SINK Packet Specification (v0.1)
Formatting and Structure for AI-Interrogable Knowledge Bundles
Purpose of This Specification
This document defines minimum formatting and structural conventions for SINK packets so they remain interoperable, interrogable, and resilient across different AI systems and readers.
It does not prescribe content, conclusions, or ideology.
It specifies how ideas should be packaged so they can be reliably explored using AI without sacrificing human readability.
This spec exists because format is not neutral in AI-mediated cognition.
1. Section Header Conventions (Semantic Anchors)
Headers in SINK documents function as semantic anchors, not decorative titles. They allow AI systems to index, reference, and triangulate ideas while preserving narrative flow for humans.
Header Levels
H1 (Document-Level)
Used only for the document title. Signals the scope and role of the document within the packet (e.g., foundational, applied, synthesis).
H2 (Concept-Level)
Represents a major concept, claim, or structural component. Each H2 section should be intelligible as a standalone unit.
H3 (Sub-Concept / Clarification-Level)
Used for refinements, implications, or internal distinctions within an H2 concept.
Avoid deeper nesting unless absolutely necessary.
Declarative vs. Interrogative Headers
Declarative headers (Thermodynamic Alignment as Constraint) are preferred for core claims and explanations.
Interrogative headers (What Does Alignment Demand of Distribution?) may be used sparingly to frame explicit tensions or open questions.
Headers should signal what kind of thinking occurs beneath them.
2. Emphasis Protocol (Semantic Weight vs. Rhetorical Emphasis)
Emphasis is a shared channel between human readers and AI systems. It must be used deliberately.
Bold Text
Use bold to mark:
- foundational concepts
- terms that recur across multiple documents phrases that serve as structural landmarks
- Bold should signal semantic importance, not emotional emphasis.
Italics
Use italics for:
- clarifications boundary-setting language conditional or contingent statements
- Avoid using italics to introduce new core concepts.
Emphasis Discipline
- Do not overuse emphasis.
- If everything is emphasized, nothing is indexed.
3. START_HERE Document (The Map Kernel)
Every SINK packet must include a START_HERE document.
This document is written for both the reader and the AI, and functions as the semantic kernel of the packet.
Required Sections
The START_HERE document should include:
What This Packet Is
A brief description of the intellectual system represented.
Document Map
A list of included documents with 12 sentence descriptions of each and their role (foundational, applied, synthesis).
Conceptual Relationships
A short explanation of how ideas recur or build across documents.
AI Instruction Block
Explicit instructions to the AI on how to treat the packet.
AI Instruction Block (Template)
You are being provided with a bundle of documents representing a single intellectual system.
Do not summarize these documents individually.
Use this document as a map to understand how they interconnect.
When the user asks a question, triangulate your response across the entire bundle and note where concepts recur, align, or conflict.
This block should be clearly labeled and visually separated.
4. Cross-Document Linking and Concept Reuse
SINK packets are networked by design.
Naming Conventions
- Use consistent names for recurring concepts across documents.
- Avoid synonyms that obscure continuity unless explicitly explained.
Repetition vs. Reference
- Repeat short definitions of foundational concepts when needed for clarity.
- Reference deeper treatments rather than re-explaining them fully.
- AI systems benefit from redundancy with consistency, not compression at all costs.
5. Length and Scope Guidelines
SINK documents are summaries, not exhaustively argued papers.
General Targets
- Foundational summaries: 3-6 pages
- Applied domain summaries: 6-10 pages
- Synthesis documents: flexible, but restrained Split documents when: a section begins introducing multiple independent concepts narrative coherence degrades under length interrogation would benefit from modularity
- Appendices may be used for operational detail or technical depth.
6. Interrogation Seed Questions
Each document should either include or be accompanied by interrogation seed questions.
Question Types
Learning questions:
(Explain the core claim of this document.)
Stress-test questions:
(What assumptions would cause this argument to fail?)
Tension-exposing questions:
(Where does this framework create tradeoffs or paradoxes?) Questions should invite challenge, not confirmation.
What SINK Is Not (Boundary Conditions)
SINK is not a reading list a persuasion tool a replacement for primary sources optimized for skimming a guarantee of consensus or correctness
SINK is a navigation layer for complex ideas under AI mediation.
Understanding emerges through interrogation, not agreement.
Versioning Note
This specification is intentionally minimal and provisional.
As AI systems and usage patterns evolve, this spec may change.
Interoperability matters more than perfection.