Decision Rights¶
This document defines who can make and approve which kinds of decisions in this repository. Roles are provisional and role-based (not tied to named individuals) as described in document-metadata-standard.md. See docs/ai/agent-role-definitions.md for the fuller catalog of AI-agent-facing operating roles; this document focuses specifically on approval authority.
Provisional roles¶
- Founder — ultimate decision authority for Devonshire Digital LLC and, absent a delegated Venture Lead, for its ventures. Approves company policy, company ADRs, legal and financial commitments, and anything not otherwise delegated.
- Product Owner — owns a specific product's roadmap and specification decisions within a venture, subject to venture and company constraints.
- Venture Lead — owns venture-level strategy, venture standards, and venture ADRs, subject to company-level constraints (see authority-and-inheritance.md).
- Repository Maintainer — owns repository structure, tooling, documentation standards, and the mechanics of how documents move through their lifecycle. Does not by itself have authority to approve policy content.
- Domain Reviewer — a subject-matter reviewer for a specific domain (e.g. a nutrition/food domain reviewer for Shelfery, a financial-modeling reviewer for Digital Products) who signs off on domain-specific accuracy.
- Security and Privacy Reviewer — reviews and approves security and privacy-relevant changes, including any handling of classified information (see information-classification.md).
- AI Agent — prepares drafts, research, recommendations, and routine documentation work under the constraints in docs/ai/. Never holds final approval authority.
What an AI agent may and may not do¶
An AI agent operating in this repository may prepare, draft, analyze, and recommend. It may not independently approve:
- Company policy.
- Legal claims (including claims about licensing, compliance, or contractual obligations).
- Trademark conclusions.
- Security exceptions.
- Financial advice.
- Medical advice.
- Food-safety claims.
- Public commercialization claims (marketing copy asserting capability, availability, or results).
- Destructive repository changes (deletion or irreversible rewriting of authoritative history or documents).
For any of the above, the agent's output is a recommendation or draft with status: draft or
status: proposed. Moving such a document to status: approved, or acting on the substance of
the recommendation in a way that has real-world effect, requires the appropriate human role above.
See docs/ai/human-review-and-approval.md for how that review
is requested and recorded, and
docs/ai/task-classification-and-routing.md for how
this maps onto the risk classification of a task.
Escalation¶
If it is unclear which role has authority over a given decision, or if a decision spans multiple roles' territory, escalate to the Founder rather than guessing. See docs/ai/escalation-and-ambiguity.md.