Lucent-Financial-Group · AceHack · May 17, 2026 · May 17, 2026 · May 17, 2026
diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md
@@ -310,7 +310,7 @@ are closed (status: closed in frontmatter)._
 - [x] **[B-0505](backlog/P1/B-0505-b0442-slice5c-docs-autonomous-loop-acceptance-close-2026-05-14.md)** B-0442 slice 5c — docs update (AUTONOMOUS-LOOP.md + bg/README.md) + B-0442 acceptance close
 - [x] **[B-0507](backlog/P1/B-0507-b0448-slice1-cloud-routines-api-research-2026-05-14.md)** B-0448 slice 1 — Research Cloud Routines auth + registration API surface (resolve unknowns)
 - [x] **[B-0508](backlog/P1/B-0508-b0448-slice2-cloud-schedule-json-schema-2026-05-14.md)** B-0448 slice 2 — Define cloud-schedule.json schema for tools/routines/<id>/
-- [ ] **[B-0509](backlog/P1/B-0509-b0448-slice3-install-ts-cloud-schedule-extension-2026-05-14.md)** B-0448 slice 3 — Extend tools/routines/install.ts to detect + surface cloud-schedule.json
+- [x] **[B-0509](backlog/P1/B-0509-b0448-slice3-install-ts-cloud-schedule-extension-2026-05-14.md)** B-0448 slice 3 — Extend tools/routines/install.ts to detect + surface cloud-schedule.json
 - [x] **[B-0510](backlog/P1/B-0510-b0448-slice4-autonomous-loop-cloud-schedule-json-2026-05-14.md)** B-0448 slice 4 — Author autonomous-loop/cloud-schedule.json (first Cloud Routine declaration)
 - [ ] **[B-0511](backlog/P1/B-0511-b0448-slice5-register-cloud-routine-empirical-fire-2026-05-14.md)** B-0448 slice 5 — Register autonomous-loop as Cloud Routine + empirical first-fire observation
 - [ ] **[B-0512](backlog/P1/B-0512-b0448-slice6-readme-4-layer-table-2026-05-14.md)** B-0448 slice 6 — Update tools/routines/README.md with 4-layer catch-43 table
@@ -600,6 +600,7 @@ are closed (status: closed in frontmatter)._
 - [ ] **[B-0580](backlog/P2/B-0580-enterprise-ruleset-management-2026-05-16.md)** Enterprise GitHub ruleset management — new layer above org/individual mapping (composes with prior ruleset-divergence smell decomposition)
 - [ ] **[B-0583](backlog/P2/B-0583-cross-machine-account-scoped-scarcity-bus-2026-05-16.md)** Cross-machine account-scoped scarcity bus — refine B-0570 from machine-local per-agent files to account-scoped timestamped surface
 - [ ] **[B-0610](backlog/P2/B-0610-amazon-orders-extract-v3-design-pass-2026-05-16.md)** Amazon orders extract — v3 design pass (8 deferred reviewer-thread findings)
+- [ ] **[B-0611](backlog/P2/B-0611-dangling-memory-refs-cleanup-35-refs-6-surfaces-2026-05-17.md)** Dangling memory-refs cleanup — 35 refs across 6 substrate surfaces (use PR #4042 audit tool)
 
 ## P3 — convenience / deferred
 

diff --git a/...backlog/P2/B-0611-dangling-memory-refs-cleanup-35-refs-6-surfaces-2026-05-17.md b/...backlog/P2/B-0611-dangling-memory-refs-cleanup-35-refs-6-surfaces-2026-05-17.md
@@ -0,0 +1,162 @@
+---
+id: B-0611
+title: Dangling memory-refs cleanup — 35 refs across 6 substrate surfaces (use PR #4042 audit tool)
+status: open
+priority: P2
+created: 2026-05-17
+last_updated: 2026-05-17
+type: chore
+composes_with:
+  - B-0510  # prior B-NNNN substrate work via Otto-CLI; uses the same audit tool surface
+depends_on: []
+---
+
+# Dangling memory-refs cleanup — 35 refs across 6 substrate surfaces
+
+## Substrate-honest context
+
+PR #4031 (closed audit memo, 2026-05-15) surfaced 5 dangling
+`memory/feedback_*.md` refs in `.claude/rules/`. PR #4041
+([merged 2026-05-17](https://github.com/Lucent-Financial-Group/Zeta/pull/4041))
+extended the audit to find 29 across 4 substrate surfaces
+(`.claude/rules` + `.claude/skills` + `docs/research` + `docs/backlog`).
+PR #4042 ([merged 2026-05-17](https://github.com/Lucent-Financial-Group/Zeta/pull/4042))
+mechanized the methodology as
+[`tools/hygiene/audit-dangling-memory-refs.ts`](../../../tools/hygiene/audit-dangling-memory-refs.ts).
+
+This row captures the **cleanup work itself** — the audit memos and
+the audit tool exist, but the dangling refs have not been resolved.
+
+Running the audit tool 2026-05-17T06:23Z (this row's filing tick)
+surfaces **35 dangling refs across 6 surfaces**:
+
+| Surface | scanned | uniqueRefs | uniqueDanglingRefs |
+|---|---:|---:|---:|
+| `.claude/agents` | 19 | 0 | 0 |
+| `.claude/skills` | 262 | 14 | 1 |
+| `.claude/rules` | 56 | 74 | 5 |
+| `docs/research` | 355 | 186 | 8 |
+| `docs/backlog` | 674 | 205 | 17 |
+| `memory/persona` | 288 | 60 | 4 |
+| **Total** | **1654** | **539** | **35** |
+
+The drift has grown (29 → 35) between PR #4041 audit and today's
+run — new substrate authoring continues to add citations to
+user-scope memory paths that don't exist in-repo.
+
+## Why this matters
+
+The maintainer's Otto-CLI auto-loads user-scope memory at
+`~/.claude/projects/.../memory/`, so citations like
+`memory/feedback_<name>.md` resolve transparently for the maintainer.
+A cold-boot agent on a fresh checkout (different machine, new
+contributor, CI agent, fresh-clone-CI run) does NOT have user-scope
+memory and follows the path to find nothing. The link is dead.
+
+Per [`.claude/rules/substrate-or-it-didnt-happen.md`](../../../.claude/rules/substrate-or-it-didnt-happen.md):
+"chat is weather, substrate is durable." If a substrate file
+points to a memory file that exists only in user-scope, the rule
+is partially weather for cold-boot consumers.
+
+## Scope
+
+For each of the 35 dangling refs, pick one of three resolutions:
+
+1. **In-repo projection (preferred for load-bearing claims)**:
+   create an in-repo summary of the user-scope content (or move the
+   file in-repo if the maintainer's PII discipline allows) and update
+   the citation to point at the in-repo file. Follows the pattern
+   already in use by some rules (e.g.,
+   `.claude/rules/codeql-no-source-on-docs-only-pr-is-broken-commit-canary.md`
+   which has a "cold-boot agents on fresh checkouts: this rule's own
+   body above is the canonical in-repo projection" pattern).
+
+2. **Footnote-the-user-scope-with-fallback**: keep the citation but
+   add a one-line note that the file is user-scope only AND name the
+   in-repo fallback substrate (e.g., `memory/CURRENT-otto.md` or the
+   rule body itself). Follows the pattern in
+   `.claude/rules/m-acc-multi-oracle-end-user-moral-invariants.md`
+   which explicitly footnotes user-scope-only references.
+
+3. **Remove the citation**: if the cited memory was an artifact of
+   an earlier iteration that no longer load-bears the surrounding
+   substrate, just delete the citation. Less common; substrate-
+   honest only when the citation is truly orphaned.
+
+## Acceptance
+
+- [ ] All 35 dangling refs resolved via one of the 3 patterns above
+- [ ] `bun tools/hygiene/audit-dangling-memory-refs.ts` exits 0
+  (no dangling refs on any of the 6 default surfaces)
+- [ ] Audit tool wired into CI (separate row OR same; see
+  Workflow notes)
+- [ ] Composing rule update (if needed): clarify the
+  in-repo-projection pattern in
+  `.claude/rules/substrate-or-it-didnt-happen.md` or sibling
+  rule so future authors avoid the pattern proactively
+- [ ] **Audit tool semi-automation** (verified 2026-05-17T06:44Z:
+  tool has only `--json` and `--surfaces` flags; no `--fix`,
+  `--allowlist`, or `--suggest` logic in source): consider adding
+  one of: (a) `--allowlist <path>` flag reading a `.audit-allowlist`
+  file with `<surface>:<line>:<ref>` entries marking known-
+  intentional dangling refs (composes with Option A from the
+  slice-1 recipe memo at
+  `memory/feedback_otto_cli_b0611_slice1_audit_recipe_*_2026_05_17.md`);
+  (b) `--suggest` mode that prints the 4-option resolution menu
+  per dangling ref; OR (c) inline annotation parser (e.g., HTML
+  comment `<!-- audit-allowlist: user-scope-intentional -->`
+  adjacent to the citation, opt-in per ref). Decision composes
+  with the design choice from the slice-1 recipe memo.
+
+## Workflow notes
+
+The cleanup is bounded but tedious — 35 individual decisions, each
+small. Suggested decomposition:
+
+- **Slice 1 (P2)**: `.claude/skills` (1 ref) + `.claude/rules`
+  (5 refs) — 6 refs, smallest scope, highest-leverage (rules
+  auto-load at session start so cleanup here ripples furthest)
+- **Slice 2 (P2)**: `memory/persona` (4 refs) — bounded, persona-
+  scoped
+- **Slice 3 (P2)**: `docs/research` (8 refs) — verbatim-preservation
+  files; resolutions likely lean toward "keep citation + footnote"
+- **Slice 4 (P3)**: `docs/backlog` (17 refs) — largest scope, lowest
+  urgency (backlog rows are work-in-progress citations, churn fast)
+
+Each slice can be a separate small PR or grouped commits within
+one PR. The audit tool's `--surfaces` flag scopes the verification.
+
+The CI integration is a separate concern: the audit tool exits 1
+on dangling refs, so wiring it into CI would block authoring until
+33 of 35 are resolved (the tool reports 35 today; new authoring will
+add more). Best practice: wire to CI with an allowlist that
+ratchets down OR run as advisory-only initially.
+
+## Composes with
+
+- PR #4042 — the audit tool (`tools/hygiene/audit-dangling-memory-refs.ts`)
+- PR #4041 — extended audit memo (29 refs)
+- PR #4031 — original audit memo (5 refs in rules)
+- [`.claude/rules/substrate-or-it-didnt-happen.md`](../../../.claude/rules/substrate-or-it-didnt-happen.md)
+  — the rule defining substrate vs weather
+- [`.claude/rules/wake-time-substrate.md`](../../../.claude/rules/wake-time-substrate.md)
+  — auto-loaded rules MUST be discoverable from cold-boot; dangling
+  refs in auto-loaded rules degrade cold-boot discoverability
+- [`.claude/rules/honor-those-that-came-before.md`](../../../.claude/rules/honor-those-that-came-before.md)
+  — preserves memory; cleanup must NOT delete user-scope content
+  in the maintainer's home directory; only resolve citations from
+  in-repo surfaces
+
+## Substrate-honest framing
+
+The dangling refs are not "broken" in the maintainer's
+day-to-day operational sense — his Otto-CLI auto-loads the files
+and follows the citations correctly. The drift is between the
+maintainer's local substrate (rich) and the in-repo substrate
+(thinner). Cleanup elevates the in-repo substrate to match what
+the maintainer sees, OR documents the asymmetry explicitly via
+the footnote pattern.
+
+This is NOT a P0/P1 because no shipping pipeline is blocked. It
+is P2 because cold-boot agents (the framework's design audience)
+hit dead links and lose substrate context.