Phase 3: 008_phase3_compaction.sql (10.1 KB) T3.1: exact_dedup_record, stale_gc_record (Tier 1 exact dedup + GC) T3.2: semantic_dedup_record (Tier 2 LLM-gated semantic dedup) T3.3: compaction_audit, compaction_dryrun_result (audit trail + dry-run) T3.4: compaction_schedule (CronJob configuration) Features: - Pre-filter tracking (60-70% candidate elimination) - Dry-run support for all operations - Reversible actions with rollback instructions - LLM cost tracking and approval workflow - Scheduled execution with resource limits Phase 4: 006_phase4_community_detection.sql (8.7 KB) T4.1: label_propagation_run, community_member_map, community_statistics (Label propagation algorithm + membership tracking + quality metrics) T4.2: community_summary_generation (LLM-based summary versioning) T4.3: (CronJob config deferred to T3.4-style scheduling) Features: - Algorithm execution tracking (convergence, iterations) - Entity→Community membership with label strength - Cohesion metrics (density, modularity, quality scores) - Community merge history with rollback capability - Summary versioning with LLM cost tracking Aligned with original design: - PHASE3_COMPACTION_PLAN.md (21.9 KB) - phase-4-community/ (T4.1-T4.3 specifications) - PHASE2_EXTENSIONS.md (community detection section) All tables include: - Bi-temporal tracking (created_at, processed_at) - Project isolation (project_id in all tables) - Dry-run mode support - Audit trail + reversibility - Proper indexes for query performance Next: Implement T3.1-T3.4 + T4.1-T4.2 algorithms (3-4 weeks)
