Phase 3 — GoalsCopy Anchor Link

Track A — Memory Schema & Data FoundationCopy Anchor Link

Goal 3.A: Establish the memory schema with HNSW, temporal validity, and multi-label taxonomyCopy Anchor Link

Milestones:

• 3.1.1 — Postgres schema v2 (completed): expand foundation ibex_core.memories (from G5.M1 / ADR-0047) with HNSW vector index (m=16, ef_construction=64), quality columns, generated search_vector, bounded embedding_model/metadata CHECKs. Labels + primary sync already shipped in ADR-0048; expand ADR is ADR-0052 (not ADR-0040).

Acceptance criteria:

• Migration applies cleanly from zero via make db-migrate
• HNSW index type confirmed in CI assertion (not just present — type-checked)
• RLS integration test proves cross-org denial on memories (labels covered in 2.5.G5.M2)
• Primary-category sync verified via existing trigger (insert 2 labels, differing confidence, correct one wins)

---

Track B — Vector Store & Embedding IntegrationCopy Anchor Link

Goal 3.B: Pluggable vector store with category-conditional composite scoringCopy Anchor Link

Milestones:

• 3.2.1 — VectorStore interface + pgvector-HNSW implementation: Shipped — VectorStore ABC, PgVectorStore, composite scoring v2, embedding client, HNSW benchmarks (ADR-0053).

Acceptance criteria:

• VectorStore ABC with PgVectorStore implementation, ef_search set per-transaction
• Composite scoring v2 unit-tested against documented formula
• Benchmark run at 10K/100K rows committed (1M full-profile optional per ADR-0061)
• embedding_client.py has zero direct ML library imports

---

Track C — Memory Write PipelineCopy Anchor Link

Goal 3.C: 9-step write pipeline with PII detection, dedup, and temporal conflict resolutionCopy Anchor Link

Milestones:

• 3.C.1 — PII detection and redaction: Shipped — Presidio two-tier detection, redaction before embed (ADR-0054).
• 3.C.2 — Content and near-duplicate detection: Shipped — exact + near-dup dedup (ADR-0055).
• 3.C.3 — Temporal-interval-aware conflict detection: Shipped — interval overlap + sequential supersede (ADR-0056).
• 3.C.5 — Write-path persistence & orchestration: Shipped — ADR-0057.
• 3.C.4 — Multi-label memory classification: Shipped — ADR-0048 (extraction-worker prompt is Phase 3.5).

Acceptance criteria:

• Sequential-fact fixture classified as supersedes with zero LLM calls
• Genuinely overlapping-interval contradiction still escalates to LLM correctly
• PII detection: >95% recall on structured PII labeled fixture
• Redaction happens strictly before embedding (pipeline ordering test)

---

Track D — Read/Ranking Pipeline & Hot CacheCopy Anchor Link

Goal 3.D: Semantic search, composite scoring, and hot cacheCopy Anchor Link

Milestones:

• 3.D.1 — Semantic search read path: Shipped — VectorStore.search(), FTS fallback (ADR-0058).
• 3.D.2 — Composite scoring in search ranking: Shipped — composite ordering in search (ADR-0053).
• 3.D.3 — Hot cache (Redis sorted set): Shipped — MemoryHotCacheReader, atomic top-50 trim (ADR-0059).

Acceptance criteria:

• find_similar uses HNSW index (confirmed via EXPLAIN ANALYZE)
• Cross-org result leakage impossible (integration test)
• Hot cache score formula reuses CATEGORY_HALF_LIFE_DAYS

---

Track E — Exit GateCopy Anchor Link

Goal 3.E: Prove the composed system enforces isolation and meets quality barsCopy Anchor Link

Milestones:

• 3.E.1 — Memory security integration suite (TestMemory_ISO_*): Shipped — 12 ISO cases, memory-security-integration CI, Tier-1 read-path PII guard.
• 3.E.2 — End-to-end memory lifecycle test: Shipped — make e2e-smoke-p3-memory, e2e-smoke-p3-memory CI job.
• 3.E.3 — Retrieval-quality benchmark sign-off: Shipped — HNSW + ranking + write CI regression gates (ADR-0061).
• 3.E.4 — ADR consolidation: Shipped — ADR-0060, ADR-0061; phase register/index reconciled (decisions).
• 3.E.5 — Phase 3 exit audit: Shipped — gap register, summary; zero open P0 gaps.

Acceptance criteria:

• All ISO-* cases pass with real containers
• memory-security-integration CI job required on main
• Benchmark baseline committed; CI fails on regression
• Zero open P0 gaps in exit audit