Phase 3 core memory substrate

Standalone gap register with severity-tiered (P0–P3) tracking, following the Phase 1 exit-audit precedent. Covers schema drift, HNSW tuning evidence, PII false-negative rates, and benchmark baseline staleness.

Milestone 3.E.5 — Phase 3 Exit Audit Document

Status: Planned
Goal: Track E — Exit Gate
Phase: 3 — Core Memory Substrate
Estimated effort: 1 day


Why This Milestone Exists

Following the Phase 1 precedent exactly — a standalone gap register with severity tiers, not folded into the roadmap page.


Document Structure

web/engineering/archive/0XX-phase3-exit-audit.md

## Severity definitions (unchanged from Phase 1 precedent)
P0 — Blocker, should be green before Phase 3.5
P1 — High, fix at exit gate or immediately after
P2 — Hygiene, fix before phase sign-off PR
P3 — Defer, document as Phase 3.5+, not hidden

Audit scope (Tracks A–E)

  • Track A/B gaps: schema drift between migration files and DATABASE_SCHEMA.md, HNSW parameter tuning evidence completeness
  • Track C/D gaps: PII detector false-negative rate on the labeled test set, conflict-detection precision on ambiguous cases
  • Track E gaps: any ISO-* test flakiness, benchmark baseline staleness

Phase 3 Exit Criteria (Final Consolidated List)

  1. Memory write pipeline (9 steps) passes all unit + ISO-* integration tests
  2. HNSW recall@10 ≥ 98% and p99 < 100ms at 1M-row benchmark scale
  3. Temporal-interval-aware conflict detection correctly distinguishes sequential-fact vs. true-contradiction cases, with zero unnecessary LLM calls on non-overlapping intervals
  4. Retrieval-quality benchmark running in CI with a committed baseline
  5. search_memory/write_memory MCP tools function end-to-end with full ClickHouse tracing
  6. All 10+ ISO-* cross-tenant isolation cases pass, including the memory_relationships cross-org edge case
  7. GDPR delete cascade verified end-to-end, including the MinIO-partial-failure orphan-detection path
  8. All Phase 1 and Phase 2 security regression tests still pass with zero regressions
  9. make e2e-smoke-p3-memory exits 0
  10. Phase 3 memory ADRs published and cross-linked (next free numbers after ADR-0047)
  11. No proxy hot-path changes — explicitly re-verified, since that's Phase 3.5's boundary

Orientation (indicative)

Named paths, package layouts, libraries, schemas, env vars, and commands anywhere on this page are rough sketches for orientation — inspiration and a baseline, not a required change list.

During implementation, expect to:

  • open the live tree and follow existing patterns before inventing new ones
  • research current constraints (latency, tenancy, deploy shape, libraries) more deeply than this page can
  • advance the design beyond the sketch where measurement or code reality says so
  • land work in different filenames, merged packages, deferred docs, or new surfaces when the situation calls for it

Prefer outcomes over matching any particular file tree or command sequence.

Areas that may be involved (situational — not a checklist):

  • Evaluation / benchmarks
  • Memory service / repositories

Success signals

Outcome-oriented signals that the milestone is in good shape. Exact filenames, package layouts, and commands may differ from any sketches above.

  • Exit audit document committed with all severity-tiered gaps populated
  • All P0 items resolved before Phase 3.5 begins
  • Exit criteria list verified against actual test/benchmark results
  • Repo guards / CI checks still pass

Prerequisites

  • Milestones 3.E.1–3.E.4
Edit on GitHub

Last updated on

On this page

0%