proffesor-for-testing__agentic-qe/docs/V3-PERSISTENCE-ANALYSIS.md

V3 Persistence & Self-Learning System Analysis

Date: 2026-02-03 Status: Investigation Complete

Database Tables Analysis

V3 Core Tables (ESSENTIAL - defined in unified-memory.ts)

Table	Purpose	Self-Learning Role
schema_version	Migration tracking	Infrastructure
kv_store	Key-value storage (v2 compatible)	Pattern storage, metrics
vectors	Vector embeddings for HNSW search	Semantic pattern matching
rl_q_values	Reinforcement learning Q-values	Agent action learning
goap_goals	Goal-Oriented Action Planning	Planning system
goap_actions	GOAP actions	Action library
goap_plans	GOAP execution plans	Plan reuse
goap_execution_steps	GOAP step execution	Execution tracking
goap_plan_signatures	Plan similarity matching	Plan deduplication
concept_nodes	Dream engine concept graph	Pattern discovery
concept_edges	Dream engine relationships	Association learning
dream_cycles	Dream cycle records	Consolidation tracking
dream_insights	Generated insights	Novel discoveries
qe_patterns	QE patterns (ReasoningBank)	Core learning storage
qe_pattern_embeddings	Pattern embeddings	Semantic search
qe_pattern_usage	Pattern usage tracking	Effectiveness learning
qe_trajectories	Learning trajectories	Experience capture
embeddings	Embedding cache	Performance optimization
execution_results	Plan execution results	Success tracking
executed_steps	Step execution tracking	Granular metrics
mincut_snapshots	MinCut graph snapshots	System health
mincut_history	MinCut history	Trend analysis
mincut_weak_vertices	Bottleneck detection	Self-healing
mincut_alerts	MinCut alerts	Alerting
mincut_healing_actions	Self-healing history	Recovery learning
mincut_observations	Strange Loop observations	Self-organization
sona_patterns	Neural backbone patterns	Deep learning
hypergraph_nodes	Hypergraph storage	Complex relationships
hypergraph_edges	Hypergraph edges	Multi-way associations

V2 Legacy Tables (Can be migrated/archived)

Table	Origin	Status
memory_entries	V2 memory store	Superseded by kv_store
patterns	V2 patterns	Superseded by qe_patterns
events	V2 events	Superseded by kv_store events namespace
sessions	V2 sessions	Empty (0 records)
hints	V2 hints	Stale (Dec 2025)
learning_experiences	V2 learning	Superseded by qe_trajectories
learning_history	V2 learning	Empty (0 records)
learning_metrics	V2 metrics	Empty (0 records)
experiences	V2 experiences	Superseded by captured_experiences
agent_registry	V2 agent tracking	Superseded by fleet system
performance_metrics	V2 performance	Empty (0 records)
consensus_state	V2 consensus	Unused
workflow_state	V2 workflow	Empty (0 records)
q_values	V2 Q-learning	Superseded by rl_q_values

Background Workers for Self-Learning

Core Workers (11 total)

Worker	Interval	Learning Role
LearningConsolidationWorker	30 min	Critical: Pattern consolidation, dream cycles
TestHealthWorker	5 min	Test suite health metrics
CoverageTrackerWorker	10 min	Coverage gap detection
FlakyDetectorWorker	15 min	Flaky test pattern learning
SecurityScanWorker	30 min	Vulnerability pattern learning
QualityGateWorker	5 min	Quality metric learning
DefectPredictorWorker	15 min	ML defect prediction
RegressionMonitorWorker	10 min	Regression pattern learning
PerformanceBaselineWorker	1 hour	Performance trend learning
ComplianceCheckerWorker	30 min	ADR/DDD compliance learning
CloudSyncWorker	5 min	Cross-instance sync

Worker Status

• Workers directory: Created at .agentic-qe/workers/
• Registry: Created at .agentic-qe/workers/registry.json
• Daemon script: Created at .agentic-qe/workers/start-daemon.sh

MCP Server Persistence

How It Works

1. HybridMemoryBackend (v3/src/kernel/hybrid-backend.ts) connects to memory.db
2. All MCP tools share the backend via getSharedMemoryBackend()
3. Fleet handlers use kernel memory which wraps the backend
4. Tool base class ensures all tools persist to the same database

Connection Flow

MCP Tool → getSharedMemoryBackend() → HybridMemoryBackend → UnifiedMemoryManager → memory.db

Current Issues Found

1. Workers not running: Directory didn't exist (now created)
2. Dream scheduler inactive: Last cycle Jan 2, 2026
3. No processes calling persistence: MCP server needs to be running
4. 47 dream cycles stuck in "running": Previous crashes without cleanup

Recommendations

Immediate Actions

1. ✅ Created workers directory and registry
2. ✅ Created daemon startup script
3. Start MCP server: npm run mcp
4. Run dream cycle: qe/learning/dream { action: "dream" }

Init Integration

The aqe init --auto should:

1. Create .agentic-qe/ directory structure
2. Initialize memory.db with v3 schema
3. Create workers registry
4. Configure MCP server in .mcp.json

Tables to Keep (V3 Essential)

All 28 tables defined in unified-memory.ts are essential for v3 self-learning.

Tables Safe to Archive

The 14 V2 legacy tables can be archived to a separate database for historical reference.