Seed
Minimal unit of cognition:
- conversation fragment
- hypothesis
- reflection
- code change
- test output
Role: initiates graph growth
Stream
Continuous temporal flow of seeds:
- chronological + causal + semantic mixture
- preserves “thinking in motion”
Knowledge Graph Node Types
- Concept Node: stabilized idea (recurring patterns)
- Hypothesis Node: structured claim about relationships
- Execution Node: runtime event (“what happened”)
- Reflection Node: interpretation (“what it meant”)
- Conversation Node: raw input trace
Edge Types (semantic relations)
supportscontradictstransforms_intoinfluencesimplementsderives_fromtriggersvalidates / invalidates
Edges are not structural only—they encode epistemic meaning
Signal
Observational evidence derived from execution or tests:
- intensity
- drift
- alignment
- propagation effects
Signal replaces binary correctness.
Hypothesis Object
Structured reasoning unit:
- claim about graph structure
- expected behavior
- evaluation method
- confidence + uncertainty
Seed Stream (as system substrate)
All interaction becomes:
“continuous emission of cognitively meaningful events”
No distinction between:
HOW THE CONCEPT WORKS
1. Continuous Ingestion
Every event becomes a seed:
- user message
- AI response
- test execution
- runtime trace
- system log
Each seed is:
- normalized
- timestamped
- embedded
- linked into graph
2. Immediate Graph Mutation
Instead of batch processing:
- each seed updates graph topology in real time
- edges are added as:
- causal links
- semantic similarity
- hypothesis-generated relations
Graph is never “rebuilt”—it is continuously drifting
3. Seed → Hypothesis → Signal Loop
A recursive cycle:
- Seed arrives
- AI generates hypotheses about structure
- Hypotheses are tested via:
- execution probes
- signal-based tests
- Signals feed back into graph
- New seeds emerge from results
This creates:
a self-feeding epistemic engine
4. Signal-Based Testing Layer
Tests become:
- probes into conceptual structure
- not pass/fail checks
Outputs:
- drift scores
- alignment vectors
- concept activation intensity
Failures become:
“unresolved tension fields in the graph”
5. Dual-Layer Storage Model
- Postgres (Seed Layer): immutable event stream
- Neo4j (Graph Layer): evolving interpretation structure
Separation:
- truth log vs interpretive model
6. AI as Graph Evolution Operator
AI performs:
- node creation
- edge inference
- clustering
- contradiction detection
- hypothesis generation
- structural compression
Not a chatbot:
a continuous graph rewriting engine
7. Reflection Layer (Recursive Meta-Graph)
The system stores:
- “what was meant”
- “why it was believed”
- “how it changed”
Reflections:
- modify graph structure
- not just describe it
8. Concept Stabilization
Nodes become concepts when:
- repeated recurrence across streams
- cross-context reinforcement
- persistent signal alignment
But instability is preserved intentionally:
- contradictions remain visible
- multiple interpretations coexist