Composition Science Knowledge Graph Specification

Vision

The project is not a collection of documents. It is a connected knowledge graph where every concept, observation, experiment, law, and implementation can be traced to evidence.

Core Node Types

Prefix Type Purpose

GN Genome Node Canonical concepts LAW Candidate Law Testable hypotheses EVD Evidence Research papers and findings OBS Observation Raw observations EXP Experiment Validation studies MET Metric Quantitative measures CASE Case Study Real-world examples RULE Design Rule Practical guidance COMP Component UI/design components IMPL Implementation HTML, CSS, tokens GOV Governance Project governance

Relationship Types

Every relationship should be directional.

• supports
• contradicts
• derives_from
• belongs_to
• influences
• measures
• validates
• predicts
• implements
• references
• supersedes

Example

EVD-042 --supports--> LAW-013
LAW-013 --belongs_to--> GN-511
LAW-013 --influences--> GN-520
MET-007 --measures--> GN-511
EXP-003 --validates--> LAW-013
RULE-004 --derives_from--> LAW-013
COMP-021 --implements--> RULE-004

Required Metadata

Every node should have:

• Stable ID
• Title
• Type
• Version
• Status
• Confidence
• Author
• Creation Date
• Last Updated
• References

Confidence Model

Confidence applies to nodes and relationships.

• Very High
• High
• Moderate
• Low
• Very Low
• Unknown

Design Principles

1. Every claim should be traceable.
2. Relationships are first-class data.
3. Contradictory evidence is preserved.
4. Nothing becomes a design rule without evidence.
5. Every implementation traces back to biological or cognitive mechanisms whenever possible.

Roadmap

Phase 1: - Complete ontology. - Assign IDs. - Link existing documents.

Phase 2: - Build relationship graph. - Add evidence and confidence.

Phase 3: - Generate visual graph. - Enable semantic queries.

Long-Term Goal

The graph becomes the reasoning engine behind future tools that explain, evaluate, and generate composition decisions instead of merely copying existing designs.