SpillwaveSolutions · RichardHightower · Sep 2, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -236,11 +236,63 @@ jobs:
           print("smoke ok", d["metric"], "questions", d["total_questions"])
           PY
 
+  bench-cli-smoke:
+    name: Benchmark CLI Isolation Smoke
+    runs-on: ubuntu-24.04
+    needs: [build]
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install system dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y protobuf-compiler libclang-dev
+
+      - name: Install Rust
+        uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: "1.97"
+
+      - name: Cache cargo registry
+        uses: Swatinem/rust-cache@v2
+        with:
+          shared-key: "bench-cli-smoke"
+
+      - name: Build daemon, CLI, and bench
+        run: cargo build -p memory-daemon -p memory-cli -p memory-bench
+
+      - name: Live-backend locomo smoke (spawn-per-conversation)
+        run: |
+          export PATH="$PWD/target/debug:$PATH"
+          cargo run -p memory-bench -- locomo \
+            --dataset benchmarks/fixtures/locomo-smoke.json \
+            --backend cli \
+            --scorer mock \
+            --isolation daemon-per-conversation \
+            --output /tmp/locomo-cli-smoke.json
+          python3 - <<'PY'
+          import json
+          d = json.load(open("/tmp/locomo-cli-smoke.json"))
+          assert d["conversations"] == 1, d
+          assert d["isolation"] == "per-conversation daemon", d
+          assert d["metric"] == "context_hit_rate", d
+          conv = d["per_conversation"][0]
+          assert "drain_wait_ms" in conv, conv
+          assert conv["drain_wait_ms"] >= 0
+          print("cli smoke ok", d["isolation"], "drain_wait_ms", conv["drain_wait_ms"])
+          PY
+
+      - name: Isolation bleed test
+        env:
+          MEMORY_BENCH_LIVE: "1"
+        run: |
+          export PATH="$PWD/target/debug:$PATH"
+          cargo test -p memory-bench cli_isolated_daemons_do_not_bleed -- --nocapture
 
   # Summary job that depends on all other jobs
   ci-success:
     name: CI Success
-    needs: [fmt, clippy, test, build, doc, e2e, benchmark-smoke, release-guards]
+    needs: [fmt, clippy, test, build, doc, e2e, benchmark-smoke, bench-cli-smoke, release-guards]
     runs-on: ubuntu-24.04
     if: always()
     steps:
@@ -253,6 +305,7 @@ jobs:
              [[ "${{ needs.doc.result }}" != "success" ]] || \
              [[ "${{ needs.e2e.result }}" != "success" ]] || \
              [[ "${{ needs.benchmark-smoke.result }}" != "success" ]] || \
+             [[ "${{ needs.bench-cli-smoke.result }}" != "success" ]] || \
              [[ "${{ needs.release-guards.result }}" != "success" ]]; then
             echo "One or more jobs failed"
             exit 1

diff --git a/.planning/MILESTONES.md b/.planning/MILESTONES.md
@@ -8,11 +8,12 @@ and a repo whose backlog is public.
 
 **Spec:** `docs/plans/v3.2-prove-it-plan.md`
 
-**Phases:** 59 Guardrails and Inventory (executing), 60 Real Numbers, 61
-Operate It, 62 Cross-encoder rerank (conditional on #39).
+**Phases:** 59 Guardrails and Inventory (complete #45), 60 Real Numbers
+(60-01 executing), 61 Operate It, 62 Cross-encoder rerank (conditional on #39).
 
-**Known Gaps (now issues):** #39 LOCOMO run, #40 vector/topic quality, #41
-backfill, #42 install-service, #43 TOC rebuild, #44 cross-encoder.
+**Known Gaps (issues labelled `v3.2`):** #39 LOCOMO run, #40 vector quality,
+#47 topic quality, #41 backfill, #42 install-service, #43 TOC rebuild,
+#48 uninstall/status, #44 cross-encoder.
 
 ---
 

diff --git a/.planning/PROJECT.md b/.planning/PROJECT.md
@@ -15,9 +15,9 @@ them provable.
 **Target work:**
 - Release pipeline guards (tag on main, crate version matches tag, all five platforms) — Phase 59
 - Committed LOCOMO LLM-judge result on the real dataset — Phase 60 / #39
-- Quality fixtures for vector search and the topic graph — Phase 60 / #40
+- Quality fixtures for vector search (#40) and the topic graph (#47) — Phase 60
 - Backfill, `install-service`, offline TOC rebuild, panic audit — Phase 61 / #41 #42 #43
-- Claude Code plugin registration + installer uninstall/status — Phase 61
+- Claude Code plugin registration + installer uninstall/status — Phase 61 / #48
 - Cross-encoder rerank only if 60-02 says retrieval is the bottleneck — Phase 62 / #44
 
 **Previous version:** v3.1.0 (Shipped 2026-09-01) — Make It True. No new
@@ -37,8 +37,8 @@ The system implements a complete 6-layer cognitive stack with control plane, mul
 - Layer 6: Ranking Policy (salience, usage, novelty, lifecycle) + StaleFilter (time-decay, supersession)
 - Control: Retrieval Policy (intent routing, tier detection, fallbacks) + MemoryOrchestrator (RRF fusion, optional LLM rerank, explainability)
 - Dedup: InFlightBuffer + HNSW composite gate, configurable threshold, fail-open
-- Installer: memory-installer crate with RuntimeConverter trait, 5 converters (Claude, Gemini, Codex, Copilot, generic skills), tool mapping tables
-- Adapters: Claude Code, Gemini CLI, Copilot CLI, Codex CLI (via installer). OpenCode removed in v3.1 Phase 57 — the converter reported success and wrote nothing
+- Installer: memory-installer crate with RuntimeConverter trait, converters for Claude, Gemini, Codex, Copilot, generic skills; tool mapping tables
+- Adapters: Claude Code, Gemini CLI, Copilot CLI, Codex CLI (via installer). Supported surfaces are those four; registration for Gemini/Codex/Copilot is v3.3+
 - Discovery: ListAgents, GetAgentActivity, agent-filtered topics
 - Testing: 1,205 workspace + 60 e2e cargo tests; 114 bats CLI tests; Tier 1 (Claude Code, Codex) gates PRs, Tier 2 (Gemini, Copilot) weekly
 - CI/CD: Dedicated E2E job + CLI matrix report; rust-toolchain pinned to 1.97

diff --git a/.planning/REQUIREMENTS.md b/.planning/REQUIREMENTS.md
@@ -1,6 +1,6 @@
-# Requirements: Agent Memory v3.0
+# Requirements: Agent Memory
 
-**Defined:** 2026-03-22
+**Defined:** 2026-03-22 (v3.0); extended 2026-09-01 (v3.2)
 **Core Value:** Agent can answer "what were we talking about last week?" without scanning everything
 
 ## v3.0 Requirements
@@ -42,59 +42,105 @@ Requirements for the Competitive Parity & Benchmarks milestone. Each maps to roa
 - [x] **BENCH-07**: CI runs benchmark suite (non-blocking, skips LOCOMO without `--dataset` flag)
 - [x] **BENCH-08**: JSON + markdown report output for all benchmark types
 
-## Future Requirements (v3.1+)
+## v3.2 Requirements (Prove It)
 
-- **ORCH-F01**: Cross-encoder reranking (requires new inference path in memory-embeddings)
+Requirements for making v3.1's claims provable and operable. Each maps to
+exactly one plan in `docs/plans/v3.2-prove-it-plan.md`.
+
+### Release pipeline (REL)
+
+- [x] **REL-01**: Tagged commit must be an ancestor of `origin/main` (59-01)
+- [x] **REL-02**: `workspace.package.version` must equal the tag minus `v` (59-01)
+- [x] **REL-03**: Any failed platform build → no GitHub release (59-01)
+- [x] **REL-04**: Release body is the matching CHANGELOG section (59-01)
+
+### Benchmarks (BENCH)
+
+- [x] **BENCH-10**: Per-conversation isolation on `--backend cli` (60-01)
+- [x] **BENCH-11**: Deterministic drain wait (poll checkpoints, no blind sleep) (60-01)
+- [ ] **BENCH-12**: Committed `locomo_llm_judge` full-dataset result (60-02 / #39)
+- [ ] **BENCH-13**: Layer switch `bm25|vector|hybrid` on the custom harness (60-03)
+
+### Quality evidence (QUAL)
+
+- [ ] **QUAL-01**: Semantic fixture set, ≥15 tests (60-03 / #40)
+- [ ] **QUAL-02**: Topic clustering purity + ARI artifact (60-03 / #47)
+- [ ] **QUAL-03**: README "Solid" rows cite committed artifacts (60-03)
+
+### Operate it (OPS)
+
+- [ ] **OPS-01**: `admin backfill-index` resumable, idempotent (61-01 / #41)
+- [ ] **OPS-02**: `install-service`/`uninstall-service` macOS+Linux (61-02 / #42)
+- [ ] **OPS-03**: Zero fallible `unwrap` on request paths (61-03)
+- [ ] **OPS-04**: Hostile-input e2e over all RPCs (61-03)
+- [ ] **OPS-05**: `admin rebuild-toc` real (61-04 / #43)
+
+### Installer (INST)
+
+- [ ] **INST-01**: Claude Code plugin registration (CREG/META) (61-05)
+- [ ] **INST-02**: `memory-installer uninstall` (61-05 / #48)
+- [ ] **INST-03**: `memory-installer status` (61-05 / #48)
+
+## Future Requirements (v3.3+)
+
+- **ORCH-F01**: Cross-encoder reranking — Phase 62 *if* 60-02 shows retrieval is the bottleneck (#44)
 - **CLI-F01**: REST/HTTP endpoint wrapping CLI commands
 - **CLI-F02**: Python SDK wrapping CLI binary
 - **BENCH-F01**: Continuous benchmark regression tracking in CI
+- **REG-F01**: Gemini/Codex/Copilot plugin registration
+- **INST-F01**: `--for all` / `--all` installer flags
+- **OPS-F01**: Windows service install; true double-fork daemonization
 
 ## Out of Scope
 
 | Feature | Reason |
 |---------|--------|
-| REST/HTTP endpoint | Future milestone — CLI-first for v3.0 |
+| REST/HTTP endpoint | Future milestone — CLI-first |
 | Python SDK | Future milestone — wraps CLI |
 | Memory views UI | Future milestone |
-| Cross-encoder reranking | Requires new inference path in memory-embeddings; extension point only |
+| Cross-encoder reranking | Conditional Phase 62; extension point only until then |
 | Multi-agent shared memory changes | Shipped in v2.1 |
 
 ## Traceability
 
+### v3.0 (complete)
+
 | Requirement | Phase | Status |
 |-------------|-------|--------|
-| ORCH-01 | Phase 51 | Complete |
-| ORCH-02 | Phase 51 | Complete |
-| ORCH-03 | Phase 51 | Complete |
-| ORCH-04 | Phase 51 | Complete |
-| ORCH-05 | Phase 51 | Complete |
-| ORCH-06 | Phase 51 | Complete |
-| ORCH-07 | Phase 51 | Complete |
-| ORCH-08 | Phase 51 | Complete |
-| CLI-01 | Phase 52 | Complete |
-| CLI-02 | Phase 52 | Complete |
-| CLI-03 | Phase 52 | Complete |
-| CLI-04 | Phase 52 | Complete |
-| CLI-05 | Phase 52 | Complete |
-| CLI-06 | Phase 52 | Complete |
-| CLI-07 | Phase 52 | Complete |
-| CLI-08 | Phase 52 | Complete |
-| CLI-09 | Phase 52 | Complete |
-| CLI-10 | Phase 52 | Complete |
-| BENCH-01 | Phase 53 | Complete |
-| BENCH-02 | Phase 53 | Complete |
-| BENCH-03 | Phase 53 | Complete |
-| BENCH-04 | Phase 53 | Complete |
-| BENCH-05 | Phase 53 | Complete |
-| BENCH-06 | Phase 53 | Complete |
-| BENCH-07 | Phase 53 | Complete |
-| BENCH-08 | Phase 53 | Complete |
+| ORCH-01..08 | Phase 51 | Complete |
+| CLI-01..10 | Phase 52 | Complete |
+| BENCH-01..08 | Phase 53 | Complete |
+
+### v3.2
+
+| Requirement | Plan | Status |
+|-------------|------|--------|
+| REL-01 | 59-01 | Complete (#45) |
+| REL-02 | 59-01 | Complete (#45) |
+| REL-03 | 59-01 | Complete (#45) |
+| REL-04 | 59-01 | Complete (#45) |
+| BENCH-10 | 60-01 | In progress |
+| BENCH-11 | 60-01 | In progress |
+| BENCH-12 | 60-02 | Open (#39) |
+| BENCH-13 | 60-03 | Open (#40) |
+| QUAL-01 | 60-03 | Open (#40) |
+| QUAL-02 | 60-03 | Open (#47) |
+| QUAL-03 | 60-03 | Open |
+| OPS-01 | 61-01 | Open (#41) |
+| OPS-02 | 61-02 | Open (#42) |
+| OPS-03 | 61-03 | Open |
+| OPS-04 | 61-03 | Open |
+| OPS-05 | 61-04 | Open (#43) |
+| INST-01 | 61-05 | Open |
+| INST-02 | 61-05 | Open (#48) |
+| INST-03 | 61-05 | Open (#48) |
 
 **Coverage:**
-- v3.0 requirements: 26 total
-- Mapped to phases: 26
+- v3.0 requirements: 26 total, all complete
+- v3.2 requirements: 19 total, 4 complete (REL), 15 open
 - Unmapped: 0 ✓
 
 ---
 *Requirements defined: 2026-03-22*
-*Last updated: 2026-03-22 after spec review*
+*Last updated: 2026-09-01 — v3.2 IDs added from the adopted Prove It plan*
+
diff --git a/.planning/ROADMAP.md b/.planning/ROADMAP.md
@@ -13,7 +13,7 @@
 - ✅ **v2.7 Multi-Runtime Portability** — Phases 45-50 (shipped 2026-03-22)
 - **v3.0 Competitive Parity & Benchmarks** — Phases 51-53 + Phase 51.5 (shipped 2026-05-14; Phase 53 merged as #30)
 - ✅ **v3.1 Make It True** — Phases 54-58 (shipped 2026-09-01 as v3.1.0)
-- **v3.2 Prove It** — Phases 59-62 (in progress; Phase 59 executing 2026-09-01)
+- **v3.2 Prove It** — Phases 59-62 (Phase 59 complete 2026-09-01; Phase 60 executing)
 
 ## Phases
 
@@ -230,7 +230,7 @@ Phases execute in numeric order: 51 -> 51.5 (merged out-of-band) -> 52 -> 53
 | v2.7 Multi-Runtime Portability | 45-50 | 11/11 | Complete | 2026-03-22 |
 | v3.0 Competitive Parity | 51-53 + 51.5, 53.5 | 10/10 | Complete | 2026-05-14 |
 | v3.1 Make It True | 54-58 | 14/14 | Complete | 2026-09-01 |
-| v3.2 Prove It | 59-62 | 3/13 | In progress | Phase 59 executing |
+| v3.2 Prove It | 59-62 | 3/13 | In progress | Phase 59 complete; 60-01 executing |
 
 ---
 
@@ -315,29 +315,29 @@ v3.1 made the claims true. v3.2 makes them provable: a real benchmark number,
 evidence behind every "Solid", and a daemon someone can run for a week.
 Nothing on this list is a new capability except Phase 62, which is conditional.
 
-### Phase 59: Guardrails and Inventory (3/3 plans) — IN EXECUTION 2026-09-01
+### Phase 59: Guardrails and Inventory (3/3 plans) — COMPLETE 2026-09-01 (#45)
 
 - [x] 59-01: Release pipeline checks (ancestor of main, crate version, all five platforms, CHANGELOG notes)
-- [x] 59-02: Orphan branch triage (`docs/plans/phase-59-orphan-branch-triage.md`)
-- [x] 59-03: Planning truth (PROJECT.md, ROADMAP, STATE, GitHub issues #39–#44)
+- [x] 59-02: Orphan branch triage (`docs/plans/phase-59-orphan-branch-triage.md`); OpenCode branch deleted
+- [x] 59-03: Planning truth (PROJECT.md, ROADMAP, STATE, 8 GitHub issues labelled `v3.2`)
 
-### Phase 60: Real Numbers (0/3)
+### Phase 60: Real Numbers (0/3) — 60-01 executing
 
-- [ ] 60-01: Live-backend isolation for `memory-bench locomo --backend cli`
+- [ ] 60-01: Live-backend isolation for `memory-bench locomo --backend cli` (this PR)
 - [ ] 60-02: The run — maintainer, needs API key + documented machine (#39)
-- [ ] 60-03: Vector and topic quality fixtures (#40)
+- [ ] 60-03: Vector (#40) and topic (#47) quality fixtures
 
 ### Phase 61: Operate It (0/5)
 
 - [ ] 61-01: Backfill (`admin backfill-index`) (#41); cherry-pick export/import from 59-02
 - [ ] 61-02: `install-service` launchd/systemd (#42)
 - [ ] 61-03: Panic audit (`unwrap()`/`expect()` on request paths)
 - [ ] 61-04: Offline TOC rebuild (#43)
-- [ ] 61-05: Installer register / uninstall / status (CREG/META from 59-02)
+- [ ] 61-05: Installer register / uninstall / status (CREG/META from 59-02; #48)
 
 ### Phase 62: Cross-encoder rerank (conditional) (#44)
 
 - [ ] Only if 60-02 shows retrieval, not generation, is the bottleneck
 
-*Updated: 2026-09-01 — v3.1.0 released; v3.2 Prove It adopted; Phase 59 executing*
+*Updated: 2026-09-02 — Phase 59 complete (#45); expanded v3.2 plan adopted; 60-01 executing*
 
diff --git a/.planning/STATE.md b/.planning/STATE.md
@@ -3,11 +3,11 @@ gsd_state_version: 1.0
 milestone_name: Prove It
 status: executing
 stopped_at: null
-last_updated: "2026-09-01T23:30:00.000Z"
-last_activity: 2026-09-01 — v3.1.0 released; v3.2 adopted; Phase 59 Guardrails and Inventory executing
+last_updated: "2026-09-02T01:40:00.000Z"
+last_activity: 2026-09-02 — Phase 60-01 live-backend isolation verified (spawn-per-conversation, checkpoint drain, CLI smoke)
 progress:
   total_phases: 4
-  completed_phases: 0
+  completed_phases: 1
   total_plans: 13
   completed_plans: 3
   percent: 23
@@ -20,39 +20,42 @@ progress:
 See: .planning/PROJECT.md (updated 2026-09-01)
 
 **Core value:** Agent can answer "what were we talking about last week?" without scanning everything
-**Current focus:** v3.2 Phase 59 — Guardrails and Inventory. Release pipeline cannot repeat the stale-tag incident; March gsd/ line inventoried; planning docs match v3.1.0 shipped.
+**Current focus:** v3.2 Phase 60-01 live-backend isolation ready to merge. Phase 59 Guardrails and Inventory is on `main` (#45).
 
 ## Current Position
 
-Phase: 59 of 62 (Guardrails and Inventory)
-Status: v3.1.0 shipped 2026-09-01 (5 of 5 platforms). v3.2 Prove It adopted. Phase 59 in execution.
-Last activity: 2026-09-01 — issues #39–#44 opened; release guards + orphan triage + PROJECT.md rewrite
+Phase: 60 of 62 (Real Numbers) — plan 60-01 verified locally
+Status: v3.1.0 shipped 2026-09-01 (5 of 5 platforms). v3.2 Prove It adopted (expanded spec). Phase 59 complete.
+Last activity: 2026-09-02 — 60-01 spawn-per-conversation + GetIndexCheckpoints drain + live CLI smoke + isolation bleed test
 
-Progress: [██░░░░░░░░] 3/13 plans (Phase 59). Phases 60–62 not started.
+Progress: [███░░░░░░░] 3/13 plans (Phase 59 complete). Phase 60-01 verified, awaiting merge.
 
 ## Out-of-band Work
 
 ### Open PRs
 
 | PR | What | Status |
 |---|---|---|
-| _(this branch)_ | Phase 59 Guardrails and Inventory | Open |
+| _(this branch)_ | Phase 60-01 live-backend isolation + remaining 59 gaps | Open |
 
 ### Open issues (the v3.2 backlog)
 
 | Issue | What | Phase |
 |---|---|---|
 | #39 | Real LOCOMO LLM-judge run | 60-02 |
-| #40 | Vector and topic-graph quality fixtures | 60-03 |
+| #40 | Vector quality fixtures | 60-03 / QUAL-01 |
+| #47 | Topic-graph quality (purity + ARI) | 60-03 / QUAL-02 |
 | #41 | Backfill BM25/vector for pre-v3.1 events | 61-01 |
 | #42 | `install-service` (launchd/systemd) | 61-02 |
 | #43 | Offline TOC rebuild | 61-04 |
+| #48 | Installer uninstall + status | 61-05 |
 | #44 | Cross-encoder rerank (conditional) | 62 |
 
 ### Recently Merged
 
 | PR | What | Merged |
 |---|---|---|
+| #45 | Phase 59 Guardrails and Inventory | 2026-09-01 |
 | #38 | docs: correct the "no tags" claim and record the release blocker | 2026-08-31 |
 | #37 | chore(v3.1): release prep — version 3.1.0, changelog, working release archives | 2026-08-31 |
 | #36 | Phase 57 Shop Window & Positioning | 2026-08-31 |
@@ -65,7 +68,8 @@ Progress: [██░░░░░░░░] 3/13 plans (Phase 59). Phases 60–62
 ## Decisions
 
 - v3.2 scope: Prove It — no new capabilities except conditional Phase 62
-- Maintainer decisions 2026-09-01: cherry-pick March export/import + CREG/META by feature (not by branch); skip OpenCode converter; blog now / product posts after #39; daemonization is unit files not double-fork
+- Maintainer decisions 2026-09-01 (all four accepted): cherry-pick March export/import + CREG/META by feature; skip OpenCode (branch deleted); blog now / product posts after #39; daemonization is unit files not double-fork; backfill is stopped-daemon CLI only
+- Canonical spec: `docs/plans/v3.2-prove-it-plan.md` (expanded GSD form)
 - v3.1.0 first tag push shipped `acc7294` (Cargo.toml 2.7.0) for 17 minutes — Phase 59-01 exists because of that
 - March `gsd/phase-56-import-bootstrap` is 88 ahead / 20 behind; naïve merge regresses orchestrator and bench. Inventory: `docs/plans/phase-59-orphan-branch-triage.md`
 - HOLD comparison marketing until #39 lands a `locomo_llm_judge` artifact