Memory Core Concept Discovery & Glossary Enhancement

Objective

Use Memory Core session summaries, tickets, and the Knowledge Base to discover concepts not yet in the ontology — the "unknown unknowns."

Context

Phase 3, Step 2 of the Concept Ontology epic (#10030). This is the ongoing maintenance and discovery layer. Depends on the integrated concept graph from #10035.

Deliverables

1. Session-Based Concept Mining

Query the Memory Core for:

• Session summaries containing recurring architectural terms
• High strategic_weight topics from get_context_frontier
• Concepts that appear across 3+ sessions (Hebbian reinforcement signal)

Cross-reference against existing concept nodes. New terms that pass the Teaching Test → candidate concepts.

2. Ticket-Based Concept Mining

Scan GitHub issues (especially epics and architecture-labeled tickets) for:

• Architectural vocabulary not yet in the ontology
• Feature areas implicitly defining new concepts

3. KB-Assisted Classification

For UNCATEGORIZED classes (from #10032), use ask_knowledge_base:

"What architectural concept does Neo.core.EffectManager implement?"
→ "Pull-based reactivity / automatic dependency tracking"

If the answer identifies a concept not in the ontology → discovered unknown unknown.

4. Confidence Gating

• LLM-discovered concepts start at tier: 3, confidence: 0.5
• Only human curation or repeated session reinforcement promotes to higher tiers
• Prevents LLM noise from flooding the ontology

5. Glossary Enhancement (Downstream)

Once the concept graph is established, generate a diff showing:

• Concepts that should be in the Glossary but aren't
• Glossary terms that don't map to any concept (candidates for removal or reclassification)

Acceptance Criteria

• Script or service can query Memory Core for concept candidates
• Discovered concepts correctly flagged with low confidence
• Human review gate exists before promotion
• Glossary gap report generated

Note

This sub-issue requires the KB MCP server timeout issue to be resolved first — ask_knowledge_base currently times out due to ChromaDB cold-start latency exceeding the 60s MCP SDK default.

Origin Session ID: 496c0120-29aa-4b02-a804-d5a32ff6b5eb