diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index ad4aa2ba01..5e76a606ee 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -188,6 +188,14 @@ jobs:
         if: needs.changes.outputs.static_heavy_changed == 'true'
         run: npm run check:upload-limit-parity
 
+      # The generated lexicon review is a clinical-facing artefact. It was in the
+      # local verify:pr-local chain but in no CI job, so it rotted on main and
+      # failed every local preflight while every required check stayed green
+      # (#333). Same static-heavy condition verify-pr-local.mjs already uses.
+      - name: Medication lexicon report freshness
+        if: needs.changes.outputs.static_heavy_changed == 'true'
+        run: npm run check:medication-lexicon-report
+
       - name: GitHub Actions pin check
         if: needs.changes.outputs.workflow_changed == 'true'
         run: npm run check:github-actions
diff --git a/docs/branch-review-records/d50a877214261c376d150342864090de1d5f90e53c3187b03f4165df702e9b5d.record.md b/docs/branch-review-records/d50a877214261c376d150342864090de1d5f90e53c3187b03f4165df702e9b5d.record.md
new file mode 100644
index 0000000000..a4f7f63491
--- /dev/null
+++ b/docs/branch-review-records/d50a877214261c376d150342864090de1d5f90e53c3187b03f4165df702e9b5d.record.md
@@ -0,0 +1 @@
+| 2026-08-15 | claude/ledger-guard-ci-followups | 8fe2a50b9bce29e24710fdf72ce8b5f187c0569e | Lexicon-report CI freshness and database remediation sequencing | Fixed P1 prerequisite order and P2 generated-report scope gap; merged current base with no conflicts | git diff --check; ci-change-scope direct-report classification and self-test; GitHub Actions pin check; remediation-order/workflow-contract assertions; ledger-inbox; outstanding-issues; branch-review-ledger; ledger-write-discipline; focused Vitest unavailable (node_modules absent) |
diff --git a/docs/database-remediation-plan.md b/docs/database-remediation-plan.md
index 9bb9a18561..277b07e09c 100644
--- a/docs/database-remediation-plan.md
+++ b/docs/database-remediation-plan.md
@@ -3,14 +3,36 @@
 Owner: operator (Josh) + specialist session. Source findings: open ledger `#248` (whose causal
 conclusion awaits a read-only history and audit check) and the queued P1 live-drift follow-up.
 Companion evidence: live-drift Actions runs
-`30763871562` (2026-08-02) and
-`31330856982` (2026-08-09), PR #1614, `supabase/migrations/20260804110240_restore_rag_search_health_indexes.sql`.
-
-**Scope.** Fixes, in dependency order: the live-vs-repo schema gap (21 missing indexes, 2
-unexpected indexes, 10 diverged `match_*` RPC bodies), the unresolved cause of the affected
-migration history, drift-detection routing, and the surrounding database debt (`#102`, `#011`, `#036`,
-`#022`, `#025`, `#056`/`#057`, `#183`, `#188`/`#196`–`#200`, `#191`, `#098`/`#099`). Ends with
-standing protections so unverified history repairs remain visible and cannot silently mask drift.
+`30763871562` (2026-08-02), `31330856982` (2026-08-09) and
+`31813064485` (2026-08-14, the current measurement), PR #1614,
+`supabase/migrations/20260804110240_restore_rag_search_health_indexes.sql`.
+
+> **Status as of 2026-08-14.** Phase 0 is **complete** (routing, post-migration trigger, evidence
+> file, and the forced-dispatch proof — which auto-created issue **#1963**). Phase 1 is **partial**:
+> 1.1 and 1.3 were run in an owner-authorised incident window, but **1.2, the RPC divergence
+> dossier, is outstanding and is the gate for Phase 3**. Phase 4 is **partial**: the two
+> retrieval-critical trigram indexes were restored, ~20 remain.
+>
+> **Next step is Phase 1.2, not Phase 3.** Complete and review the RPC divergence dossier first,
+> then run the Phase 2 staging-parity rehearsal. Only after both prerequisites are complete may an
+> appropriately approved Phase 3 reconciliation proceed. Before starting any phase, check the
+> open-PR list for the surface first (`#292`): Phase 0 was independently built twice on 2026-08-14,
+> and Phases 3 and 4 spend approved production windows and eval-canary budget rather than just tokens.
+
+**Scope.** Fixes, in dependency order: the live-vs-repo schema gap, the unresolved cause of the
+affected migration history, drift-detection routing, and the surrounding database debt (`#102`,
+`#011`, `#036`, `#022`, `#025`, `#056`/`#057`, `#183`, `#188`/`#196`–`#200`, `#191`, `#098`/`#099`).
+Ends with standing protections so unverified history repairs remain visible and cannot silently mask
+drift.
+
+The schema gap, measured 2026-08-14 (superseding the 2026-08-09 figures this plan was written
+against):
+
+| Category                      | 2026-08-09 | 2026-08-14         |
+| ----------------------------- | ---------- | ------------------ |
+| diverged `match_*` RPC bodies | 10         | **10 — unchanged** |
+| `missing_live` indexes        | 21         | **20**             |
+| `unexpected_live` indexes     | 2          | **2 — unchanged**  |
 
 **Standing rules for every phase.** No hosted mutation without explicit approval for that phase.
 Never raw-SQL a drift fix — every live change is codified (migration + `schema.sql` mirror +
@@ -63,7 +85,7 @@ hypothesis into a named, evidence-backed conclusion.
 noise** (whitespace/qualifier only). No edits. This is the highest-stakes unknown in the whole
 plan — live retrieval behaviour may currently depend on bodies the repo does not contain.
 
-1.3 **Index inventory & sizing.** For the 21 missing and 2 unexpected indexes: owning-table
+1.3 **Index inventory & sizing.** For the 20 missing and 2 unexpected indexes: owning-table
 `pg_relation_size`, and `EXPLAIN (ANALYZE, BUFFERS)` for the known hot queries
 (`documents` title ILIKE, `document_chunks` content search, `rag_retrieval_logs` miss scan).
 These are the before-measurements for Phase 4, including the `#231` latency hypothesis.
diff --git a/docs/database-remediation-playbook.md b/docs/database-remediation-playbook.md
index d9d089aac1..0c8deceb92 100644
--- a/docs/database-remediation-playbook.md
+++ b/docs/database-remediation-playbook.md
@@ -19,19 +19,32 @@ migrations show that history advanced, but neither point distinguishes skipped D
 history from indexes that were created and later dropped. The Phase 1 read-only history and audit
 check must establish that cause before the row is closed or remediation is attributed to it.
 
-**Current live state** (from scheduled `live-drift.yml` Actions run `31330856982`, 2026-08-09):
+**Current live state** (from `live-drift.yml` Actions run `31813064485`, 2026-08-14 — this
+supersedes the 2026-08-09 run `31330856982` the plan was originally written against):
 
-- 21 `missing_live` indexes across many migrations — tables: `audit_logs` (2), `api_rate_limits`,
-  `document_chunks` (2, incl. `document_chunks_content_trgm_idx`), `document_images` (3),
-  `document_index_quality`, `document_index_units`, `document_publication_approvals`,
-  `document_summaries`, `documents` (2, incl. `documents_title_trgm_idx`),
+- 20 `missing_live` indexes across many migrations — tables: `audit_logs` (2), `api_rate_limits`,
+  `document_chunks`, `document_images` (3), `document_index_quality`, `document_index_units`,
+  `document_publication_approvals`, `document_summaries`, `documents`,
   `image_caption_cache`, `indexing_v3_agent_jobs`, `ingestion_job_stages`, `medication_records`,
   `rag_aliases`, `rag_queries`, `rag_query_misses`, `storage_cleanup_jobs`.
+  **`documents_title_trgm_idx` and `document_chunks_content_trgm_idx` are no longer among them** —
+  both were restored in the 2026-08-14 incident window and re-verified `indisvalid`/`indisready`.
 - 2 `unexpected_live` indexes: `document_table_facts_document_id_idx`,
   `storage_cleanup_jobs_owner_id_idx`.
-- `def_hash` mismatches on 10 `match_*` retrieval RPCs (protected RAG surface; live bodies vs
-  repo — direction unknown until diffed).
-- The weekly `live-drift` run has been red since 2026-07-26 with no notification routing.
+- `def_hash` mismatches on 10 `match_*` retrieval RPCs — **unchanged, and entirely outstanding**
+  (protected RAG surface; live bodies vs repo — direction unknown until diffed). This is now the
+  highest-stakes remaining unknown and the reason Phase 1.2 comes next.
+- Drift-failure routing is **live**: a failed run creates or updates the pinned issue
+  "Live drift check failing" (currently **#1963**) and a green run closes it.
+
+**Phase status.** Phase 0 complete. Phase 1 partial — 1.1 and 1.3 done in the owner-authorised
+incident window, **1.2 (RPC dossier) outstanding and gating Phase 3**. Phase 4 partial — two
+indexes restored, ~20 remain. **Next: complete Phase 1.2, then the Phase 2 staging-parity
+rehearsal; only then may an appropriately approved Phase 3 proceed.**
+
+**Before starting any phase, check the open-PR list for the surface** (`#292`). Phase 0 was built
+twice independently on 2026-08-14; a duplicate in Phase 3 or 4 wastes an approved production window
+and eval-canary budget, not just tokens.
 
 **Prior repair to imitate.** PR #1614 / migration `20260804110240_restore_rag_search_health_indexes.sql`
 is the approved pattern: operator prebuilds indexes with `CREATE INDEX CONCURRENTLY` outside any
@@ -70,7 +83,10 @@ mistakes. Opus is sufficient for Phases 0, 2, 4, 5, 7 — execution against this
 
 ---
 
-## Phase 0 — Enablement (repo-only; no approval window needed) · Opus · 2–4 h
+## Phase 0 — Enablement (repo-only; no approval window needed) · Opus · 2–4 h — **COMPLETE 2026-08-14**
+
+Delivered in PRs #1938, #1939 and #1951; the forced-dispatch proof is Actions run `31813064485`,
+which auto-created issue #1963. The prompt below is retained as history.
 
 Deliverables: drift-failure routing, post-migration drift trigger, evidence file scaffold.
 Definition of done: PR merged; `check:github-actions` and `verify:pr-local` green; a forced
@@ -119,7 +135,7 @@ conclusion for `#248`; no writes performed.
 > classify each as live-ahead, repo-ahead, or normalization noise, quoting the decisive diff hunks
 > — this is a protected RAG surface, so classification accuracy matters more than speed, and any
 > ambiguous diff is recorded as UNCLASSIFIED with the ambiguity explained, never guessed; (3) for
-> the 21 missing and 2 unexpected indexes, record owning-table pg_relation_size and run EXPLAIN
+> the 20 missing and 2 unexpected indexes, record owning-table pg_relation_size and run EXPLAIN
 > (ANALYZE, BUFFERS) for the documents title ILIKE query, the document_chunks content search, and
 > the rag_retrieval_logs miss scan as before-baselines. Write all evidence with dates and run IDs
 > into docs/audit/live-drift-forensics-2026-08.md, update the live-drift tracking item, commit, push, PR (docs-only;
@@ -169,7 +185,8 @@ allowlist entries); eval evidence attached for any behaviour-changing deploy.
 
 ## Phase 4 — Index restoration (approved off-peak production window) · Opus · 2–3 h active
 
-Prerequisites: Phases 1–3. Deliverables: 21 indexes restored + validated, 2 unexpected indexes
+Prerequisites: Phases 1–3. Deliverables: the ~20 still-missing indexes restored + validated (the two
+trigram indexes were already restored on 2026-08-14), 2 unexpected indexes
 dispositioned, guard migrations landed, live-drift green. Definition of done: green live-drift
 dispatch output pasted; `search_schema_health()` still `ok: true`.
 
diff --git a/scripts/ci-change-scope.mjs b/scripts/ci-change-scope.mjs
index 431749391f..03307a9e6d 100644
--- a/scripts/ci-change-scope.mjs
+++ b/scripts/ci-change-scope.mjs
@@ -121,6 +121,11 @@ const docPatterns = [
   /^LICENSE(?:\..*)?$/i,
 ];
 
+// This Markdown file is generated from the medication interaction lexicon. A
+// direct edit must run its freshness check; otherwise the ordinary docs-only
+// classification would let a stale clinical-facing report through CI.
+const generatedMedicationLexiconReport = "docs/medication-interaction-lexicon-review.md";
+
 const workflowPatterns = [
   ".github/workflows",
   ".github/actions",
@@ -371,6 +376,7 @@ function isExecutableWorkflowSurfacePath(filePath) {
  * YAML/policy under workflow surfaces stay light; executable files there do not.
  */
 function isRecognisedLightPath(filePath) {
+  if (filePath === generatedMedicationLexiconReport) return false;
   if (pathMatches(filePath, docPatterns)) return true;
   if (!pathMatches(filePath, workflowPatterns)) return false;
   return !isExecutableWorkflowSurfacePath(filePath);
@@ -413,6 +419,7 @@ function classify(files, { readLedger = readFlakeLedger } = {}) {
   const docsOnly =
     normalized.length > 0 &&
     normalized.every((file) => pathMatches(file, docPatterns)) &&
+    !normalized.includes(generatedMedicationLexiconReport) &&
     !sourceChanged &&
     !workflowChanged;
   const workflowOnly = workflowChanged && !staticHeavyChanged;
@@ -767,6 +774,12 @@ function selfTest() {
     build_changed: false,
     lockfile_changed: false,
   });
+  assertScope("generated-medication-lexicon-report-stays-heavy", [generatedMedicationLexiconReport], {
+    docs_only: false,
+    docs_changed: true,
+    static_heavy_changed: true,
+    coverage_changed: true,
+  });
   assertScope("tests-only", ["tests/rag-routing.test.ts"], {
     source_changed: true,
     coverage_changed: true,
diff --git a/tests/ci-cache-safety.test.ts b/tests/ci-cache-safety.test.ts
index ffd387b525..480df51044 100644
--- a/tests/ci-cache-safety.test.ts
+++ b/tests/ci-cache-safety.test.ts
@@ -99,6 +99,12 @@ describe("CI cache safety", () => {
     expect(workflow).toContain("run: npm run check:verification-plan");
   });
 
+  it("runs the generated medication lexicon freshness check through static-heavy scope", () => {
+    expect(workflow).toMatch(
+      /name: Medication lexicon report freshness\n\s+if: needs\.changes\.outputs\.static_heavy_changed == 'true'\n\s+run: npm run check:medication-lexicon-report/,
+    );
+  });
+
   it("does not repeat focused workflow contracts inside the full coverage run", () => {
     expect(workflow).toContain(
       "if: needs.changes.outputs.workflow_changed == 'true' && needs.changes.outputs.coverage_changed != 'true'",