 Profa
|
e154b6bf40
|
feat(v3): implement RuVector Neural Backbone with persistent learning (ADR-050)
GOAP Implementation - 8/8 Actions Complete:
1. ML Observability Layer - Track ML vs fallback usage
2. PersistentQLearningRouter - Q-values persist across sessions with EWC++
3. PersistentSONAEngine - SONA patterns persist across sessions
4. Remove Silent Fallbacks - All fallbacks now observable
5. Hypergraph Schema - SQLite-backed code graph storage
6. HypergraphEngine - Query engine for code intelligence
7. Coordinator Integration - All 7 domain coordinators + provider use persistent engines
8. RuVector Server Client - Stubbed with supportsVectorOperations() guard
Key Changes:
- All domain coordinators migrated to PersistentSONAEngine
- DefaultRuVectorClient uses PersistentQLearningRouter
- RuVectorServiceProvider.getGlobalSONA() now async with persistence
- HypergraphEngine integrated in code-intelligence coordinator
- 491+ tests passing, TypeScript clean
README Updates:
- Performance table updated with verified claims only
- Removed unverified speedup claims
- Added persistent learning features (EWC++, Q-value persistence)
New Files:
- persistent-q-router.ts - Q-Learning with SQLite persistence
- sona-persistence.ts - SONA with pattern persistence
- hypergraph-schema.ts - Graph node/edge schema
- hypergraph-engine.ts - Code intelligence queries
- server-client.ts - RuVector server integration
- shared-memory.ts - Fleet shared memory
- observability.ts - ML usage tracking
Documentation:
- ADR-050: RuVector Neural Backbone specification
- BRUTAL-HONESTY-GOAP-REVIEW.md - Integration verification
- Migrated ADRs to standardized MADR format
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
|
2026-01-20 10:46:42 +00:00 |
|