Rahul Vir’s Agentic PM frameworks providing structural guidance
- Stanford’s 4-axis evaluation model for comprehensive metrics
- Chain-of-Draft research enabling 92% token reduction
- Community patterns from LangChain, Anthropic, and others
Without systematic attribution, we risk:
- Appearing to claim others’ work as our own
- Missing opportunities to contribute back to the community
- Accumulating “attribution debt” similar to technical debt
- Damaging credibility and relationships with the broader AI community
Decision
We will adopt an Attribution-First Development methodology where acknowledging sources is as fundamental as writing tests or documentation.
Core Components
- CITATIONS.md Maintenance
- Living document in
docs/governance/
- Structured format with categories (Frameworks, Research, Code Patterns, Tools)
- Version controlled with semantic versioning
- Regular audits for completeness
- Pattern → Citation Mapping
# Example: When implementing CoD optimization
# Citation: Chain-of-Draft (arXiv:2502.18600v1)
def compress_reasoning(prompt: str) -> str:
"""Compress reasoning to 5-word expressions per CoD methodology"""
...
- Attribution Workflow Integration
- Pre-commit hooks check for unattributed patterns
- PR template includes “New Influences” section
- Weekly Pattern Sweep includes citation discovery
- Monthly attribution audit against codebase
- Attribution Debt Tracking
- Treat missing attributions like technical debt
- Create GitHub issues for attribution gaps
- Track metrics: attributed vs. unattributed patterns
- Regular debt reduction sprints
- Operational Tooling
```yaml
.pre-commit-config.yaml
- id: attribution-check
name: Verify Attribution
entry: scripts/validation/citation_checker.py
files: .(py|md)$
```
Attribution Standards
Required Attribution:
- External frameworks adapted or implemented
- Research papers influencing architecture
- Code patterns from other projects
- Significant algorithmic approaches
- Theoretical concepts applied
Attribution Format:
- In code: Comment with citation at point of use
- In docs: Inline references with full citation in CITATIONS.md
- In ADRs: Explicit “Influences” section
- In commits: Note new attributions in message
Not Required (but encouraged):
- Common programming patterns
- Standard library usage
- Trivial implementations
- General knowledge
Consequences
Positive
- Ethical Leadership: Sets industry standard for AI development integrity
- Community Building: Creates goodwill and collaboration opportunities
- Knowledge Lineage: Clear tracking of intellectual heritage
- Legal Protection: Reduces risk of plagiarism claims
- Learning Acceleration: The Attribution Flywheel:
Discover → Attribute → Share → Community Learns →
→ They Share Back → We Discover More
Negative
- Development Overhead: ~5-10 minutes per PR for attribution review
[Confidence: Medium - Estimated based on similar review processes, needs measurement]
- Initial Debt: Need to retrofit existing code with attributions
- Tooling Investment: Time to build attribution infrastructure
- False Positives: Attribution checker may flag common patterns
Neutral
- Cultural Shift: Team must internalize attribution as core practice
- Process Evolution: Workflows will adapt around attribution needs
- External Visibility: Our attribution practices become public
Implementation Plan
Phase 1: Foundation (Week 1)
- Create initial CITATIONS.md with known influences
- Add attribution section to PR template
- Document attribution standards in contributing guide
Phase 2: Automation (Week 2)
- Implement basic citation_checker.py
- Add pre-commit hook (warning only)
- Integrate with Pattern Sweep
Phase 3: Enforcement (Week 3)
- Citation checker becomes blocking
- Attribution debt dashboard
- Monthly audit process established
Phase 4: Maturity (Month 2+)
- Pattern → Citation mapping database
- Automated attribution suggestions
- Community attribution contributions
Alternatives Considered
Alternative 1: Ad-hoc Attribution
Description: Attribute on case-by-case basis without system
Rejected Because: Inconsistent, easily forgotten under pressure
Alternative 2: Legal Minimum Only
Description: Only attribute where legally required
Rejected Because: Misses opportunity for community building and ethical leadership
Alternative 3: Post-hoc Attribution
Description: Add attributions during quarterly reviews
Rejected Because: Attribution debt accumulates, context lost
References and Influences
- Agent Charter Framework: Rahul Vir’s “The Agentic PM’s Guide” (2025)
- Technical Debt Metaphor: Ward Cunningham (1992)
- Open Source Attribution Practices: Apache Foundation, Linux Kernel
- Academic Citation Standards: ACM, IEEE guidelines
- Ethics in AI Development: Partnership on AI recommendations
- ADR-015: Wild Claim Verification Protocol (ensures cited metrics are valid)
- ADR-000: Meta-Platform Vision (attribution as platform differentiator)
- Agent Charter v1.0: Section 10 (Attribution and Acknowledgments)
Notes
This ADR itself demonstrates attribution-first development by explicitly crediting all influences on our attribution methodology. The recursive nature (attributing our attribution approach) exemplifies the principle.
The 5-10 minute overhead per PR is estimated based on:
- 2 minutes: Developer self-check
- 2 minutes: Automated tool run
- 1 minute: Review suggestions
- 5 minutes: Add missing attributions (worst case)
Success metrics:
- 100% of external patterns attributed within 30 days
[Target: Aspirational but achievable with systematic review]
- Zero attribution debt growth after Month 1
[Target: Requires disciplined process]
- Positive community feedback on attribution practices
[Measure: Survey and social media sentiment]
- At least one upstream contribution based on attribution relationships
[Target: Demonstrates reciprocal value]