From 2a8ec2ef9f0662b9d6386d8d42d68dd082516f2c Mon Sep 17 00:00:00 2001
From: qnbs <155236708+qnbs@users.noreply.github.com>
Date: Wed, 10 Jun 2026 08:30:36 +0200
Subject: [PATCH 1/3] =?UTF-8?q?feat(proforge):=20audit=20&=20perfection=20?=
 =?UTF-8?q?cycle=20=E2=80=94=20apply=20edits,=20persistence,=208-stage=20s?=
 =?UTF-8?q?upervisor,=20semantic=20RAG?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Full audit of the ProForge pipeline (flag `enableProForge`, default off) surfaced latent
correctness bugs, dead code, doc drift, and i18n/a11y gaps. This cycle fixes them and adds
the requested enhancements. Local gate green: lint + typecheck + i18n parity (11 locales);
417 ProForge + 169 LoRA unit tests pass.

Correctness:
- Fix hook getState self-recursion + stale state — read live store via new app/storeRef.ts
  (lightweight, no reducer-graph import); live config getter; per-project orchestrator rebuild.
- Reset AbortController in startPipeline so abort→restart works.
- SupervisorAgent now detects the real isFallback flag (was a stale qualityScore===50 sentinel).
- Extract headSnapshotId() helper (dedupes 4× inline lookup).

Apply accepted edits (core gap): new services/proForge/applyReviewEdits.ts writes accepted
ReviewItems back into the manuscript (offset-safe back-to-front, text-match fallback, stale-skip)
in submitReview before snapshotting, editing stages only; redux-undo makes them reversible.

Remove dead code: toolRegistry (zero consumers), StyleAdaptedWriterAgent (unwired),
PipelineConfig.loraAdapterId, and their tests.

Enhancements:
- Persist run history per project to IDB (proForgeHistoryStore, cap 20); hydrate on load.
- SupervisorAgent gates all 8 stages (was 3).
- Feedback-injected retries: BaseAgent.setRetryFeedback prepends supervisor reasons + reflection.
- Memory-bank semantic RAG: search honours ragMode (lexical/semantic/hybrid via MiniLM), keyword fallback.

i18n/a11y: PipelineReviewPanel fully t()-ized (~60 proforge.* keys, en+de translated, 10 locales
propagated) with aria-hidden emoji, sr-only severity, aria-pressed/expanded, useAnnounce on submit.

Docs: correct PROFORGE-PIPELINE.md (SupervisionDecision shape, supervisor rules, removed
tool-calling section) and CLAUDE.md; rebuild graphify.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 CLAUDE.md                                     |  11 +-
 app/store.ts                                  |   8 +-
 app/storeRef.ts                               |  12 +
 components/proForge/PipelineReviewPanel.tsx   | 139 ++++----
 docs/PROFORGE-PIPELINE.md                     |  90 +++--
 features/proForge/types.ts                    |   2 -
 graphify-out/GRAPH_REPORT.md                  | 324 ++++++++---------
 hooks/useProForgeOrchestrator.ts              |  54 ++-
 locales/ar/common.json                        |  55 +++
 locales/de/common.json                        |  55 +++
 locales/el/common.json                        |  55 +++
 locales/en/common.json                        |  55 +++
 locales/es/common.json                        |  55 +++
 locales/es/settings.json                      |  36 +-
 locales/fr/common.json                        |  55 +++
 locales/fr/settings.json                      |  36 +-
 locales/he/common.json                        |  55 +++
 locales/it/common.json                        |  55 +++
 locales/it/settings.json                      |  36 +-
 locales/ja/common.json                        |  55 +++
 locales/pt/common.json                        |  55 +++
 locales/zh/common.json                        |  55 +++
 public/locales/ar/bundle.json                 |  55 +++
 public/locales/de/bundle.json                 |  55 +++
 public/locales/el/bundle.json                 |  55 +++
 public/locales/en/bundle.json                 |  55 +++
 public/locales/es/bundle.json                 |  89 ++++-
 public/locales/fr/bundle.json                 |  89 ++++-
 public/locales/he/bundle.json                 |  55 +++
 public/locales/it/bundle.json                 |  89 ++++-
 public/locales/ja/bundle.json                 |  55 +++
 public/locales/pt/bundle.json                 |  55 +++
 public/locales/zh/bundle.json                 |  55 +++
 services/proForge/applyReviewEdits.ts         | 133 +++++++
 services/proForge/pipelineAgents/baseAgent.ts |  22 +-
 .../pipelineAgents/styleAdaptedWriterAgent.ts | 119 -------
 .../pipelineAgents/supervisorAgent.ts         | 132 ++++++-
 .../proForge/pipelineTools/toolRegistry.ts    | 303 ----------------
 services/proForge/proForgeHistoryStore.ts     |  65 ++++
 services/proForge/proForgeMemoryBank.ts       |  95 +++--
 services/proForge/proForgeOrchestrator.ts     |  88 ++++-
 .../unit/lora/styleAdaptedWriterAgent.test.ts | 173 ---------
 tests/unit/proForge/applyReviewEdits.test.ts  | 127 +++++++
 .../components/PipelineReviewPanel.test.tsx   |  27 +-
 .../proForge/pipelineAgents/baseAgent.test.ts |  18 +
 .../pipelineAgents/supervisorAgent.test.ts    | 139 +++++++-
 .../proForge/proForgeHistoryStore.test.ts     |  70 ++++
 .../unit/proForge/proForgeMemoryBank.test.ts  |  56 ++-
 .../proForge/proForgeOrchestrator.test.ts     | 176 +++++++++
 tests/unit/proForge/toolRegistry.test.ts      | 333 ------------------
 50 files changed, 2752 insertions(+), 1384 deletions(-)
 create mode 100644 app/storeRef.ts
 create mode 100644 services/proForge/applyReviewEdits.ts
 delete mode 100644 services/proForge/pipelineAgents/styleAdaptedWriterAgent.ts
 delete mode 100644 services/proForge/pipelineTools/toolRegistry.ts
 create mode 100644 services/proForge/proForgeHistoryStore.ts
 delete mode 100644 tests/unit/lora/styleAdaptedWriterAgent.test.ts
 create mode 100644 tests/unit/proForge/applyReviewEdits.test.ts
 create mode 100644 tests/unit/proForge/proForgeHistoryStore.test.ts
 delete mode 100644 tests/unit/proForge/toolRegistry.test.ts

diff --git a/CLAUDE.md b/CLAUDE.md
index 659c131ff..23a82dad3 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -91,8 +91,9 @@ services/         → External adapters; key sub-dirs:
                      duckdb/      (duckdbClient, duckdbSchema, duckdbAnalytics, duckdbMigration, ragVectorMigration)
                      help/        (helpCatalog, helpSearch, helpDocRetrieval)
                      keyboard/    (shortcut normalization, conflict detection)
-                     proForge/    (proForgeOrchestrator, proForgeMemoryBank, pipelineAgents/ — baseAgent,
-                                   supervisorAgent + 8 stage agents; pipelineOutput/, pipelinePrompts/, pipelineTools/)
+                     proForge/    (proForgeOrchestrator, proForgeMemoryBank, proForgeHistoryStore,
+                                   applyReviewEdits, pipelineAgents/ — baseAgent, supervisorAgent + 8
+                                   stage agents; pipelineOutput/)
                      storage/     (idbCore, idbProjectStore, idbSnapshotStore, idbKeyStore, idbCodexStore,
                                    idbAssetStore, storageEncryptionService — AES-256-GCM at-rest via B-1)
                      voice/       (voiceCommandService, voiceTypes, stt/tts/vad/wakeWord/intent engines,
@@ -300,11 +301,11 @@ All `.md` guides listed in **[`README.md`](README.md#-documentation-hub) § Docu
 
 ### ProForge Pipeline
 
-8-stage agentic editing pipeline. Flag: `enableProForge`. Full docs: `docs/PROFORGE-PIPELINE.md`. Stages: `intake` → `structural` → `lineProse` → `copyEdit` → `proof` → `production` → `publishing` → `analytics`. Pauses at `awaitingReview` — manuscript never auto-modified without user approval.
+8-stage agentic editing pipeline. Flag: `enableProForge`. Full docs: `docs/PROFORGE-PIPELINE.md`. Stages: `intake` → `structural` → `lineProse` → `copyEdit` → `proof` → `production` → `publishing` → `analytics`. Pauses at `awaitingReview` — manuscript never auto-modified without user approval. On review submit, accepted `ReviewItem`s for **editing stages** are applied back into the manuscript via `applyReviewEdits.ts` (offset-safe, back-to-front, stale-match skip) before the post-stage snapshot; redux-undo makes them reversible.
 
-**Redux slice** (`features/proForge/proForgeSlice.ts`, root key `proForge`, NOT undo-wrapped). **Types** in `features/proForge/types.ts`. **Orchestrator:** `services/proForge/proForgeOrchestrator.ts` — call via `hooks/useProForgeOrchestrator.ts`, never instantiate in components. Agents (all 8 extend `BaseAgent`) lazy-loaded per stage; call `this.buildAiOpts({ maxTokens: N })` — `AIRequestOptions` requires `model` + `provider`. `BaseAgent` provides `requireProject()`, `getMemoryBank()`, `selfReflect()`, `elapsed()`.
+**Redux slice** (`features/proForge/proForgeSlice.ts`, root key `proForge`, NOT undo-wrapped, **ephemeral**). Completed/aborted runs persist per project to IDB `proforge-run-history` (`proForgeHistoryStore.ts`, cap 20) and rehydrate via `useProForgeOrchestrator`. **Types** in `features/proForge/types.ts`. **Orchestrator:** `services/proForge/proForgeOrchestrator.ts` — call via `hooks/useProForgeOrchestrator.ts` (reads live state via `appStoreRef` from `app/storeRef.ts`), never instantiate in components. Agents (all 8 extend `BaseAgent`) lazy-loaded per stage; call `this.buildAiOpts({ maxTokens: N })` — `AIRequestOptions` requires `model` + `provider`. `BaseAgent` provides `requireProject()`, `getMemoryBank()`, `selfReflect()`, `elapsed()`, `setRetryFeedback()` (supervisor reasons injected into retry prompts).
 
-**SupervisorAgent**: heuristic gate, no AI calls. `evaluate(stage, result)` → `{verdict: 'pass'|'retry'|'fail'}`. Hard gate: intake `qualityScore < 30` → fail. All `createFallback*` use 0 scores + `isFallback: true` — never fake mid-range values. **Memory Bank:** IDB `proforge-memory-bank`. **View:** `ProForgeViewContext` + `useProForgeViewContext()`.
+**SupervisorAgent**: heuristic gate, no AI calls, gates **all 8 stages**. `evaluate(stage, result)` → `SupervisionDecision { pass, retryRecommended, qualityScore, reasons }`. Detects `isFallback: true` (intake/structural/proof). Hard gate: intake `qualityScore < 30` → fail. All `createFallback*` use 0 scores + `isFallback: true` — never fake mid-range values. **Memory Bank:** IDB `proforge-memory-bank`; `search(query, limit, mode)` honours `ragMode` (`lexical` | `semantic` | `hybrid` via MiniLM embeddings, keyword fallback). **View:** `ProForgeViewContext` + `useProForgeViewContext()`.
 
 ### Scene-level services
 
diff --git a/app/store.ts b/app/store.ts
index a17cb4bc8..9a72f9ba1 100644
--- a/app/store.ts
+++ b/app/store.ts
@@ -204,7 +204,7 @@ const _tempStore = configureStore({ reducer: rootReducer });
 export type RootState = ReturnType<typeof _tempStore.getState>;
 export type AppDispatch = typeof _tempStore.dispatch;
 
-// Global store reference for non-React consumers (e.g., voice service)
-export const appStoreRef = {
-  current: null as { getState(): RootState; dispatch: AppDispatch } | null,
-};
+// Global store reference for non-React consumers (e.g., voice service, ProForge orchestrator).
+// QNBS-v3: Defined in app/storeRef.ts so consumers can import the ref without pulling the whole
+// reducer graph; re-exported here for backward compatibility.
+export { appStoreRef } from './storeRef';
diff --git a/app/storeRef.ts b/app/storeRef.ts
new file mode 100644
index 000000000..3f39cc1ba
--- /dev/null
+++ b/app/storeRef.ts
@@ -0,0 +1,12 @@
+/**
+ * Global store reference for non-React consumers (e.g., voice service, ProForge orchestrator).
+ * QNBS-v3: Lives in its own module so importing the ref does NOT pull the entire store/reducer
+ * graph into the consumer — keeps hooks/services lightweight and unit-testable in isolation.
+ * Type-only imports from './store' are erased at runtime, so there is no runtime dependency cycle.
+ */
+
+import type { AppDispatch, RootState } from './store';
+
+export const appStoreRef = {
+  current: null as { getState(): RootState; dispatch: AppDispatch } | null,
+};
diff --git a/components/proForge/PipelineReviewPanel.tsx b/components/proForge/PipelineReviewPanel.tsx
index 238d90295..0318a53c7 100644
--- a/components/proForge/PipelineReviewPanel.tsx
+++ b/components/proForge/PipelineReviewPanel.tsx
@@ -6,9 +6,11 @@
 
 import type React from 'react';
 import { useCallback, useMemo, useState } from 'react';
+import { useAnnounce } from '../../contexts/LiveRegionContext';
 import { useProForgeViewContext } from '../../contexts/ProForgeViewContext';
 import { proForgeActions } from '../../features/proForge/proForgeSlice';
 import type { ReviewItem, ReviewItemStatus } from '../../features/proForge/types';
+import { useTranslation } from '../../hooks/useTranslation';
 
 const SEVERITY_ICONS: Record<string, string> = {
   critical: '🔴',
@@ -16,42 +18,30 @@ const SEVERITY_ICONS: Record<string, string> = {
   info: '🔵',
 };
 
-const TYPE_LABELS: Record<string, string> = {
-  structuralEdit: 'Structural',
-  proseEdit: 'Prose',
-  grammarEdit: 'Grammar',
-  styleEdit: 'Style',
-  repetitionHit: 'Repetition',
-  consistencyIssue: 'Consistency',
-  plotHole: 'Plot Hole',
-  legalWarning: 'Legal',
-  technicalIssue: 'Technical',
-  pacingIssue: 'Pacing',
-  arcIssue: 'Arc',
-};
-
 const SEVERITY_GROUPS = [
   {
     key: 'critical' as const,
-    label: 'Critical Actions',
+    labelKey: 'proforge.review.group.critical',
     headerClass: 'text-[var(--sc-error)] border-[var(--sc-error-muted)]',
     badgeClass: 'bg-[var(--sc-error-muted)] text-[var(--sc-error)]',
   },
   {
     key: 'warning' as const,
-    label: 'Warnings',
+    labelKey: 'proforge.review.group.warning',
     headerClass: 'text-[var(--sc-warning,#d97706)] border-[var(--sc-warning-muted,#fef3c7)]',
     badgeClass: 'bg-[var(--sc-warning-muted,#fef3c7)] text-[var(--sc-warning,#d97706)]',
   },
   {
     key: 'info' as const,
-    label: 'Suggestions',
+    labelKey: 'proforge.review.group.info',
     headerClass: 'text-[var(--sc-accent)] border-[var(--sc-accent-muted,#eff6ff)]',
     badgeClass: 'bg-[var(--sc-accent-muted,#eff6ff)] text-[var(--sc-accent)]',
   },
 ] as const;
 
 export const PipelineReviewPanel: React.FC = () => {
+  const { t } = useTranslation();
+  const announce = useAnnounce();
   const {
     currentRun,
     activeStageResult,
@@ -65,6 +55,8 @@ export const PipelineReviewPanel: React.FC = () => {
   const [filter, setFilter] = useState<'all' | 'pending' | 'accepted' | 'rejected'>('all');
 
   const stage = activeStageResult?.stage;
+  // QNBS-v3: Localised stage label; falls back to the raw stage id if no key exists.
+  const stageLabel = stage ? t(`proforge.stageName.${stage}`) : '';
 
   const pendingCount = currentStageReviewItems.filter((i) => i.status === 'pending').length;
   const acceptedCount = currentStageReviewItems.filter((i) => i.status === 'accepted').length;
@@ -138,8 +130,10 @@ export const PipelineReviewPanel: React.FC = () => {
     const items = activeStageResult?.reviewItems ?? [];
     const decisions = items.map((item) => ({ itemId: item.id, status: item.status }));
     void submitReview(stage, decisions);
+    // QNBS-v3: WCAG live-region announce so screen-reader users hear the stage was submitted.
+    announce(t('proforge.review.announceSubmitted', { stage: stageLabel }), 'polite');
     setActiveView('dashboard');
-  }, [stage, activeStageResult, submitReview, setActiveView]);
+  }, [stage, stageLabel, activeStageResult, submitReview, setActiveView, announce, t]);
 
   const handleSkip = useCallback(() => {
     if (!stage) return;
@@ -150,7 +144,7 @@ export const PipelineReviewPanel: React.FC = () => {
   if (!currentRun || !stage) {
     return (
       <div className="rounded-sc-lg bg-[var(--sc-surface-elevated)] border border-[var(--sc-border-subtle)] p-6 text-center">
-        <p className="text-sm text-[var(--sc-text-secondary)]">No review items available.</p>
+        <p className="text-sm text-[var(--sc-text-secondary)]">{t('proforge.review.noItems')}</p>
       </div>
     );
   }
@@ -160,9 +154,15 @@ export const PipelineReviewPanel: React.FC = () => {
       {/* Header */}
       <div className="flex items-center justify-between px-4 py-3 border-b border-[var(--sc-border-subtle)]">
         <div>
-          <h3 className="text-sm font-medium capitalize">Review: {stage}</h3>
+          <h3 className="text-sm font-medium">
+            {t('proforge.review.heading', { stage: stageLabel })}
+          </h3>
           <p className="text-xs text-[var(--sc-text-secondary)] mt-0.5">
-            {pendingCount} pending · {acceptedCount} accepted · {rejectedCount} rejected
+            {t('proforge.review.counts', {
+              pending: pendingCount,
+              accepted: acceptedCount,
+              rejected: rejectedCount,
+            })}
           </p>
         </div>
         <div className="flex items-center gap-2 flex-wrap justify-end">
@@ -171,9 +171,9 @@ export const PipelineReviewPanel: React.FC = () => {
               type="button"
               onClick={handleQuickAcceptHighConfidence}
               className="px-2.5 py-1 text-xs rounded-sc-md bg-[var(--sc-accent-muted,#eff6ff)] text-[var(--sc-accent)] hover:opacity-80 transition-opacity"
-              title="Accept all high-confidence non-critical suggestions"
+              title={t('proforge.review.quickAcceptTitle')}
             >
-              Quick Accept ({quickAcceptCount})
+              {t('proforge.review.quickAccept', { count: quickAcceptCount })}
             </button>
           )}
           <button
@@ -181,14 +181,14 @@ export const PipelineReviewPanel: React.FC = () => {
             onClick={handleAcceptAll}
             className="px-2.5 py-1 text-xs rounded-sc-md bg-[var(--sc-success-muted)] text-[var(--sc-success)] hover:opacity-80 transition-opacity"
           >
-            Accept All
+            {t('proforge.review.acceptAll')}
           </button>
           <button
             type="button"
             onClick={handleRejectAll}
             className="px-2.5 py-1 text-xs rounded-sc-md bg-[var(--sc-error-muted)] text-[var(--sc-error)] hover:opacity-80 transition-opacity"
           >
-            Reject All
+            {t('proforge.review.rejectAll')}
           </button>
         </div>
       </div>
@@ -198,15 +198,20 @@ export const PipelineReviewPanel: React.FC = () => {
         <div className="mx-4 mt-3 p-3 rounded-sc-md border border-[var(--sc-error-muted)] bg-[var(--sc-error-muted)]">
           <div className="flex items-center justify-between mb-2">
             <span className="text-xs font-semibold text-[var(--sc-error)]">
-              🔴 {criticalPending.length} Critical{' '}
-              {criticalPending.length === 1 ? 'Issue' : 'Issues'} Need Attention
+              <span aria-hidden="true">🔴 </span>
+              {t(
+                criticalPending.length === 1
+                  ? 'proforge.review.criticalHeadingOne'
+                  : 'proforge.review.criticalHeadingOther',
+                { count: criticalPending.length },
+              )}
             </span>
             <button
               type="button"
               onClick={handleAcceptAllCritical}
               className="px-2 py-0.5 text-xs rounded-sc-sm bg-[var(--sc-error)] text-white hover:opacity-80 transition-opacity"
             >
-              Accept All Critical
+              {t('proforge.review.acceptAllCritical')}
             </button>
           </div>
           {topCritical.map((item) => (
@@ -216,7 +221,7 @@ export const PipelineReviewPanel: React.FC = () => {
           ))}
           {criticalPending.length > 3 && (
             <div className="text-xs text-[var(--sc-error)] mt-1 opacity-70">
-              +{criticalPending.length - 3} more critical issues below
+              {t('proforge.review.moreCritical', { count: criticalPending.length - 3 })}
             </div>
           )}
         </div>
@@ -224,24 +229,27 @@ export const PipelineReviewPanel: React.FC = () => {
 
       {/* Filter Tabs — secondary position */}
       <div className="flex gap-1 px-4 pt-3">
-        {(['all', 'pending', 'accepted', 'rejected'] as const).map((f) => (
-          <button
-            type="button"
-            key={f}
-            onClick={() => setFilter(f)}
-            className={`px-2.5 py-1 text-xs rounded-sc-md capitalize transition-colors ${
-              filter === f
-                ? 'bg-[var(--sc-accent)] text-white'
-                : 'bg-[var(--sc-surface-base)] text-[var(--sc-text-secondary)] hover:text-[var(--sc-text-primary)]'
-            }`}
-          >
-            {f} (
-            {f === 'all'
+        {(['all', 'pending', 'accepted', 'rejected'] as const).map((f) => {
+          const count =
+            f === 'all'
               ? currentStageReviewItems.length
-              : currentStageReviewItems.filter((i) => i.status === f).length}
-            )
-          </button>
-        ))}
+              : currentStageReviewItems.filter((i) => i.status === f).length;
+          return (
+            <button
+              type="button"
+              key={f}
+              onClick={() => setFilter(f)}
+              aria-pressed={filter === f}
+              className={`px-2.5 py-1 text-xs rounded-sc-md transition-colors ${
+                filter === f
+                  ? 'bg-[var(--sc-accent)] text-white'
+                  : 'bg-[var(--sc-surface-base)] text-[var(--sc-text-secondary)] hover:text-[var(--sc-text-primary)]'
+              }`}
+            >
+              {t(`proforge.review.filter.${f}`)} ({count})
+            </button>
+          );
+        })}
       </div>
 
       {/* Review Items */}
@@ -254,7 +262,7 @@ export const PipelineReviewPanel: React.FC = () => {
             return (
               <div key={group.key}>
                 <div className={`flex items-center gap-2 mb-2 pb-1 border-b ${group.headerClass}`}>
-                  <span className="text-xs font-semibold">{group.label}</span>
+                  <span className="text-xs font-semibold">{t(group.labelKey)}</span>
                   <span className={`text-xs px-1.5 py-0.5 rounded-sc-sm ${group.badgeClass}`}>
                     {items.length}
                   </span>
@@ -269,7 +277,7 @@ export const PipelineReviewPanel: React.FC = () => {
           })
         ) : filteredItems.length === 0 ? (
           <p className="text-sm text-[var(--sc-text-secondary)] text-center py-8">
-            No items match this filter.
+            {t('proforge.review.noItemsForFilter')}
           </p>
         ) : (
           filteredItems.map((item) => (
@@ -285,7 +293,7 @@ export const PipelineReviewPanel: React.FC = () => {
           onClick={handleSkip}
           className="px-3 py-1.5 text-xs rounded-sc-md text-[var(--sc-text-secondary)] hover:text-[var(--sc-text-primary)] transition-colors"
         >
-          Skip Stage
+          {t('proforge.review.skipStage')}
         </button>
         <button
           type="button"
@@ -294,7 +302,9 @@ export const PipelineReviewPanel: React.FC = () => {
           className="px-4 py-1.5 text-xs font-medium rounded-sc-md text-white disabled:opacity-50 transition-opacity"
           style={{ backgroundColor: 'var(--sc-accent)' }}
         >
-          {pendingCount > 0 ? `${pendingCount} Pending` : 'Submit & Continue'}
+          {pendingCount > 0
+            ? t('proforge.review.pendingCount', { count: pendingCount })
+            : t('proforge.review.submitContinue')}
         </button>
       </div>
     </div>
@@ -310,12 +320,13 @@ function ReviewItemCard({
   item: ReviewItem;
   onStatusChange: (id: string, status: ReviewItemStatus) => void;
 }) {
+  const { t } = useTranslation();
   const [expanded, setExpanded] = useState(false);
 
-  const statusButtons: { status: ReviewItemStatus; label: string; color: string }[] = [
-    { status: 'accepted', label: 'Accept', color: 'var(--sc-success)' },
-    { status: 'rejected', label: 'Reject', color: 'var(--sc-error)' },
-    { status: 'ignored', label: 'Ignore', color: 'var(--sc-text-tertiary)' },
+  const statusButtons: { status: ReviewItemStatus; labelKey: string; color: string }[] = [
+    { status: 'accepted', labelKey: 'proforge.review.accept', color: 'var(--sc-success)' },
+    { status: 'rejected', labelKey: 'proforge.review.reject', color: 'var(--sc-error)' },
+    { status: 'ignored', labelKey: 'proforge.review.ignore', color: 'var(--sc-text-tertiary)' },
   ];
 
   return (
@@ -329,11 +340,15 @@ function ReviewItemCard({
       }`}
     >
       <div className="flex items-start gap-2">
-        <span className="text-sm mt-0.5">{SEVERITY_ICONS[item.severity]}</span>
+        {/* QNBS-v3: Emoji is decorative; severity is conveyed to AT via the sr-only text below. */}
+        <span className="text-sm mt-0.5" aria-hidden="true">
+          {SEVERITY_ICONS[item.severity]}
+        </span>
+        <span className="sr-only">{t(`proforge.review.severity.${item.severity}`)}</span>
         <div className="flex-1 min-w-0">
           <div className="flex items-center gap-2 flex-wrap">
             <span className="text-xs font-medium px-1.5 py-0.5 rounded-sc-sm bg-[var(--sc-surface-elevated)]">
-              {TYPE_LABELS[item.type] ?? item.type}
+              {t(`proforge.review.type.${item.type}`)}
             </span>
             {item.sectionTitle && (
               <span className="text-xs text-[var(--sc-text-secondary)] truncate">
@@ -341,7 +356,7 @@ function ReviewItemCard({
               </span>
             )}
             <span className="text-xs text-[var(--sc-text-tertiary)] ml-auto">
-              {Math.round(item.confidence * 100)}% confidence
+              {t('proforge.review.confidence', { percent: Math.round(item.confidence * 100) })}
             </span>
           </div>
           <p className="text-sm mt-1.5">{item.description}</p>
@@ -349,10 +364,10 @@ function ReviewItemCard({
           {item.original && item.proposed && expanded && (
             <div className="mt-2 space-y-1.5 text-xs">
               <div className="p-2 rounded-sc-sm bg-[var(--sc-error-muted)] text-[var(--sc-error)]">
-                <span className="font-medium">Original:</span> {item.original}
+                <span className="font-medium">{t('proforge.review.original')}</span> {item.original}
               </div>
               <div className="p-2 rounded-sc-sm bg-[var(--sc-success-muted)] text-[var(--sc-success)]">
-                <span className="font-medium">Proposed:</span> {item.proposed}
+                <span className="font-medium">{t('proforge.review.proposed')}</span> {item.proposed}
               </div>
             </div>
           )}
@@ -361,9 +376,10 @@ function ReviewItemCard({
             <button
               type="button"
               onClick={() => setExpanded(!expanded)}
+              aria-expanded={expanded}
               className="text-xs text-[var(--sc-accent)] mt-1 hover:underline"
             >
-              {expanded ? 'Show less' : 'Show details'}
+              {expanded ? t('proforge.review.showLess') : t('proforge.review.showDetails')}
             </button>
           )}
 
@@ -374,6 +390,7 @@ function ReviewItemCard({
                 type="button"
                 key={btn.status}
                 onClick={() => onStatusChange(item.id, btn.status)}
+                aria-pressed={item.status === btn.status}
                 className={`px-2 py-0.5 text-xs rounded-sc-sm border transition-all ${
                   item.status === btn.status
                     ? 'text-white'
@@ -385,7 +402,7 @@ function ReviewItemCard({
                     : {}
                 }
               >
-                {btn.label}
+                {t(btn.labelKey)}
               </button>
             ))}
           </div>
diff --git a/docs/PROFORGE-PIPELINE.md b/docs/PROFORGE-PIPELINE.md
index 224c0f683..66e1fd164 100644
--- a/docs/PROFORGE-PIPELINE.md
+++ b/docs/PROFORGE-PIPELINE.md
@@ -88,7 +88,9 @@ features/proForge/
 
 services/proForge/
   proForgeOrchestrator.ts     # Central orchestrator + executeStageWithSupervision loop
-  proForgeMemoryBank.ts       # IndexedDB-backed project memory
+  proForgeMemoryBank.ts       # IndexedDB-backed project memory (keyword / semantic / hybrid recall)
+  proForgeHistoryStore.ts     # IndexedDB run-history persistence (survives reload)
+  applyReviewEdits.ts         # Applies accepted ReviewItems back into manuscript text
   pipelineAgents/
     baseAgent.ts              # Abstract base class (~200 LOC saved across 8 agents)
     supervisorAgent.ts        # Heuristic quality gate — no AI calls
@@ -100,8 +102,6 @@ services/proForge/
     productionAgent.ts        # Stage 6
     publishingAgent.ts        # Stage 7
     analyticsAgent.ts         # Stage 8
-  pipelineTools/
-    toolRegistry.ts           # Tool-calling registry + built-in tools
   pipelineOutput/
     structuredOutput.ts       # Zod schemas for all AI outputs
 
@@ -159,17 +159,24 @@ export class SupervisorAgent {
 `SupervisionDecision` shape:
 ```typescript
 interface SupervisionDecision {
-  verdict: 'pass' | 'retry' | 'fail';
-  reason: string;
-  retryHint?: string;
+  pass: boolean;
+  retryRecommended: boolean;
+  qualityScore: number;   // 0–100
+  reasons: string[];
 }
 ```
 
-**Decision rules per stage:**
-- `intake`: Detects uniform 50/100 fallback sentinel (all scores = exactly 50) → retry. `qualityScore < 30` → fail (hard gate).
-- `structural`: If edit count > word count / 10 → retry (over-aggressive edits).
-- `proof`: If grammar issue count > word count / 20 → retry (implausible density).
-- All stages: `isFallback: true` on output → retry (up to `maxRetries` times).
+**Decision rules per stage** (all 8 stages are gated):
+- `intake`: `isFallback: true` → fail (`qualityScore` 0). `qualityScore < 30` → hard-fail the pipeline.
+- `structural`: `isFallback: true` → retry. Zero edits + zero review items on a >1000-word manuscript → retry.
+- `lineProse`: Zero prose edits + zero review items on a >1000-word manuscript → retry.
+- `copyEdit`: Zero grammar/style/repetition/format findings on a >1500-word manuscript → retry.
+- `proof`: `isFallback: true` → retry. `overallPass` with zero grammar issues on a >500-word manuscript → retry.
+- `production`: Zero export artifacts → retry.
+- `publishing`: Missing book-title metadata or back-cover blurb → retry.
+- `analytics`: Never blocks (terminal, informational); flags a missing metrics block in `reasons`.
+
+A non-passing decision below the retry budget is attached to the `StageResult` as `supervisorDecision` so the reviewer sees why the stage was flagged. Only `intake` with `qualityScore < 30` hard-fails.
 
 ---
 
@@ -180,17 +187,23 @@ interface SupervisionDecision {
 ```
 for each selected stage:
   attempt = 0
+  retryFeedback = ''
   loop:
+    agent = new AgentClass(context)
+    if retryFeedback: agent.setRetryFeedback(retryFeedback)   # corrective guidance
     result = await agent.execute(signal)
     decision = supervisorAgent.evaluate(stage, result)
-    if decision.verdict === 'pass' OR attempt >= maxRetries:
-      dispatch stageCompleted / stageAwaitingReview
+    if decision.pass OR attempt >= maxRetries:
+      dispatch stageCompleted (with supervisorDecision if !pass)
       break
+    retryFeedback = decision.reasons + reflectionNotes        # fed into next prompt
     attempt++
-    (retry with updated context)
 ```
 
 `maxRetries` (default `1`) is set in `PipelineConfig`. Set to `0` to disable supervision retries.
+On retry, the supervisor's `reasons` (plus any `reflectionNotes`) are injected as a corrective
+preamble into the agent's next prompt via `BaseAgent.setRetryFeedback`, so a retry is materially
+different rather than an identical re-roll.
 
 ---
 
@@ -248,29 +261,27 @@ Every edit, issue, or suggestion produced by an agent becomes a `ReviewItem`:
 - Per-item rationale and confidence score
 - Quick Accept for high-confidence non-critical items (≥ 0.85)
 
+### Applying Accepted Edits
+When a stage's review is submitted, accepted `ReviewItem`s for **editing stages**
+(`intake`, `structural`, `lineProse`, `copyEdit`, `proof`) are written back into the manuscript
+before the post-stage snapshot is taken (`services/proForge/applyReviewEdits.ts`):
+- Edits are applied **back-to-front by offset** so earlier edits never invalidate later offsets.
+- If an offset range no longer matches the recorded `original`, the applier falls back to locating
+  the `original` text; if it can't be found, the edit is **skipped** (counted, never force-applied).
+- Because the `project` slice is `redux-undo` wrapped, applied edits are natively undoable.
+- Non-editing stages (`production`, `publishing`, `analytics`) record decisions only.
+
 ### Snapshots
 - **Pre-pipeline snapshot:** Auto-created at start
 - **Pre-stage snapshot:** Created before each stage
-- **Post-stage snapshot:** Created after review acceptance
+- **Post-stage snapshot:** Created after accepted edits are applied (captures the edited text)
 - Rollback restores any pre-stage snapshot
 
----
-
-## Tool Calling
-
-Agents use the `toolRegistry` (`services/proForge/pipelineTools/toolRegistry.ts`) to interact with the manuscript:
-
-| Tool | Purpose | Stages |
-|------|---------|--------|
-| `readSection` | Read full section content | All |
-| `readAllSections` | List all sections (metadata) | All |
-| `readProjectMeta` | Read title, logline, characters, worlds | All |
-| `searchLore` | Search memory bank for lore | All |
-| `analyzePacing` | Compute tension scores per section | Intake, Structural |
-| `countWords` | Word count statistics | All |
-| `generateReport` | Save structured report to memory bank | All |
-| `getMemoryContext` | Retrieve relevant memory context | All |
-| `proposeEdit` | Queue an edit for Human-in-the-Loop review | Structural, Prose, CopyEdit |
+### Run History Persistence
+Completed and aborted runs are persisted per project to IndexedDB
+(`proforge-run-history`, capped at 20 most-recent) by `proForgeHistoryStore.ts`, and rehydrated on
+load by `useProForgeOrchestrator`. The `proForge` Redux slice itself stays ephemeral; this lets the
+`AnalyticsAgent`'s cross-run comparisons survive a reload.
 
 ---
 
@@ -280,8 +291,10 @@ The `ProForgeMemoryBank` (`services/proForge/proForgeMemoryBank.ts`) provides pe
 
 - **Storage:** IndexedDB (`proforge-memory-bank`)
 - **Categories:** `lore` | `character` | `style` | `feedback` | `edit` | `meta`
-- **Retrieval:** Keyword search + stage-aware filtering
-- **Context Building:** Automatically assembles memory blocks for prompt injection
+- **Stage-context recall:** `buildContextString` returns prior-stage + lore/character entries (chronological).
+- **Targeted search:** `search(query, limit, mode)` honours the run's `ragMode` — `lexical` (keyword),
+  or `semantic` / `hybrid` via the local MiniLM embedding service (`embedText` + cosine similarity),
+  with automatic keyword fallback if embeddings are unavailable (offline / model not loaded).
 
 ---
 
@@ -319,11 +332,12 @@ interface QualityGateReport { isFallback?: boolean; ... }
 // (on DiagnosticReport and StructuralEditPlan)
 reflectionNotes?: string;
 
-// Supervisor decision — attached to StageResult after supervision
+// Supervisor decision — attached to StageResult when a stage is flagged
 interface SupervisionDecision {
-  verdict: 'pass' | 'retry' | 'fail';
-  reason: string;
-  retryHint?: string;
+  pass: boolean;
+  retryRecommended: boolean;
+  qualityScore: number;   // 0–100
+  reasons: string[];
 }
 // On StageResult:
 supervisorDecision?: SupervisionDecision;
diff --git a/features/proForge/types.ts b/features/proForge/types.ts
index 96d26117f..6d02187ee 100644
--- a/features/proForge/types.ts
+++ b/features/proForge/types.ts
@@ -95,8 +95,6 @@ export interface PipelineConfig {
   language: string;
   /** Max supervisor-triggered retries per stage (0 = no retry, 1 = one retry) */
   maxRetries?: 0 | 1;
-  /** Active LoRA adapter ID — enables the optional StyleAdaptedWriterAgent stage. */
-  loraAdapterId?: string;
 }
 
 export const DEFAULT_PIPELINE_CONFIG: PipelineConfig = {
diff --git a/graphify-out/GRAPH_REPORT.md b/graphify-out/GRAPH_REPORT.md
index e737b34fd..4b385dde4 100644
--- a/graphify-out/GRAPH_REPORT.md
+++ b/graphify-out/GRAPH_REPORT.md
@@ -1,12 +1,12 @@
-# Graph Report - StoryCraft-Studio  (2026-06-05)
+# Graph Report - StoryCraft-Studio  (2026-06-10)
 
 ## Corpus Check
-- 948 files · ~585,302 words
+- 998 files · ~1,160,450 words
 - Verdict: corpus is large enough that graph structure adds value.
 
 ## Summary
-- 2945 nodes · 3896 edges · 58 communities detected
-- Extraction: 68% EXTRACTED · 32% INFERRED · 0% AMBIGUOUS · INFERRED: 1228 edges (avg confidence: 0.8)
+- 4475 nodes · 8039 edges · 54 communities detected
+- Extraction: 77% EXTRACTED · 23% INFERRED · 0% AMBIGUOUS · INFERRED: 1870 edges (avg confidence: 0.8)
 - Token cost: 0 input · 0 output
 
 ## Community Hubs (Navigation)
@@ -37,186 +37,182 @@
 - [[_COMMUNITY_Community 24|Community 24]]
 - [[_COMMUNITY_Community 25|Community 25]]
 - [[_COMMUNITY_Community 26|Community 26]]
-- [[_COMMUNITY_Community 29|Community 29]]
-- [[_COMMUNITY_Community 30|Community 30]]
+- [[_COMMUNITY_Community 27|Community 27]]
+- [[_COMMUNITY_Community 28|Community 28]]
 - [[_COMMUNITY_Community 31|Community 31]]
 - [[_COMMUNITY_Community 32|Community 32]]
 - [[_COMMUNITY_Community 34|Community 34]]
 - [[_COMMUNITY_Community 36|Community 36]]
 - [[_COMMUNITY_Community 37|Community 37]]
-- [[_COMMUNITY_Community 38|Community 38]]
+- [[_COMMUNITY_Community 41|Community 41]]
 - [[_COMMUNITY_Community 42|Community 42]]
-- [[_COMMUNITY_Community 43|Community 43]]
+- [[_COMMUNITY_Community 44|Community 44]]
 - [[_COMMUNITY_Community 45|Community 45]]
 - [[_COMMUNITY_Community 46|Community 46]]
-- [[_COMMUNITY_Community 47|Community 47]]
-- [[_COMMUNITY_Community 50|Community 50]]
-- [[_COMMUNITY_Community 58|Community 58]]
-- [[_COMMUNITY_Community 59|Community 59]]
-- [[_COMMUNITY_Community 66|Community 66]]
-- [[_COMMUNITY_Community 71|Community 71]]
-- [[_COMMUNITY_Community 72|Community 72]]
-- [[_COMMUNITY_Community 80|Community 80]]
-- [[_COMMUNITY_Community 85|Community 85]]
-- [[_COMMUNITY_Community 88|Community 88]]
-- [[_COMMUNITY_Community 90|Community 90]]
-- [[_COMMUNITY_Community 101|Community 101]]
-- [[_COMMUNITY_Community 133|Community 133]]
-- [[_COMMUNITY_Community 146|Community 146]]
-- [[_COMMUNITY_Community 151|Community 151]]
-- [[_COMMUNITY_Community 182|Community 182]]
-- [[_COMMUNITY_Community 224|Community 224]]
-- [[_COMMUNITY_Community 257|Community 257]]
-- [[_COMMUNITY_Community 262|Community 262]]
+- [[_COMMUNITY_Community 48|Community 48]]
+- [[_COMMUNITY_Community 55|Community 55]]
+- [[_COMMUNITY_Community 57|Community 57]]
+- [[_COMMUNITY_Community 61|Community 61]]
+- [[_COMMUNITY_Community 68|Community 68]]
+- [[_COMMUNITY_Community 73|Community 73]]
+- [[_COMMUNITY_Community 75|Community 75]]
+- [[_COMMUNITY_Community 84|Community 84]]
+- [[_COMMUNITY_Community 115|Community 115]]
+- [[_COMMUNITY_Community 129|Community 129]]
+- [[_COMMUNITY_Community 134|Community 134]]
+- [[_COMMUNITY_Community 164|Community 164]]
+- [[_COMMUNITY_Community 204|Community 204]]
+- [[_COMMUNITY_Community 237|Community 237]]
+- [[_COMMUNITY_Community 242|Community 242]]
 
 ## God Nodes (most connected - your core abstractions)
-1. `fn()` - 47 edges
-2. `StorageManager` - 36 edges
-3. `CloudSyncBackend` - 36 edges
-4. `retryFs()` - 31 edges
-5. `t()` - 29 edges
-6. `useTranslation()` - 28 edges
-7. `VoiceCommandService` - 28 edges
-8. `useAppDispatch()` - 25 edges
-9. `CollaborationService` - 24 edges
-10. `collectLibraryBackupPayload()` - 22 edges
+1. `mt()` - 103 edges
+2. `Bv` - 74 edges
+3. `fn()` - 47 edges
+4. `Ze()` - 43 edges
+5. `wx()` - 41 edges
+6. `xA` - 40 edges
+7. `t()` - 40 edges
+8. `CloudSyncBackend` - 39 edges
+9. `StorageManager` - 36 edges
+10. `tA()` - 34 edges
 
 ## Surprising Connections (you probably didn't know these)
-- `App()` --calls--> `useApp()`  [INFERRED]
-  App.tsx → hooks/useApp.ts
+- `checkForUpdate()` --calls--> `update()`  [INFERRED]
+  register-sw.ts → e2e-deep-report/trace/sw.bundle.js
 - `handleWebllmDownload()` --calls--> `generateLocalText()`  [INFERRED]
   components/settings/AiSections.tsx → services/localAiFacade.ts
 - `ViewLoader()` --calls--> `useTranslation()`  [INFERRED]
   App.tsx → hooks/useTranslation.ts
+- `App()` --calls--> `useApp()`  [INFERRED]
+  App.tsx → hooks/useApp.ts
 - `App()` --calls--> `useGlobalKeyboardShortcuts()`  [INFERRED]
   App.tsx → hooks/useGlobalKeyboardShortcuts.ts
-- `App()` --calls--> `usePushToTalk()`  [INFERRED]
-  App.tsx → hooks/usePushToTalk.ts
 
 ## Communities
 
 ### Community 0 - "Community 0"
-Cohesion: 0.02
-Nodes (54): AiInferenceCacheService, hashKey(), assertCloudAiAllowed(), assertCloudAiAllowedSync(), assertLoraLocalOnly(), binderDepth(), CloudSyncBackend, CloudSyncClient (+46 more)
+Cohesion: 0.01
+Nodes (301): _0, _2(), A0, a2(), aA(), ac(), ad(), Ah() (+293 more)
 
 ### Community 1 - "Community 1"
-Cohesion: 0.03
-Nodes (48): decryptCloudPayload(), deriveCloudSyncKey(), encryptCloudPayload(), loadStoryCodex(), IdbAssetStore, IdbCodexStore, compressData(), IdbConnectionManager (+40 more)
+Cohesion: 0.01
+Nodes (171): af(), ef(), ff(), Ja(), lf(), mt(), nf(), of() (+163 more)
 
 ### Community 2 - "Community 2"
-Cohesion: 0.03
-Nodes (45): createCancellationToken(), createAttentionPipeline(), createComputePipeline(), createKvCachePipeline(), createMlpPipeline(), createSimilarityBuffers(), createSimilarityPipeline(), encodeSimilarityUniforms() (+37 more)
+Cohesion: 0.01
+Nodes (117): recordLatency(), AiInferenceCacheService, hashKey(), assertCloudAiAllowed(), assertCloudAiAllowedSync(), assertLoraLocalOnly(), _cleanupPendingRequest(), _clearPendingRequestsForTest() (+109 more)
 
 ### Community 3 - "Community 3"
-Cohesion: 0.02
-Nodes (54): makeContext(), makeContext(), renderSheet(), makeDeps(), renderPanel(), makeStoreState(), createFakeAdapter(), createFakeDevice() (+46 more)
+Cohesion: 0.01
+Nodes (82): handleCopyForNotion(), handleDocxImport(), handleExport(), handlePasteImport(), handleBuildLocalRag(), handleWebllmDownload(), isCustomOllamaModel(), handleAddFolder() (+74 more)
 
 ### Community 4 - "Community 4"
 Cohesion: 0.02
-Nodes (49): AnalyticsBootstrap(), App(), ViewLoader(), Header(), useAppDispatch(), useAppSelectorShallow(), IdbUnlockModal(), useAnnounce() (+41 more)
+Nodes (63): pipeline(), pipeline(), getFocusable(), onKeyDown(), onPointerUp(), handleKeyDown(), applyPreset(), async() (+55 more)
 
 ### Community 5 - "Community 5"
 Cohesion: 0.02
-Nodes (27): buildKeyModuleMap(), loadBundleKeys(), loadModuleData(), getLocalUser(), getRandomColor(), sanitizeRoomInput(), stripControlChars(), loadFeatureFlagsState() (+19 more)
+Nodes (35): hasMigrationMarker(), legacyDatabaseListed(), migrateLegacyStorycraftDbIfNeeded(), openLegacyDatabase(), promisifyRequest(), readAllFromStore(), setMigrationMarker(), stateDbHasProjectOrSettings() (+27 more)
 
 ### Community 6 - "Community 6"
-Cohesion: 0.04
-Nodes (40): generateJson(), attachCause(), cleanPrompt(), sanitizePromptBlock(), stripControlChars(), stripJsonFences(), AnalyticsAgent, buildAiOpts() (+32 more)
+Cohesion: 0.02
+Nodes (73): CloudSyncBackend, decryptCloudPayload(), deriveCloudSyncKey(), encryptCloudPayload(), decryptDuckDbData(), encryptDuckDbData(), initDuckDbEncryption(), translate() (+65 more)
 
 ### Community 7 - "Community 7"
-Cohesion: 0.03
-Nodes (48): handleCopyForNotion(), handleDocxImport(), handleExport(), handlePasteImport(), handleRemoveKey(), handleSaveKey(), handleTestConnection(), handleAddFolder() (+40 more)
+Cohesion: 0.02
+Nodes (53): assertNoSeriousViolations(), AudioNavigator, navigateToCollaborationSettings(), ab(), B_(), br(), bs(), Bv (+45 more)
 
 ### Community 8 - "Community 8"
-Cohesion: 0.04
-Nodes (16): EcoModeService, FeedbackService, handleEcoToggle(), LoraEnvReport, LoraTrainPayload, train_lora(), ConsentRequiredError, createSttEngine() (+8 more)
+Cohesion: 0.02
+Nodes (50): item(), glossaryTranslate(), loadCheckpoint(), loadGlossary(), main(), parseArgs(), saveCheckpoint(), sleep() (+42 more)
 
 ### Community 9 - "Community 9"
-Cohesion: 0.03
-Nodes (29): CircuitBreaker, translate(), minimalProject(), getNotifications(), handleEvaluate(), ScoreGauge(), comparePromptOutputs(), computeStyleConsistencyScore() (+21 more)
+Cohesion: 0.02
+Nodes (63): clampRetryAfter(), computeRetryDelayMs(), delay(), parseRetryAfterMs(), retryAfterStringToMs(), withTransientRetry(), makeContext(), makeContext() (+55 more)
 
 ### Community 10 - "Community 10"
-Cohesion: 0.04
-Nodes (47): handleBuildLocalRag(), handleWebllmDownload(), isCustomOllamaModel(), duckdbCodexWrite(), duckdbCrossProjectWrite(), duckdbDualWrite(), duckdbRagWrite(), esc() (+39 more)
+Cohesion: 0.02
+Nodes (53): AnalyticsBootstrap(), App(), ViewLoader(), Header(), useAppDispatch(), useAppSelectorShallow(), IdbUnlockModal(), useAnnounce() (+45 more)
 
 ### Community 11 - "Community 11"
 Cohesion: 0.04
-Nodes (34): deleteIdb(), formatStorageError(), initializeStorage(), resetAllDatabases(), routeTask(), addDebouncedListener(), initAdaptiveAiOnStartup(), initWorkerBusOnStartup() (+26 more)
+Nodes (54): generateJson(), attachCause(), cleanPrompt(), sanitizePromptBlock(), stripControlChars(), stripJsonFences(), AnalyticsAgent, handleRemoveKey() (+46 more)
 
 ### Community 12 - "Community 12"
-Cohesion: 0.04
-Nodes (35): generateMessageId(), getWorker(), send(), getDuckDb(), initDuckDb(), isOPFSSupported(), buildTimeoutSignal(), createStoryCraftFetch() (+27 more)
+Cohesion: 0.03
+Nodes (47): AdaptiveAiEngine, _clearLatencyHistory(), estimateLatency(), getTaskConfig(), selectModelForBackend(), start(), getLastBenchmarkResults(), loadResults() (+39 more)
 
 ### Community 13 - "Community 13"
-Cohesion: 0.04
-Nodes (21): pipeline(), pipeline(), start(), decrypt(), decryptJson(), encrypt(), encryptJson(), GpuResourceManager (+13 more)
+Cohesion: 0.03
+Nodes (35): CollabEncryptionRequiredError, CollaborationService, resolveWebRtcSignalingUrls(), MockDoc, MockWebrtcProvider, createAttentionPipeline(), createComputePipeline(), createKvCachePipeline() (+27 more)
 
 ### Community 14 - "Community 14"
-Cohesion: 0.09
-Nodes (16): FsAssetStore, FsCodexStore, countProjectWords(), decompressData(), decryptText(), deriveFileSystemCryptoKey(), encryptText(), FsCore (+8 more)
+Cohesion: 0.03
+Nodes (45): check(), green(), grep(), hasRuntimeConsumption(), read(), red(), collect(), id (+37 more)
 
 ### Community 15 - "Community 15"
 Cohesion: 0.05
-Nodes (38): applyPreset(), async(), close(), countWords(), enrichProjectIndex(), extractCharacterNames(), getDb(), indexProject() (+30 more)
+Nodes (42): NT, getDuckDb(), handleExec(), handleQuery(), handleShutdown(), initDuckDb(), isOPFSSupported(), duckdbCodexWrite() (+34 more)
 
 ### Community 16 - "Community 16"
-Cohesion: 0.06
-Nodes (28): _clearLatencyHistory(), _clearPendingRequestsForTest(), clearCommunityTemplateCache(), clearServiceWorkerCaches(), deleteAllIndexedDBDatabases(), wipeAllAppData(), clearEmbeddingCache(), detectOnnxExecutionProviders() (+20 more)
+Cohesion: 0.08
+Nodes (20): FsAssetStore, FsCodexStore, deleteIdb(), formatStorageError(), initializeStorage(), resetAllDatabases(), countProjectWords(), decompressData() (+12 more)
 
 ### Community 17 - "Community 17"
-Cohesion: 0.07
-Nodes (37): _cleanupPendingRequest(), _deduplicateRequest(), generateText(), generateTextSingleProvider(), _pendingKey(), streamAiHelpResponse(), streamAnthropic(), streamGrok() (+29 more)
+Cohesion: 0.04
+Nodes (27): DeadLetterQueue, openDlqDb(), storeClear(), storeGetAll(), k2, download_artifact(), get_failed_logs(), get_latest_failed_run() (+19 more)
 
 ### Community 18 - "Community 18"
-Cohesion: 0.07
-Nodes (28): AdaptiveAiEngine, estimateLatency(), recordLatency(), selectModelForBackend(), getLastBenchmarkResults(), loadResults(), runAllBenchmarks(), runInferenceBenchmark() (+20 more)
+Cohesion: 0.04
+Nodes (38): countWords(), enrichProjectIndex(), extractCharacterNames(), getDb(), indexProject(), listIndexedProjects(), removeProjectIndex(), semanticSearchProjects() (+30 more)
 
 ### Community 19 - "Community 19"
-Cohesion: 0.07
-Nodes (25): check(), green(), grep(), hasRuntimeConsumption(), read(), red(), formatArgs(), formatLogsForReport() (+17 more)
+Cohesion: 0.06
+Nodes (26): mockT(), aE, iE, lE(), rE, sE, analyzeSentiment(), classifyWritingTopic() (+18 more)
 
 ### Community 20 - "Community 20"
-Cohesion: 0.11
-Nodes (1): StorageManager
+Cohesion: 0.09
+Nodes (38): analyze_stryker_failure(), analyze_vitest_failure(), analyze_with_llm(), format_for_vscode(), get_openrouter_client(), main(), Send preprocessed errors to LLM for analysis., Format errors for VS Code problem matcher. (+30 more)
 
 ### Community 21 - "Community 21"
-Cohesion: 0.09
-Nodes (5): CollabEncryptionRequiredError, CollaborationService, resolveWebRtcSignalingUrls(), MockDoc, MockWebrtcProvider
+Cohesion: 0.08
+Nodes (27): generateTextSingleProvider(), _pendingKey(), streamAiHelpResponse(), streamAnthropic(), streamGrok(), streamOpenAI(), streamProvider(), testAIConnection() (+19 more)
 
 ### Community 22 - "Community 22"
-Cohesion: 0.14
-Nodes (21): handleToggle(), handleDelete(), handleFileChange(), activateAdapter(), clearDatasetEntries(), deactivateAdapter(), deleteAdapter(), exportAdapter() (+13 more)
+Cohesion: 0.1
+Nodes (1): StorageManager
 
 ### Community 23 - "Community 23"
-Cohesion: 0.16
-Nodes (14): buildExcerpt(), extractCharacters(), extractManuscriptSections(), searchAcrossProjectIndex(), searchAcrossProjects(), normalizeSearch(), scoreAgainstQuery(), subsequenceScore() (+6 more)
+Cohesion: 0.21
+Nodes (4): LS, Th(), xn(), aa
 
 ### Community 24 - "Community 24"
-Cohesion: 0.14
-Nodes (12): navigateToCollaborationSettings(), clickNavItem(), ensureBlankProject(), flushWriterDebounce(), seedGeminiApiKey(), selectFirstEnabledWriterSection(), waitForMainChrome(), waitForSpaReady() (+4 more)
+Cohesion: 0.16
+Nodes (11): cE(), fr(), Go(), jS(), ri(), v_(), wb(), Xd (+3 more)
 
 ### Community 25 - "Community 25"
-Cohesion: 0.18
-Nodes (14): analyze_text(), count_sentences(), count_syllables(), counts_words_chars_and_spaces(), empty_text_is_all_zero(), flesch_score_is_finite_for_real_prose(), run_text_analyze(), RustTaskRequest (+6 more)
+Cohesion: 0.14
+Nodes (21): handleToggle(), handleDelete(), handleFileChange(), activateAdapter(), clearDatasetEntries(), deactivateAdapter(), deleteAdapter(), exportAdapter() (+13 more)
 
 ### Community 26 - "Community 26"
+Cohesion: 0.14
+Nodes (14): buildExcerpt(), extractCharacters(), extractManuscriptSections(), searchAcrossProjectIndex(), searchAcrossProjects(), normalizeSearch(), scoreAgainstQuery(), subsequenceScore() (+6 more)
+
+### Community 27 - "Community 27"
+Cohesion: 0.35
+Nodes (2): cc, Gb()
+
+### Community 28 - "Community 28"
 Cohesion: 0.19
 Nodes (6): buildNormManuscriptExport(), paginateNormLines(), stripLightMarkdown(), wrapParagraphToLines(), wrapPlainTextToNormLines(), UsageAnalyticsService
 
-### Community 29 - "Community 29"
+### Community 31 - "Community 31"
 Cohesion: 0.22
 Nodes (4): accessibilityPresetDefaults(), normalizeAccessibilitySettings(), applyPreset(), baseSettings()
 
-### Community 30 - "Community 30"
-Cohesion: 0.29
-Nodes (1): PriorityTaskQueue
-
-### Community 31 - "Community 31"
-Cohesion: 0.33
-Nodes (1): AudioNavigator
-
 ### Community 32 - "Community 32"
-Cohesion: 0.46
+Cohesion: 0.42
 Nodes (6): emit(), main(), merge(), ProgressCallback, Emits JSON progress events on each training log step., train()
 
 ### Community 34 - "Community 34"
@@ -228,171 +224,145 @@ Cohesion: 0.53
 Nodes (4): buildWebNNExecutionProviders(), detectWebNN(), isDirectMLAvailable(), isDirectMLHeuristic()
 
 ### Community 37 - "Community 37"
-Cohesion: 0.4
-Nodes (2): getFocusable(), onKeyDown()
-
-### Community 38 - "Community 38"
 Cohesion: 0.7
 Nodes (4): check_cuda_and_vram(), check_package(), check_python_version(), main()
 
-### Community 42 - "Community 42"
+### Community 41 - "Community 41"
 Cohesion: 0.4
 Nodes (2): useDashboardContext(), DashboardHeader()
 
-### Community 43 - "Community 43"
+### Community 42 - "Community 42"
 Cohesion: 0.4
 Nodes (4): Room, SignalingConn, WebrtcConn, WebrtcProvider
 
-### Community 45 - "Community 45"
+### Community 44 - "Community 44"
 Cohesion: 0.6
 Nodes (4): applyFormula(), computeReadabilitySnapshot(), estimateSyllables(), getSyllablePattern()
 
-### Community 46 - "Community 46"
+### Community 45 - "Community 45"
 Cohesion: 0.67
 Nodes (2): sanitizeSpeechTranscript(), stripControlChars()
 
-### Community 47 - "Community 47"
+### Community 46 - "Community 46"
 Cohesion: 0.5
 Nodes (1): SpeechSynthesisUtteranceMock
 
-### Community 50 - "Community 50"
+### Community 48 - "Community 48"
 Cohesion: 0.67
 Nodes (2): makeConfig(), startPipelinePayload()
 
-### Community 58 - "Community 58"
+### Community 55 - "Community 55"
 Cohesion: 0.67
 Nodes (2): defaultProject(), setProjectData()
 
-### Community 59 - "Community 59"
+### Community 57 - "Community 57"
 Cohesion: 0.5
 Nodes (3): AsyncDuckDB, AsyncDuckDBConnection, ConsoleLogger
 
-### Community 66 - "Community 66"
-Cohesion: 0.67
-Nodes (2): getFocusable(), handleTabKey()
-
-### Community 71 - "Community 71"
+### Community 61 - "Community 61"
 Cohesion: 0.67
 Nodes (2): getQuestionsForArchetype(), getTemplateForArchetype()
 
-### Community 72 - "Community 72"
-Cohesion: 0.67
-Nodes (2): buildLcsTable(), diffTokensToOps()
-
-### Community 80 - "Community 80"
+### Community 68 - "Community 68"
 Cohesion: 0.67
 Nodes (1): makeSection()
 
-### Community 85 - "Community 85"
+### Community 73 - "Community 73"
 Cohesion: 0.67
 Nodes (1): MockGoogleGenAI
 
-### Community 88 - "Community 88"
+### Community 75 - "Community 75"
 Cohesion: 0.67
 Nodes (1): makeDeps()
 
-### Community 90 - "Community 90"
-Cohesion: 1.0
-Nodes (2): fireSwipe(), makePointerEvent()
-
-### Community 101 - "Community 101"
+### Community 84 - "Community 84"
 Cohesion: 0.67
 Nodes (1): TaskError
 
-### Community 133 - "Community 133"
+### Community 115 - "Community 115"
 Cohesion: 1.0
 Nodes (1): MockIntersectionObserver
 
-### Community 146 - "Community 146"
+### Community 129 - "Community 129"
 Cohesion: 1.0
 Nodes (1): MockWorker
 
-### Community 151 - "Community 151"
+### Community 134 - "Community 134"
 Cohesion: 1.0
 Nodes (1): MockBroadcastChannel
 
-### Community 182 - "Community 182"
+### Community 164 - "Community 164"
 Cohesion: 1.0
 Nodes (1): MockIntersectionObserver
 
-### Community 224 - "Community 224"
+### Community 204 - "Community 204"
 Cohesion: 1.0
 Nodes (1): MockWorker
 
-### Community 257 - "Community 257"
+### Community 237 - "Community 237"
 Cohesion: 1.0
 Nodes (1): FileSystemService
 
-### Community 262 - "Community 262"
+### Community 242 - "Community 242"
 Cohesion: 1.0
 Nodes (1): IndexedDBService
 
 ## Knowledge Gaps
-- **26 isolated node(s):** `Emits JSON progress events on each training log step.`, `MockIntersectionObserver`, `MockWorker`, `MockGoogleGenAI`, `MockBroadcastChannel` (+21 more)
+- **47 isolated node(s):** `Emits JSON progress events on each training log step.`, `Remove ANSI escape codes from text.`, `Remove timestamp strings from text.`, `Replace long base64 strings with placeholder.`, `Remove NPM/pnpm warning lines.` (+42 more)
   These have ≤1 connection - possible missing edges or undocumented components.
-- **Thin community `Community 20`** (36 nodes): `storageService.ts`, `StorageManager`, `.clearApiKey()`, `.clearGeminiApiKey()`, `.constructor()`, `.deleteAllBinderAssetsForProject()`, `.deleteBinderAsset()`, `.deleteImage()`, `.deleteProject()`, `.deleteRagVectors()`, `.deleteSnapshot()`, `.deleteStoryCodex()`, `.getApiKey()`, `.getBackend()`, `.getBinderAsset()`, `.getGeminiApiKey()`, `.getImage()`, `.getRagVectors()`, `.getSnapshotData()`, `.getStoryCodex()`, `.hasSavedData()`, `.initializeBackend()`, `.listBinderAssetIds()`, `.listProjects()`, `.listSnapshots()`, `.loadProject()`, `.loadSettings()`, `.saveApiKey()`, `.saveBinderAsset()`, `.saveGeminiApiKey()`, `.saveImage()`, `.saveProject()`, `.saveRagVectors()`, `.saveSettings()`, `.saveSnapshot()`, `.saveStoryCodex()`
-  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 30`** (10 nodes): `taskQueue.ts`, `PriorityTaskQueue`, `.constructor()`, `.dequeue()`, `.effectivePriority()`, `.enqueue()`, `.peek()`, `.promoteStarvedTasks()`, `.stats()`, `.totalDepth()`
-  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 31`** (9 nodes): `AudioNavigator`, `.announce()`, `.focusElement()`, `.focusFirstIn()`, `.getFocusedLabel()`, `.nextLandmark()`, `.previousLandmark()`, `.scanLandmarks()`, `audioNavigator.ts`
-  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 37`** (6 nodes): `getFocusable()`, `onKeyDown()`, `onPointerDown()`, `onPointerMove()`, `onPointerUp()`, `BottomSheet.tsx`
-  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 42`** (5 nodes): `DashboardHeader.tsx`, `DashboardContext.ts`, `useDashboardContext()`, `Chip()`, `DashboardHeader()`
-  Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 46`** (4 nodes): `useSpeechRecognition.ts`, `sanitizeSpeechTranscript()`, `stripControlChars()`, `useSpeechRecognition()`
+- **Thin community `Community 22`** (37 nodes): `.initialize()`, `storageService.ts`, `StorageManager`, `.clearApiKey()`, `.clearGeminiApiKey()`, `.constructor()`, `.deleteAllBinderAssetsForProject()`, `.deleteBinderAsset()`, `.deleteImage()`, `.deleteProject()`, `.deleteRagVectors()`, `.deleteSnapshot()`, `.deleteStoryCodex()`, `.getApiKey()`, `.getBackend()`, `.getBinderAsset()`, `.getGeminiApiKey()`, `.getImage()`, `.getRagVectors()`, `.getSnapshotData()`, `.getStoryCodex()`, `.hasSavedData()`, `.initializeBackend()`, `.listBinderAssetIds()`, `.listProjects()`, `.listSnapshots()`, `.loadProject()`, `.loadSettings()`, `.saveApiKey()`, `.saveBinderAsset()`, `.saveGeminiApiKey()`, `.saveImage()`, `.saveProject()`, `.saveRagVectors()`, `.saveSettings()`, `.saveSnapshot()`, `.saveStoryCodex()`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 47`** (4 nodes): `makeStorageMock()`, `SpeechSynthesisUtteranceMock`, `.constructor()`, `setup.ts`
+- **Thin community `Community 27`** (17 nodes): `cc`, `._applyAttribute()`, `._assert()`, `.constructor()`, `._eof()`, `._isWhitespace()`, `._next()`, `.parse()`, `._peek()`, `._readAttributes()`, `._readIdentifier()`, `._readRegex()`, `._readString()`, `._readStringOrRegex()`, `._skipWhitespace()`, `._throwError()`, `Gb()`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 50`** (4 nodes): `makeConfig()`, `makeReviewItem()`, `startPipelinePayload()`, `proForgeSlice.test.ts`
+- **Thin community `Community 41`** (5 nodes): `DashboardHeader.tsx`, `DashboardContext.ts`, `useDashboardContext()`, `Chip()`, `DashboardHeader()`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 58`** (4 nodes): `useDashboard.test.ts`, `defaultProject()`, `defaultSection()`, `setProjectData()`
+- **Thin community `Community 45`** (4 nodes): `useSpeechRecognition.ts`, `sanitizeSpeechTranscript()`, `stripControlChars()`, `useSpeechRecognition()`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 66`** (4 nodes): `Drawer.tsx`, `getFocusable()`, `handleEsc()`, `handleTabKey()`
+- **Thin community `Community 46`** (4 nodes): `makeStorageMock()`, `SpeechSynthesisUtteranceMock`, `.constructor()`, `setup.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 71`** (4 nodes): `getAllTemplates()`, `getQuestionsForArchetype()`, `getTemplateForArchetype()`, `characterInterviewTemplates.ts`
+- **Thin community `Community 48`** (4 nodes): `makeConfig()`, `makeReviewItem()`, `startPipelinePayload()`, `proForgeSlice.test.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 72`** (4 nodes): `wordDiff.ts`, `buildLcsTable()`, `diffTokensToOps()`, `tokenizeWordsAndSpaces()`
+- **Thin community `Community 55`** (4 nodes): `useDashboard.test.ts`, `defaultProject()`, `defaultSection()`, `setProjectData()`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 80`** (3 nodes): `makeSection()`, `plotBoardService.test.ts`, `plotBoardService.test.ts`
+- **Thin community `Community 61`** (4 nodes): `getAllTemplates()`, `getQuestionsForArchetype()`, `getTemplateForArchetype()`, `characterInterviewTemplates.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 85`** (3 nodes): `makeStream()`, `MockGoogleGenAI`, `geminiService.test.ts`
+- **Thin community `Community 68`** (3 nodes): `makeSection()`, `plotBoardService.test.ts`, `plotBoardService.test.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 88`** (3 nodes): `makeDeps()`, `aiSuggestions.test.ts`, `aiSuggestions.test.ts`
+- **Thin community `Community 73`** (3 nodes): `makeStream()`, `MockGoogleGenAI`, `geminiService.test.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 90`** (3 nodes): `useSwipeGesture.test.ts`, `fireSwipe()`, `makePointerEvent()`
+- **Thin community `Community 75`** (3 nodes): `makeDeps()`, `aiSuggestions.test.ts`, `aiSuggestions.test.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 101`** (3 nodes): `types.ts`, `TaskError`, `.constructor()`
+- **Thin community `Community 84`** (3 nodes): `types.ts`, `TaskError`, `.constructor()`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 133`** (2 nodes): `MockIntersectionObserver`, `BookPreviewView.test.tsx`
+- **Thin community `Community 115`** (2 nodes): `MockIntersectionObserver`, `BookPreviewView.test.tsx`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 146`** (2 nodes): `MockWorker`, `duckdbClient.test.ts`
+- **Thin community `Community 129`** (2 nodes): `MockWorker`, `duckdbClient.test.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 151`** (2 nodes): `MockBroadcastChannel`, `tabLeaderElection.test.ts`
+- **Thin community `Community 134`** (2 nodes): `MockBroadcastChannel`, `tabLeaderElection.test.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 182`** (2 nodes): `useBookPreviewView.test.ts`, `MockIntersectionObserver`
+- **Thin community `Community 164`** (2 nodes): `useBookPreviewView.test.ts`, `MockIntersectionObserver`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 224`** (2 nodes): `workerPool.test.ts`, `MockWorker`
+- **Thin community `Community 204`** (2 nodes): `workerPool.test.ts`, `MockWorker`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 257`** (2 nodes): `FileSystemService`, `index.ts`
+- **Thin community `Community 237`** (2 nodes): `FileSystemService`, `index.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
-- **Thin community `Community 262`** (2 nodes): `IndexedDBService`, `index.ts`
+- **Thin community `Community 242`** (2 nodes): `IndexedDBService`, `index.ts`
   Too small to be a meaningful cluster - may be noise or needs more connections extracted.
 
 ## Suggested Questions
 _Questions this graph is uniquely positioned to answer:_
 
-- **Why does `fn()` connect `Community 3` to `Community 0`, `Community 1`, `Community 6`, `Community 14`, `Community 17`?**
-  _High betweenness centrality (0.064) - this node is a cross-community bridge._
-- **Why does `t()` connect `Community 7` to `Community 10`, `Community 2`, `Community 4`, `Community 22`?**
-  _High betweenness centrality (0.048) - this node is a cross-community bridge._
-- **Why does `useTranslation()` connect `Community 4` to `Community 7`?**
-  _High betweenness centrality (0.028) - this node is a cross-community bridge._
+- **Why does `t()` connect `Community 3` to `Community 0`, `Community 1`, `Community 2`, `Community 5`, `Community 7`, `Community 10`, `Community 11`, `Community 14`, `Community 25`?**
+  _High betweenness centrality (0.078) - this node is a cross-community bridge._
+- **Why does `mt()` connect `Community 1` to `Community 0`, `Community 2`, `Community 3`, `Community 4`, `Community 5`, `Community 7`, `Community 11`, `Community 14`, `Community 19`, `Community 23`, `Community 24`?**
+  _High betweenness centrality (0.068) - this node is a cross-community bridge._
+- **Why does `wx()` connect `Community 0` to `Community 1`, `Community 2`, `Community 3`, `Community 4`, `Community 7`, `Community 13`, `Community 16`?**
+  _High betweenness centrality (0.057) - this node is a cross-community bridge._
+- **Are the 86 inferred relationships involving `mt()` (e.g. with `pE()` and `xE()`) actually correct?**
+  _`mt()` has 86 INFERRED edges - model-reasoned connections that need verification._
 - **Are the 46 inferred relationships involving `fn()` (e.g. with `makeMediaQuery()` and `MockSpeechRecognition()`) actually correct?**
   _`fn()` has 46 INFERRED edges - model-reasoned connections that need verification._
-- **Are the 30 inferred relationships involving `retryFs()` (e.g. with `.saveProject()` and `.loadProject()`) actually correct?**
-  _`retryFs()` has 30 INFERRED edges - model-reasoned connections that need verification._
-- **Are the 28 inferred relationships involving `t()` (e.g. with `useHelpView()` and `handleSaveKey()`) actually correct?**
-  _`t()` has 28 INFERRED edges - model-reasoned connections that need verification._
-- **What connects `Emits JSON progress events on each training log step.`, `MockIntersectionObserver`, `MockWorker` to the rest of the system?**
-  _26 weakly-connected nodes found - possible documentation gaps or missing edges._
\ No newline at end of file
+- **Are the 16 inferred relationships involving `wx()` (e.g. with `for()` and `.addEventListener()`) actually correct?**
+  _`wx()` has 16 INFERRED edges - model-reasoned connections that need verification._
+- **What connects `Emits JSON progress events on each training log step.`, `Remove ANSI escape codes from text.`, `Remove timestamp strings from text.` to the rest of the system?**
+  _47 weakly-connected nodes found - possible documentation gaps or missing edges._
\ No newline at end of file
diff --git a/hooks/useProForgeOrchestrator.ts b/hooks/useProForgeOrchestrator.ts
index de4698e6b..1869b8c34 100644
--- a/hooks/useProForgeOrchestrator.ts
+++ b/hooks/useProForgeOrchestrator.ts
@@ -3,8 +3,9 @@
  * QNBS-v3: Bridges Redux state, orchestrator, and UI components.
  */
 
-import { useCallback, useMemo, useRef } from 'react';
+import { useCallback, useEffect, useMemo, useRef } from 'react';
 import { useAppDispatch, useAppSelector } from '../app/hooks';
+import { appStoreRef } from '../app/storeRef';
 import { proForgeActions } from '../features/proForge/proForgeSlice';
 import type { PipelineConfig, PipelineStage, ReviewItemStatus } from '../features/proForge/types';
 import { createProForgeOrchestrator } from '../services/proForge/proForgeOrchestrator';
@@ -50,17 +51,47 @@ export function useProForgeOrchestrator() {
     [settings, featureFlags, language],
   );
 
+  // QNBS-v3: Track which project the cached orchestrator was built for, so we rebuild
+  // it (and drop its stale AbortController/context) when the user switches projects.
+  const orchestratorProjectIdRef = useRef<string | null>(null);
+
+  // QNBS-v3: Hydrate persisted run history when a project loads — analytics comparisons across
+  // runs were lost on reload because the proForge slice is ephemeral. Best-effort.
+  const projectId = project?.id;
+  useEffect(() => {
+    if (!projectId) return;
+    let cancelled = false;
+    void (async () => {
+      try {
+        const { loadRunHistory } = await import('../services/proForge/proForgeHistoryStore');
+        const runs = await loadRunHistory(projectId);
+        if (!cancelled && runs.length > 0) {
+          dispatch(proForgeActions.loadRunHistory(runs));
+        }
+      } catch {
+        // Non-blocking: history hydration failures must not break the view.
+      }
+    })();
+    return () => {
+      cancelled = true;
+    };
+  }, [projectId, dispatch]);
+
   const getOrchestrator = useCallback(() => {
     if (!project) return null;
+    const projectId = project.id || 'default';
+
+    if (orchestratorRef.current && orchestratorProjectIdRef.current !== projectId) {
+      orchestratorRef.current.dispose();
+      orchestratorRef.current = null;
+    }
+
     if (!orchestratorRef.current) {
       orchestratorRef.current = createProForgeOrchestrator({
         dispatch,
-        getState: () =>
-          ({
-            ...orchestratorRef.current!.context.getState(),
-            proForge: proForgeState,
-          }) as import('../app/store').RootState,
-        projectId: project.id || 'default',
+        // QNBS-v3: Read the LIVE store (was self-referential → infinite recursion + stale state).
+        getState: () => appStoreRef.current!.getState(),
+        projectId,
         manuscript: project.manuscript.map((s) => ({
           id: s.id,
           title: s.title,
@@ -72,11 +103,16 @@ export function useProForgeOrchestrator() {
         worlds: Object.values(project.worlds?.entities ?? {})
           .filter(Boolean)
           .map((w) => ({ id: (w as { id: string }).id, name: (w as { name: string }).name })),
-        config: currentRun?.config ?? defaultConfig,
+        // QNBS-v3: Live config getter — agents read this.context.config at execute time,
+        // so a custom config passed to startPipeline must win over the build-time default.
+        get config() {
+          return appStoreRef.current!.getState().proForge.currentRun?.config ?? defaultConfig;
+        },
       });
+      orchestratorProjectIdRef.current = projectId;
     }
     return orchestratorRef.current;
-  }, [dispatch, project, proForgeState, defaultConfig, currentRun?.config]);
+  }, [dispatch, project, defaultConfig]);
 
   const startPipeline = useCallback(
     async (label: string, config: PipelineConfig) => {
diff --git a/locales/ar/common.json b/locales/ar/common.json
index b1c33aef2..57f5eeb8c 100644
--- a/locales/ar/common.json
+++ b/locales/ar/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "جارٍ النظر في شكل قصتك…",
   "proforge.pipeline.noneActive": "لا يوجد خط أنابيب نشط",
   "proforge.pipeline.title": "خط أنابيب المؤلّف الأمثل",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "كلمات مكتوبة في الجلسات",
   "progress.allTime.title": "كل الأوقات",
   "progress.bestDay.title": "أفضل يوم",
diff --git a/locales/de/common.json b/locales/de/common.json
index e71f3f19f..0b8e1f18d 100644
--- a/locales/de/common.json
+++ b/locales/de/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Betrachtet die Struktur deiner Geschichte…",
   "proforge.pipeline.noneActive": "Keine aktive Pipeline",
   "proforge.pipeline.title": "Ultimative Autoren-Pipeline",
+  "proforge.stageName.idle": "Leerlauf",
+  "proforge.stageName.intake": "Aufnahme & Diagnose",
+  "proforge.stageName.structural": "Strukturlektorat",
+  "proforge.stageName.lineProse": "Sprach- & Prosalektorat",
+  "proforge.stageName.copyEdit": "Schlusslektorat",
+  "proforge.stageName.proof": "Korrektorat",
+  "proforge.stageName.production": "Produktion",
+  "proforge.stageName.publishing": "Veröffentlichung",
+  "proforge.stageName.analytics": "Auswertung",
+  "proforge.stageName.archived": "Archiviert",
+  "proforge.review.heading": "Prüfung: {{stage}}",
+  "proforge.review.counts": "{{pending}} offen · {{accepted}} übernommen · {{rejected}} abgelehnt",
+  "proforge.review.quickAccept": "Schnell übernehmen ({{count}})",
+  "proforge.review.quickAcceptTitle": "Alle unkritischen Vorschläge mit hoher Konfidenz übernehmen",
+  "proforge.review.acceptAll": "Alle übernehmen",
+  "proforge.review.rejectAll": "Alle ablehnen",
+  "proforge.review.criticalHeadingOne": "{{count}} kritisches Problem erfordert Aufmerksamkeit",
+  "proforge.review.criticalHeadingOther": "{{count}} kritische Probleme erfordern Aufmerksamkeit",
+  "proforge.review.acceptAllCritical": "Alle kritischen übernehmen",
+  "proforge.review.moreCritical": "+{{count}} weitere kritische Probleme unten",
+  "proforge.review.filter.all": "Alle",
+  "proforge.review.filter.pending": "Offen",
+  "proforge.review.filter.accepted": "Übernommen",
+  "proforge.review.filter.rejected": "Abgelehnt",
+  "proforge.review.noItems": "Keine Prüfpunkte verfügbar.",
+  "proforge.review.noItemsForFilter": "Keine Einträge für diesen Filter.",
+  "proforge.review.skipStage": "Phase überspringen",
+  "proforge.review.submitContinue": "Übernehmen & fortfahren",
+  "proforge.review.pendingCount": "{{count}} offen",
+  "proforge.review.confidence": "{{percent}} % Konfidenz",
+  "proforge.review.original": "Original:",
+  "proforge.review.proposed": "Vorschlag:",
+  "proforge.review.showDetails": "Details anzeigen",
+  "proforge.review.showLess": "Weniger anzeigen",
+  "proforge.review.accept": "Übernehmen",
+  "proforge.review.reject": "Ablehnen",
+  "proforge.review.ignore": "Ignorieren",
+  "proforge.review.group.critical": "Kritische Maßnahmen",
+  "proforge.review.group.warning": "Warnungen",
+  "proforge.review.group.info": "Vorschläge",
+  "proforge.review.severity.critical": "Kritisch",
+  "proforge.review.severity.warning": "Warnung",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.announceSubmitted": "Prüfung {{stage}} übermittelt",
+  "proforge.review.type.structuralEdit": "Struktur",
+  "proforge.review.type.proseEdit": "Prosa",
+  "proforge.review.type.grammarEdit": "Grammatik",
+  "proforge.review.type.styleEdit": "Stil",
+  "proforge.review.type.repetitionHit": "Wiederholung",
+  "proforge.review.type.consistencyIssue": "Konsistenz",
+  "proforge.review.type.plotHole": "Logiklücke",
+  "proforge.review.type.legalWarning": "Rechtliches",
+  "proforge.review.type.technicalIssue": "Technik",
+  "proforge.review.type.pacingIssue": "Tempo",
+  "proforge.review.type.arcIssue": "Spannungsbogen",
   "progress.allTime.label": "in Sitzungen geschriebene Wörter",
   "progress.allTime.title": "Gesamt",
   "progress.bestDay.title": "Bester Tag",
diff --git a/locales/el/common.json b/locales/el/common.json
index c6f7bb448..8b1590ea6 100644
--- a/locales/el/common.json
+++ b/locales/el/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Κοιτάζοντας τη μορφή της ιστορίας σας…",
   "proforge.pipeline.noneActive": "Δεν υπάρχει ενεργός αγωγός",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "λέξεις γραμμένες σε συνεδρίες",
   "progress.allTime.title": "Όλων των Εποχών",
   "progress.bestDay.title": "Καλύτερη μέρα",
diff --git a/locales/en/common.json b/locales/en/common.json
index f2fada71e..6810e724e 100644
--- a/locales/en/common.json
+++ b/locales/en/common.json
@@ -491,6 +491,61 @@
   "proforge.emptyState.title": "Your manuscript, refined.",
   "proforge.emptyState.description": "ProForge reads every chapter, then walks you through editing stage by stage. Nothing gets lost — every change is reversible.",
   "proforge.enabledHint": "ProForge Pipeline enabled. Open the Writer view and click the ProForge button in the tools panel to start.",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.structural": "Structural Edit",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.original": "Original:",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.accept": "Accept",
+  "proforge.review.reject": "Reject",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.arcIssue": "Arc",
   "sceneboard.emptyState.title": "Your plot board is a blank canvas",
   "sceneboard.emptyState.description": "Add your first scene or import your outline to start planning your story.",
   "characters.emptyState.title": "Your cast list is empty",
diff --git a/locales/es/common.json b/locales/es/common.json
index 0bf2ad516..87b3ec96d 100644
--- a/locales/es/common.json
+++ b/locales/es/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Observando la forma de tu historia…",
   "proforge.pipeline.noneActive": "Sin pipeline activa",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "palabras escritas en sesiones",
   "progress.allTime.title": "Total histórico",
   "progress.bestDay.title": "Mejor día",
diff --git a/locales/es/settings.json b/locales/es/settings.json
index 02c1cdd8c..9e6391f1f 100644
--- a/locales/es/settings.json
+++ b/locales/es/settings.json
@@ -546,11 +546,11 @@
   "settings.privacy.encryptionUnlockButton": "Desbloquear",
   "settings.privacy.encryptionWarning": "Si olvidas tu contraseña, tus datos no podrán recuperarse.",
   "settings.privacy.encryptionWrongPassphrase": "Contraseña incorrecta — descifrado fallido",
+  "settings.privacy.euDataResidency": "Residencia de datos en la UE",
+  "settings.privacy.euDataResidencyHint": "Mantener todos los datos en servidores de la Unión Europea cuando sea posible.",
   "settings.privacy.localStorageOnly": "Solo almacenamiento local",
   "settings.privacy.migrationComplete": "Todos los datos cifrados correctamente",
   "settings.privacy.migrationProgress": "Cifrando datos existentes…",
-  "settings.privacy.euDataResidency": "Residencia de datos en la UE",
-  "settings.privacy.euDataResidencyHint": "Mantener todos los datos en servidores de la Unión Europea cuando sea posible.",
   "settings.privacy.shareUsageData": "Compartir datos de uso",
   "settings.privacy.title": "Privacidad y seguridad",
   "settings.projectAi.activeIndicator": "IA de proyecto activa",
@@ -627,6 +627,9 @@
   "settings.voice.downloadTtsModel": "Download TTS (Kokoro)",
   "settings.voice.enableHint": "Activate voice commands, dictation, and audio feedback.",
   "settings.voice.enableLabel": "Enable Voice Support",
+  "settings.voice.engine.auto": "Automático (recomendado)",
+  "settings.voice.engine.wasm": "Modelo WASM local (sin conexión)",
+  "settings.voice.engine.webSpeech": "Web Speech API (nube)",
   "settings.voice.feedbackLevel": "Audio Feedback Level",
   "settings.voice.intro": "Control StoryCraft Studio with your voice. All voice processing runs locally for maximum privacy.",
   "settings.voice.level.minimal": "Minimal — errors only",
@@ -641,12 +644,24 @@
   "settings.voice.privacy.consentDecline": "Rechazar",
   "settings.voice.privacy.consentTitle": "Consentimiento Web Speech",
   "settings.voice.privacy.revokeConsent": "Revocar consentimiento",
+  "settings.voice.privacy.revokeConsentHint": "Revoca tu consentimiento de Web Speech. La voz usará WASM local o se desactivará.",
   "settings.voice.privacy.statusExternal": "Externo",
   "settings.voice.privacy.statusLocal": "Local",
   "settings.voice.privacyNotice": "Voice data is processed locally on your device. No audio is sent to the cloud unless you explicitly enable cloud fallback below.",
+  "settings.voice.pttHint": "Atajo Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Mantener para grabar.",
+  "settings.voice.speechRate": "Velocidad de habla",
+  "settings.voice.speechRateHint": "Qué tan rápido habla la voz de síntesis. 1.0 = velocidad normal.",
+  "settings.voice.speechVolume": "Volumen de voz",
+  "settings.voice.speechVolumeHint": "Volumen de la voz de síntesis. No afecta la entrada del micrófono.",
+  "settings.voice.sttEngine": "Motor de reconocimiento de voz (STT)",
+  "settings.voice.sttEngineHint": "Elige el motor que convierte tu voz en texto. 'Auto' selecciona la mejor opción disponible.",
   "settings.voice.title": "Voice Full Support",
+  "settings.voice.ttsEngine": "Motor de síntesis de voz (TTS)",
+  "settings.voice.ttsEngineHint": "Elige la voz que lee las respuestas en voz alta. 'Auto' selecciona la mejor opción disponible.",
   "settings.voice.ttsMuted": "Mute All Speech Output",
   "settings.voice.ttsMutedHint": "Disable text-to-speech while keeping visual indicators.",
+  "settings.voice.wakeWordPhrase": "Frase de activación",
+  "settings.voice.wakeWordPhraseHint": "La frase que dices para activar los comandos de voz sin manos.",
   "settings.voice.wasmModels": "Modelos de voz WASM",
   "settings.voice.wasmModelsNotReady": "Modelos Whisper STT (~42 MB) y Kokoro TTS (~15 MB) no descargados",
   "settings.voice.wasmModelsReady": "Descargados y listos para procesamiento de voz local",
@@ -655,20 +670,5 @@
   "voice.modelDownload.error": "Descarga fallida: {{error}}",
   "voice.modelDownload.progress": "{{percent}}% completado",
   "voice.modelDownload.retry": "Reintentar",
-  "voice.modelDownload.title": "Descarga de modelo de voz",
-  "settings.voice.sttEngine": "Motor de reconocimiento de voz (STT)",
-  "settings.voice.sttEngineHint": "Elige el motor que convierte tu voz en texto. 'Auto' selecciona la mejor opción disponible.",
-  "settings.voice.ttsEngine": "Motor de síntesis de voz (TTS)",
-  "settings.voice.ttsEngineHint": "Elige la voz que lee las respuestas en voz alta. 'Auto' selecciona la mejor opción disponible.",
-  "settings.voice.engine.auto": "Automático (recomendado)",
-  "settings.voice.engine.webSpeech": "Web Speech API (nube)",
-  "settings.voice.engine.wasm": "Modelo WASM local (sin conexión)",
-  "settings.voice.speechRate": "Velocidad de habla",
-  "settings.voice.speechRateHint": "Qué tan rápido habla la voz de síntesis. 1.0 = velocidad normal.",
-  "settings.voice.speechVolume": "Volumen de voz",
-  "settings.voice.speechVolumeHint": "Volumen de la voz de síntesis. No afecta la entrada del micrófono.",
-  "settings.voice.wakeWordPhrase": "Frase de activación",
-  "settings.voice.wakeWordPhraseHint": "La frase que dices para activar los comandos de voz sin manos.",
-  "settings.voice.pttHint": "Atajo Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Mantener para grabar.",
-  "settings.voice.privacy.revokeConsentHint": "Revoca tu consentimiento de Web Speech. La voz usará WASM local o se desactivará."
+  "voice.modelDownload.title": "Descarga de modelo de voz"
 }
diff --git a/locales/fr/common.json b/locales/fr/common.json
index 77b5182c2..a34bb87b4 100644
--- a/locales/fr/common.json
+++ b/locales/fr/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Examen de la structure de votre histoire…",
   "proforge.pipeline.noneActive": "Aucune pipeline active",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "mots écrits en sessions",
   "progress.allTime.title": "Total cumulé",
   "progress.bestDay.title": "Meilleur jour",
diff --git a/locales/fr/settings.json b/locales/fr/settings.json
index 20fa914d8..2103135ff 100644
--- a/locales/fr/settings.json
+++ b/locales/fr/settings.json
@@ -546,11 +546,11 @@
   "settings.privacy.encryptionUnlockButton": "Déverrouiller",
   "settings.privacy.encryptionWarning": "Si vous oubliez votre phrase secrète, vos données ne pourront pas être récupérées.",
   "settings.privacy.encryptionWrongPassphrase": "Phrase secrète incorrecte — déchiffrement échoué",
+  "settings.privacy.euDataResidency": "Résidence des données UE",
+  "settings.privacy.euDataResidencyHint": "Conserver toutes les données sur des serveurs de l'Union européenne si possible.",
   "settings.privacy.localStorageOnly": "Stockage local uniquement",
   "settings.privacy.migrationComplete": "Toutes les données ont été chiffrées avec succès",
   "settings.privacy.migrationProgress": "Chiffrement des données existantes…",
-  "settings.privacy.euDataResidency": "Résidence des données UE",
-  "settings.privacy.euDataResidencyHint": "Conserver toutes les données sur des serveurs de l'Union européenne si possible.",
   "settings.privacy.shareUsageData": "Partager les données d'utilisation",
   "settings.privacy.title": "Confidentialité et sécurité",
   "settings.projectAi.activeIndicator": "IA projet active",
@@ -627,6 +627,9 @@
   "settings.voice.downloadTtsModel": "Download TTS (Kokoro)",
   "settings.voice.enableHint": "Activate voice commands, dictation, and audio feedback.",
   "settings.voice.enableLabel": "Enable Voice Support",
+  "settings.voice.engine.auto": "Automatique (recommandé)",
+  "settings.voice.engine.wasm": "Modèle WASM local (hors ligne)",
+  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
   "settings.voice.feedbackLevel": "Audio Feedback Level",
   "settings.voice.intro": "Control StoryCraft Studio with your voice. All voice processing runs locally for maximum privacy.",
   "settings.voice.level.minimal": "Minimal — errors only",
@@ -641,12 +644,24 @@
   "settings.voice.privacy.consentDecline": "Refuser",
   "settings.voice.privacy.consentTitle": "Consentement Web Speech",
   "settings.voice.privacy.revokeConsent": "Révoquer le consentement",
+  "settings.voice.privacy.revokeConsentHint": "Révoque votre consentement Web Speech. La voix reviendra au WASM local ou sera désactivée.",
   "settings.voice.privacy.statusExternal": "Externe",
   "settings.voice.privacy.statusLocal": "Local",
   "settings.voice.privacyNotice": "Voice data is processed locally on your device. No audio is sent to the cloud unless you explicitly enable cloud fallback below.",
+  "settings.voice.pttHint": "Raccourci Push-to-Talk : Ctrl+Maj+V (Windows/Linux) ou ⌘+Maj+V (Mac). Maintenir pour enregistrer.",
+  "settings.voice.speechRate": "Vitesse d'élocution",
+  "settings.voice.speechRateHint": "Vitesse à laquelle la voix de synthèse vocale parle. 1,0 = vitesse normale.",
+  "settings.voice.speechVolume": "Volume de la parole",
+  "settings.voice.speechVolumeHint": "Volume de la voix de synthèse vocale. N'affecte pas l'entrée du microphone.",
+  "settings.voice.sttEngine": "Moteur de reconnaissance vocale (STT)",
+  "settings.voice.sttEngineHint": "Choisissez le moteur qui convertit votre parole en texte. 'Auto' choisit la meilleure option disponible.",
   "settings.voice.title": "Voice Full Support",
+  "settings.voice.ttsEngine": "Moteur de synthèse vocale (TTS)",
+  "settings.voice.ttsEngineHint": "Choisissez la voix qui lit les réponses à voix haute. 'Auto' choisit la meilleure option disponible.",
   "settings.voice.ttsMuted": "Mute All Speech Output",
   "settings.voice.ttsMutedHint": "Disable text-to-speech while keeping visual indicators.",
+  "settings.voice.wakeWordPhrase": "Phrase d'activation",
+  "settings.voice.wakeWordPhraseHint": "La phrase que vous dites pour activer les commandes vocales en mains libres.",
   "settings.voice.wasmModels": "Modèles vocaux WASM",
   "settings.voice.wasmModelsNotReady": "Modèles Whisper STT (~42 Mo) et Kokoro TTS (~15 Mo) non téléchargés",
   "settings.voice.wasmModelsReady": "Téléchargés et prêts pour le traitement vocal local",
@@ -655,20 +670,5 @@
   "voice.modelDownload.error": "Échec du téléchargement : {{error}}",
   "voice.modelDownload.progress": "{{percent}}% terminé",
   "voice.modelDownload.retry": "Réessayer",
-  "voice.modelDownload.title": "Téléchargement du modèle vocal",
-  "settings.voice.sttEngine": "Moteur de reconnaissance vocale (STT)",
-  "settings.voice.sttEngineHint": "Choisissez le moteur qui convertit votre parole en texte. 'Auto' choisit la meilleure option disponible.",
-  "settings.voice.ttsEngine": "Moteur de synthèse vocale (TTS)",
-  "settings.voice.ttsEngineHint": "Choisissez la voix qui lit les réponses à voix haute. 'Auto' choisit la meilleure option disponible.",
-  "settings.voice.engine.auto": "Automatique (recommandé)",
-  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
-  "settings.voice.engine.wasm": "Modèle WASM local (hors ligne)",
-  "settings.voice.speechRate": "Vitesse d'élocution",
-  "settings.voice.speechRateHint": "Vitesse à laquelle la voix de synthèse vocale parle. 1,0 = vitesse normale.",
-  "settings.voice.speechVolume": "Volume de la parole",
-  "settings.voice.speechVolumeHint": "Volume de la voix de synthèse vocale. N'affecte pas l'entrée du microphone.",
-  "settings.voice.wakeWordPhrase": "Phrase d'activation",
-  "settings.voice.wakeWordPhraseHint": "La phrase que vous dites pour activer les commandes vocales en mains libres.",
-  "settings.voice.pttHint": "Raccourci Push-to-Talk : Ctrl+Maj+V (Windows/Linux) ou ⌘+Maj+V (Mac). Maintenir pour enregistrer.",
-  "settings.voice.privacy.revokeConsentHint": "Révoque votre consentement Web Speech. La voix reviendra au WASM local ou sera désactivée."
+  "voice.modelDownload.title": "Téléchargement du modèle vocal"
 }
diff --git a/locales/he/common.json b/locales/he/common.json
index 35ed21f3a..2e0ed54f2 100644
--- a/locales/he/common.json
+++ b/locales/he/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "מתבונן במבנה הסיפור שלכם…",
   "proforge.pipeline.noneActive": "אין צינור פעיל",
   "proforge.pipeline.title": "צינור הסופר האולטימטיבי",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "מילים שנכתבו בהפעלות",
   "progress.allTime.title": "כל הזמן",
   "progress.bestDay.title": "היום הטוב ביותר",
diff --git a/locales/it/common.json b/locales/it/common.json
index 3059369eb..b8ce9d005 100644
--- a/locales/it/common.json
+++ b/locales/it/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Osservando la forma della tua storia…",
   "proforge.pipeline.noneActive": "Nessuna pipeline attiva",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "parole scritte nelle sessioni",
   "progress.allTime.title": "Totale storico",
   "progress.bestDay.title": "Giorno migliore",
diff --git a/locales/it/settings.json b/locales/it/settings.json
index 9ff182243..404442bab 100644
--- a/locales/it/settings.json
+++ b/locales/it/settings.json
@@ -546,11 +546,11 @@
   "settings.privacy.encryptionUnlockButton": "Sblocca",
   "settings.privacy.encryptionWarning": "Se dimentichi la passphrase, i tuoi dati non potranno essere recuperati.",
   "settings.privacy.encryptionWrongPassphrase": "Passphrase errata — decifratura non riuscita",
+  "settings.privacy.euDataResidency": "Residenza dati UE",
+  "settings.privacy.euDataResidencyHint": "Mantenere tutti i dati su server dell'Unione Europea ove possibile.",
   "settings.privacy.localStorageOnly": "Solo archiviazione locale",
   "settings.privacy.migrationComplete": "Tutti i dati cifrati con successo",
   "settings.privacy.migrationProgress": "Cifratura dei dati esistenti in corso…",
-  "settings.privacy.euDataResidency": "Residenza dati UE",
-  "settings.privacy.euDataResidencyHint": "Mantenere tutti i dati su server dell'Unione Europea ove possibile.",
   "settings.privacy.shareUsageData": "Condividi dati di utilizzo",
   "settings.privacy.title": "Privacy e sicurezza",
   "settings.projectAi.activeIndicator": "IA progetto attiva",
@@ -627,6 +627,9 @@
   "settings.voice.downloadTtsModel": "Download TTS (Kokoro)",
   "settings.voice.enableHint": "Attiva comandi vocali, dettatura e feedback audio.",
   "settings.voice.enableLabel": "Abilita controllo vocale",
+  "settings.voice.engine.auto": "Automatico (consigliato)",
+  "settings.voice.engine.wasm": "Modello WASM locale (offline)",
+  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
   "settings.voice.feedbackLevel": "Livello di feedback audio",
   "settings.voice.intro": "Controlla StoryCraft Studio con la tua voce. Tutto l’elaborazione vocale avviene localmente per la massima privacy.",
   "settings.voice.level.minimal": "Minimo — solo errori",
@@ -641,12 +644,24 @@
   "settings.voice.privacy.consentDecline": "Rifiuta",
   "settings.voice.privacy.consentTitle": "Consenso Web Speech",
   "settings.voice.privacy.revokeConsent": "Revoca il consenso",
+  "settings.voice.privacy.revokeConsentHint": "Revoca il tuo consenso Web Speech. La voce passerà al WASM locale o verrà disattivata.",
   "settings.voice.privacy.statusExternal": "Esterno",
   "settings.voice.privacy.statusLocal": "Locale",
   "settings.voice.privacyNotice": "I dati vocali vengono elaborati localmente sul tuo dispositivo. Nessun audio viene inviato al cloud a meno che tu non abiliti esplicitamente il fallback cloud.",
+  "settings.voice.pttHint": "Scorciatoia Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Tenere premuto per registrare.",
+  "settings.voice.speechRate": "Velocità del parlato",
+  "settings.voice.speechRateHint": "Quanto velocemente parla la voce di sintesi. 1.0 = velocità normale.",
+  "settings.voice.speechVolume": "Volume del parlato",
+  "settings.voice.speechVolumeHint": "Volume della voce di sintesi. Non influisce sull'ingresso del microfono.",
+  "settings.voice.sttEngine": "Motore di riconoscimento vocale (STT)",
+  "settings.voice.sttEngineHint": "Scegli il motore che converte il tuo parlato in testo. 'Auto' sceglie la migliore opzione disponibile.",
   "settings.voice.title": "Controllo vocale",
+  "settings.voice.ttsEngine": "Motore di sintesi vocale (TTS)",
+  "settings.voice.ttsEngineHint": "Scegli la voce che legge le risposte ad alta voce. 'Auto' sceglie la migliore opzione disponibile.",
   "settings.voice.ttsMuted": "Silenzia tutta l’uscita vocale",
   "settings.voice.ttsMutedHint": "Disabilita il text-to-speech mantenendo gli indicatori visivi.",
+  "settings.voice.wakeWordPhrase": "Frase di attivazione",
+  "settings.voice.wakeWordPhraseHint": "La frase che dici per attivare i comandi vocali a mani libere.",
   "settings.voice.wasmModels": "Modelli vocali WASM",
   "settings.voice.wasmModelsNotReady": "Modelli Whisper STT (~42 MB) e Kokoro TTS (~15 MB) non scaricati",
   "settings.voice.wasmModelsReady": "Scaricati e pronti per il riconoscimento vocale locale",
@@ -655,20 +670,5 @@
   "voice.modelDownload.error": "Download fallito: {{error}}",
   "voice.modelDownload.progress": "{{percent}}% completato",
   "voice.modelDownload.retry": "Riprova",
-  "voice.modelDownload.title": "Download modello vocale",
-  "settings.voice.sttEngine": "Motore di riconoscimento vocale (STT)",
-  "settings.voice.sttEngineHint": "Scegli il motore che converte il tuo parlato in testo. 'Auto' sceglie la migliore opzione disponibile.",
-  "settings.voice.ttsEngine": "Motore di sintesi vocale (TTS)",
-  "settings.voice.ttsEngineHint": "Scegli la voce che legge le risposte ad alta voce. 'Auto' sceglie la migliore opzione disponibile.",
-  "settings.voice.engine.auto": "Automatico (consigliato)",
-  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
-  "settings.voice.engine.wasm": "Modello WASM locale (offline)",
-  "settings.voice.speechRate": "Velocità del parlato",
-  "settings.voice.speechRateHint": "Quanto velocemente parla la voce di sintesi. 1.0 = velocità normale.",
-  "settings.voice.speechVolume": "Volume del parlato",
-  "settings.voice.speechVolumeHint": "Volume della voce di sintesi. Non influisce sull'ingresso del microfono.",
-  "settings.voice.wakeWordPhrase": "Frase di attivazione",
-  "settings.voice.wakeWordPhraseHint": "La frase che dici per attivare i comandi vocali a mani libere.",
-  "settings.voice.pttHint": "Scorciatoia Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Tenere premuto per registrare.",
-  "settings.voice.privacy.revokeConsentHint": "Revoca il tuo consenso Web Speech. La voce passerà al WASM locale o verrà disattivata."
+  "voice.modelDownload.title": "Download modello vocale"
 }
diff --git a/locales/ja/common.json b/locales/ja/common.json
index e37de140e..92aed1cb3 100644
--- a/locales/ja/common.json
+++ b/locales/ja/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "あなたの物語の形を見てみると…",
   "proforge.pipeline.noneActive": "アクティブなパイプラインがありません",
   "proforge.pipeline.title": "究極の著者パイプライン",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "セッションで書いた言葉",
   "progress.allTime.title": "オールタイム",
   "progress.bestDay.title": "最高の一日",
diff --git a/locales/pt/common.json b/locales/pt/common.json
index 0a288b7df..4477019ad 100644
--- a/locales/pt/common.json
+++ b/locales/pt/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Olhando para a forma da sua história…",
   "proforge.pipeline.noneActive": "Nenhum pipeline ativo",
   "proforge.pipeline.title": "Pipeline final do autor",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "palavras escritas em sessões",
   "progress.allTime.title": "Todos os tempos",
   "progress.bestDay.title": "Melhor dia",
diff --git a/locales/zh/common.json b/locales/zh/common.json
index 615248862..159faf13f 100644
--- a/locales/zh/common.json
+++ b/locales/zh/common.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "看看你的故事的形状……",
   "proforge.pipeline.noneActive": "无活动管道",
   "proforge.pipeline.title": "最终作者管道",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "会话中写下的单词",
   "progress.allTime.title": "所有时间",
   "progress.bestDay.title": "最好的一天",
diff --git a/public/locales/ar/bundle.json b/public/locales/ar/bundle.json
index 54f1f94c5..f8459c1aa 100644
--- a/public/locales/ar/bundle.json
+++ b/public/locales/ar/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "جارٍ النظر في شكل قصتك…",
   "proforge.pipeline.noneActive": "لا يوجد خط أنابيب نشط",
   "proforge.pipeline.title": "خط أنابيب المؤلّف الأمثل",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "كلمات مكتوبة في الجلسات",
   "progress.allTime.title": "كل الأوقات",
   "progress.bestDay.title": "أفضل يوم",
diff --git a/public/locales/de/bundle.json b/public/locales/de/bundle.json
index 94d5f1c18..92ca202f1 100644
--- a/public/locales/de/bundle.json
+++ b/public/locales/de/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Betrachtet die Struktur deiner Geschichte…",
   "proforge.pipeline.noneActive": "Keine aktive Pipeline",
   "proforge.pipeline.title": "Ultimative Autoren-Pipeline",
+  "proforge.stageName.idle": "Leerlauf",
+  "proforge.stageName.intake": "Aufnahme & Diagnose",
+  "proforge.stageName.structural": "Strukturlektorat",
+  "proforge.stageName.lineProse": "Sprach- & Prosalektorat",
+  "proforge.stageName.copyEdit": "Schlusslektorat",
+  "proforge.stageName.proof": "Korrektorat",
+  "proforge.stageName.production": "Produktion",
+  "proforge.stageName.publishing": "Veröffentlichung",
+  "proforge.stageName.analytics": "Auswertung",
+  "proforge.stageName.archived": "Archiviert",
+  "proforge.review.heading": "Prüfung: {{stage}}",
+  "proforge.review.counts": "{{pending}} offen · {{accepted}} übernommen · {{rejected}} abgelehnt",
+  "proforge.review.quickAccept": "Schnell übernehmen ({{count}})",
+  "proforge.review.quickAcceptTitle": "Alle unkritischen Vorschläge mit hoher Konfidenz übernehmen",
+  "proforge.review.acceptAll": "Alle übernehmen",
+  "proforge.review.rejectAll": "Alle ablehnen",
+  "proforge.review.criticalHeadingOne": "{{count}} kritisches Problem erfordert Aufmerksamkeit",
+  "proforge.review.criticalHeadingOther": "{{count}} kritische Probleme erfordern Aufmerksamkeit",
+  "proforge.review.acceptAllCritical": "Alle kritischen übernehmen",
+  "proforge.review.moreCritical": "+{{count}} weitere kritische Probleme unten",
+  "proforge.review.filter.all": "Alle",
+  "proforge.review.filter.pending": "Offen",
+  "proforge.review.filter.accepted": "Übernommen",
+  "proforge.review.filter.rejected": "Abgelehnt",
+  "proforge.review.noItems": "Keine Prüfpunkte verfügbar.",
+  "proforge.review.noItemsForFilter": "Keine Einträge für diesen Filter.",
+  "proforge.review.skipStage": "Phase überspringen",
+  "proforge.review.submitContinue": "Übernehmen & fortfahren",
+  "proforge.review.pendingCount": "{{count}} offen",
+  "proforge.review.confidence": "{{percent}} % Konfidenz",
+  "proforge.review.original": "Original:",
+  "proforge.review.proposed": "Vorschlag:",
+  "proforge.review.showDetails": "Details anzeigen",
+  "proforge.review.showLess": "Weniger anzeigen",
+  "proforge.review.accept": "Übernehmen",
+  "proforge.review.reject": "Ablehnen",
+  "proforge.review.ignore": "Ignorieren",
+  "proforge.review.group.critical": "Kritische Maßnahmen",
+  "proforge.review.group.warning": "Warnungen",
+  "proforge.review.group.info": "Vorschläge",
+  "proforge.review.severity.critical": "Kritisch",
+  "proforge.review.severity.warning": "Warnung",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.announceSubmitted": "Prüfung {{stage}} übermittelt",
+  "proforge.review.type.structuralEdit": "Struktur",
+  "proforge.review.type.proseEdit": "Prosa",
+  "proforge.review.type.grammarEdit": "Grammatik",
+  "proforge.review.type.styleEdit": "Stil",
+  "proforge.review.type.repetitionHit": "Wiederholung",
+  "proforge.review.type.consistencyIssue": "Konsistenz",
+  "proforge.review.type.plotHole": "Logiklücke",
+  "proforge.review.type.legalWarning": "Rechtliches",
+  "proforge.review.type.technicalIssue": "Technik",
+  "proforge.review.type.pacingIssue": "Tempo",
+  "proforge.review.type.arcIssue": "Spannungsbogen",
   "progress.allTime.label": "in Sitzungen geschriebene Wörter",
   "progress.allTime.title": "Gesamt",
   "progress.bestDay.title": "Bester Tag",
diff --git a/public/locales/el/bundle.json b/public/locales/el/bundle.json
index e68d5e6bb..cd6624588 100644
--- a/public/locales/el/bundle.json
+++ b/public/locales/el/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Κοιτάζοντας τη μορφή της ιστορίας σας…",
   "proforge.pipeline.noneActive": "Δεν υπάρχει ενεργός αγωγός",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "λέξεις γραμμένες σε συνεδρίες",
   "progress.allTime.title": "Όλων των Εποχών",
   "progress.bestDay.title": "Καλύτερη μέρα",
diff --git a/public/locales/en/bundle.json b/public/locales/en/bundle.json
index 86443951a..b7ea93d5d 100644
--- a/public/locales/en/bundle.json
+++ b/public/locales/en/bundle.json
@@ -491,6 +491,61 @@
   "proforge.emptyState.title": "Your manuscript, refined.",
   "proforge.emptyState.description": "ProForge reads every chapter, then walks you through editing stage by stage. Nothing gets lost — every change is reversible.",
   "proforge.enabledHint": "ProForge Pipeline enabled. Open the Writer view and click the ProForge button in the tools panel to start.",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.structural": "Structural Edit",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.original": "Original:",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.accept": "Accept",
+  "proforge.review.reject": "Reject",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.arcIssue": "Arc",
   "sceneboard.emptyState.title": "Your plot board is a blank canvas",
   "sceneboard.emptyState.description": "Add your first scene or import your outline to start planning your story.",
   "characters.emptyState.title": "Your cast list is empty",
diff --git a/public/locales/es/bundle.json b/public/locales/es/bundle.json
index 55810ea93..aa73bf86a 100644
--- a/public/locales/es/bundle.json
+++ b/public/locales/es/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Observando la forma de tu historia…",
   "proforge.pipeline.noneActive": "Sin pipeline activa",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "palabras escritas en sesiones",
   "progress.allTime.title": "Total histórico",
   "progress.bestDay.title": "Mejor día",
@@ -1865,11 +1920,11 @@
   "settings.privacy.encryptionUnlockButton": "Desbloquear",
   "settings.privacy.encryptionWarning": "Si olvidas tu contraseña, tus datos no podrán recuperarse.",
   "settings.privacy.encryptionWrongPassphrase": "Contraseña incorrecta — descifrado fallido",
+  "settings.privacy.euDataResidency": "Residencia de datos en la UE",
+  "settings.privacy.euDataResidencyHint": "Mantener todos los datos en servidores de la Unión Europea cuando sea posible.",
   "settings.privacy.localStorageOnly": "Solo almacenamiento local",
   "settings.privacy.migrationComplete": "Todos los datos cifrados correctamente",
   "settings.privacy.migrationProgress": "Cifrando datos existentes…",
-  "settings.privacy.euDataResidency": "Residencia de datos en la UE",
-  "settings.privacy.euDataResidencyHint": "Mantener todos los datos en servidores de la Unión Europea cuando sea posible.",
   "settings.privacy.shareUsageData": "Compartir datos de uso",
   "settings.privacy.title": "Privacidad y seguridad",
   "settings.projectAi.activeIndicator": "IA de proyecto activa",
@@ -1946,6 +2001,9 @@
   "settings.voice.downloadTtsModel": "Download TTS (Kokoro)",
   "settings.voice.enableHint": "Activate voice commands, dictation, and audio feedback.",
   "settings.voice.enableLabel": "Enable Voice Support",
+  "settings.voice.engine.auto": "Automático (recomendado)",
+  "settings.voice.engine.wasm": "Modelo WASM local (sin conexión)",
+  "settings.voice.engine.webSpeech": "Web Speech API (nube)",
   "settings.voice.feedbackLevel": "Audio Feedback Level",
   "settings.voice.intro": "Control StoryCraft Studio with your voice. All voice processing runs locally for maximum privacy.",
   "settings.voice.level.minimal": "Minimal — errors only",
@@ -1960,12 +2018,24 @@
   "settings.voice.privacy.consentDecline": "Rechazar",
   "settings.voice.privacy.consentTitle": "Consentimiento Web Speech",
   "settings.voice.privacy.revokeConsent": "Revocar consentimiento",
+  "settings.voice.privacy.revokeConsentHint": "Revoca tu consentimiento de Web Speech. La voz usará WASM local o se desactivará.",
   "settings.voice.privacy.statusExternal": "Externo",
   "settings.voice.privacy.statusLocal": "Local",
   "settings.voice.privacyNotice": "Voice data is processed locally on your device. No audio is sent to the cloud unless you explicitly enable cloud fallback below.",
+  "settings.voice.pttHint": "Atajo Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Mantener para grabar.",
+  "settings.voice.speechRate": "Velocidad de habla",
+  "settings.voice.speechRateHint": "Qué tan rápido habla la voz de síntesis. 1.0 = velocidad normal.",
+  "settings.voice.speechVolume": "Volumen de voz",
+  "settings.voice.speechVolumeHint": "Volumen de la voz de síntesis. No afecta la entrada del micrófono.",
+  "settings.voice.sttEngine": "Motor de reconocimiento de voz (STT)",
+  "settings.voice.sttEngineHint": "Elige el motor que convierte tu voz en texto. 'Auto' selecciona la mejor opción disponible.",
   "settings.voice.title": "Voice Full Support",
+  "settings.voice.ttsEngine": "Motor de síntesis de voz (TTS)",
+  "settings.voice.ttsEngineHint": "Elige la voz que lee las respuestas en voz alta. 'Auto' selecciona la mejor opción disponible.",
   "settings.voice.ttsMuted": "Mute All Speech Output",
   "settings.voice.ttsMutedHint": "Disable text-to-speech while keeping visual indicators.",
+  "settings.voice.wakeWordPhrase": "Frase de activación",
+  "settings.voice.wakeWordPhraseHint": "La frase que dices para activar los comandos de voz sin manos.",
   "settings.voice.wasmModels": "Modelos de voz WASM",
   "settings.voice.wasmModelsNotReady": "Modelos Whisper STT (~42 MB) y Kokoro TTS (~15 MB) no descargados",
   "settings.voice.wasmModelsReady": "Descargados y listos para procesamiento de voz local",
@@ -1975,21 +2045,6 @@
   "voice.modelDownload.progress": "{{percent}}% completado",
   "voice.modelDownload.retry": "Reintentar",
   "voice.modelDownload.title": "Descarga de modelo de voz",
-  "settings.voice.sttEngine": "Motor de reconocimiento de voz (STT)",
-  "settings.voice.sttEngineHint": "Elige el motor que convierte tu voz en texto. 'Auto' selecciona la mejor opción disponible.",
-  "settings.voice.ttsEngine": "Motor de síntesis de voz (TTS)",
-  "settings.voice.ttsEngineHint": "Elige la voz que lee las respuestas en voz alta. 'Auto' selecciona la mejor opción disponible.",
-  "settings.voice.engine.auto": "Automático (recomendado)",
-  "settings.voice.engine.webSpeech": "Web Speech API (nube)",
-  "settings.voice.engine.wasm": "Modelo WASM local (sin conexión)",
-  "settings.voice.speechRate": "Velocidad de habla",
-  "settings.voice.speechRateHint": "Qué tan rápido habla la voz de síntesis. 1.0 = velocidad normal.",
-  "settings.voice.speechVolume": "Volumen de voz",
-  "settings.voice.speechVolumeHint": "Volumen de la voz de síntesis. No afecta la entrada del micrófono.",
-  "settings.voice.wakeWordPhrase": "Frase de activación",
-  "settings.voice.wakeWordPhraseHint": "La frase que dices para activar los comandos de voz sin manos.",
-  "settings.voice.pttHint": "Atajo Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Mantener para grabar.",
-  "settings.voice.privacy.revokeConsentHint": "Revoca tu consentimiento de Web Speech. La voz usará WASM local o se desactivará.",
   "help.advanced.adaptiveAi.content": "El motor de IA adaptativa detecta tu hardware en tiempo de ejecución —WebGPU, WebNN, DirectML, núcleos de CPU, nivel de VRAM y batería— y enruta la inferencia local al backend más rápido disponible. Actívalo en Ajustes → Funciones de acceso anticipado y consulta la lectura en vivo en Ajustes → IA. El modo Eco elige el modelo más pequeño viable para ahorrar batería y memoria, y una elección de pestaña líder garantiza que solo una pestaña del navegador use la GPU a la vez para evitar colisiones de VRAM.",
   "help.advanced.adaptiveAi.title": "IA adaptativa, GPU y modo Eco",
   "help.advanced.cloudSync.content": "La sincronización opcional cifrada de extremo a extremo mantiene tu biblioteca al día en todos tus dispositivos mediante un backend de Cloudflare R2. Activa «Sincronización en la nube» en Ajustes → Funciones de acceso anticipado y configúrala en Ajustes → Conexiones. Los datos del proyecto se cifran localmente con AES-256-GCM antes de subirse, y tus claves de API de IA nunca se envían a la nube. La sincronización es totalmente opcional; sin ella, StoryCraft sigue siendo una app totalmente «offline-first».",
diff --git a/public/locales/fr/bundle.json b/public/locales/fr/bundle.json
index c3cb47a1b..3f5257e7f 100644
--- a/public/locales/fr/bundle.json
+++ b/public/locales/fr/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Examen de la structure de votre histoire…",
   "proforge.pipeline.noneActive": "Aucune pipeline active",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "mots écrits en sessions",
   "progress.allTime.title": "Total cumulé",
   "progress.bestDay.title": "Meilleur jour",
@@ -1865,11 +1920,11 @@
   "settings.privacy.encryptionUnlockButton": "Déverrouiller",
   "settings.privacy.encryptionWarning": "Si vous oubliez votre phrase secrète, vos données ne pourront pas être récupérées.",
   "settings.privacy.encryptionWrongPassphrase": "Phrase secrète incorrecte — déchiffrement échoué",
+  "settings.privacy.euDataResidency": "Résidence des données UE",
+  "settings.privacy.euDataResidencyHint": "Conserver toutes les données sur des serveurs de l'Union européenne si possible.",
   "settings.privacy.localStorageOnly": "Stockage local uniquement",
   "settings.privacy.migrationComplete": "Toutes les données ont été chiffrées avec succès",
   "settings.privacy.migrationProgress": "Chiffrement des données existantes…",
-  "settings.privacy.euDataResidency": "Résidence des données UE",
-  "settings.privacy.euDataResidencyHint": "Conserver toutes les données sur des serveurs de l'Union européenne si possible.",
   "settings.privacy.shareUsageData": "Partager les données d'utilisation",
   "settings.privacy.title": "Confidentialité et sécurité",
   "settings.projectAi.activeIndicator": "IA projet active",
@@ -1946,6 +2001,9 @@
   "settings.voice.downloadTtsModel": "Download TTS (Kokoro)",
   "settings.voice.enableHint": "Activate voice commands, dictation, and audio feedback.",
   "settings.voice.enableLabel": "Enable Voice Support",
+  "settings.voice.engine.auto": "Automatique (recommandé)",
+  "settings.voice.engine.wasm": "Modèle WASM local (hors ligne)",
+  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
   "settings.voice.feedbackLevel": "Audio Feedback Level",
   "settings.voice.intro": "Control StoryCraft Studio with your voice. All voice processing runs locally for maximum privacy.",
   "settings.voice.level.minimal": "Minimal — errors only",
@@ -1960,12 +2018,24 @@
   "settings.voice.privacy.consentDecline": "Refuser",
   "settings.voice.privacy.consentTitle": "Consentement Web Speech",
   "settings.voice.privacy.revokeConsent": "Révoquer le consentement",
+  "settings.voice.privacy.revokeConsentHint": "Révoque votre consentement Web Speech. La voix reviendra au WASM local ou sera désactivée.",
   "settings.voice.privacy.statusExternal": "Externe",
   "settings.voice.privacy.statusLocal": "Local",
   "settings.voice.privacyNotice": "Voice data is processed locally on your device. No audio is sent to the cloud unless you explicitly enable cloud fallback below.",
+  "settings.voice.pttHint": "Raccourci Push-to-Talk : Ctrl+Maj+V (Windows/Linux) ou ⌘+Maj+V (Mac). Maintenir pour enregistrer.",
+  "settings.voice.speechRate": "Vitesse d'élocution",
+  "settings.voice.speechRateHint": "Vitesse à laquelle la voix de synthèse vocale parle. 1,0 = vitesse normale.",
+  "settings.voice.speechVolume": "Volume de la parole",
+  "settings.voice.speechVolumeHint": "Volume de la voix de synthèse vocale. N'affecte pas l'entrée du microphone.",
+  "settings.voice.sttEngine": "Moteur de reconnaissance vocale (STT)",
+  "settings.voice.sttEngineHint": "Choisissez le moteur qui convertit votre parole en texte. 'Auto' choisit la meilleure option disponible.",
   "settings.voice.title": "Voice Full Support",
+  "settings.voice.ttsEngine": "Moteur de synthèse vocale (TTS)",
+  "settings.voice.ttsEngineHint": "Choisissez la voix qui lit les réponses à voix haute. 'Auto' choisit la meilleure option disponible.",
   "settings.voice.ttsMuted": "Mute All Speech Output",
   "settings.voice.ttsMutedHint": "Disable text-to-speech while keeping visual indicators.",
+  "settings.voice.wakeWordPhrase": "Phrase d'activation",
+  "settings.voice.wakeWordPhraseHint": "La phrase que vous dites pour activer les commandes vocales en mains libres.",
   "settings.voice.wasmModels": "Modèles vocaux WASM",
   "settings.voice.wasmModelsNotReady": "Modèles Whisper STT (~42 Mo) et Kokoro TTS (~15 Mo) non téléchargés",
   "settings.voice.wasmModelsReady": "Téléchargés et prêts pour le traitement vocal local",
@@ -1975,21 +2045,6 @@
   "voice.modelDownload.progress": "{{percent}}% terminé",
   "voice.modelDownload.retry": "Réessayer",
   "voice.modelDownload.title": "Téléchargement du modèle vocal",
-  "settings.voice.sttEngine": "Moteur de reconnaissance vocale (STT)",
-  "settings.voice.sttEngineHint": "Choisissez le moteur qui convertit votre parole en texte. 'Auto' choisit la meilleure option disponible.",
-  "settings.voice.ttsEngine": "Moteur de synthèse vocale (TTS)",
-  "settings.voice.ttsEngineHint": "Choisissez la voix qui lit les réponses à voix haute. 'Auto' choisit la meilleure option disponible.",
-  "settings.voice.engine.auto": "Automatique (recommandé)",
-  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
-  "settings.voice.engine.wasm": "Modèle WASM local (hors ligne)",
-  "settings.voice.speechRate": "Vitesse d'élocution",
-  "settings.voice.speechRateHint": "Vitesse à laquelle la voix de synthèse vocale parle. 1,0 = vitesse normale.",
-  "settings.voice.speechVolume": "Volume de la parole",
-  "settings.voice.speechVolumeHint": "Volume de la voix de synthèse vocale. N'affecte pas l'entrée du microphone.",
-  "settings.voice.wakeWordPhrase": "Phrase d'activation",
-  "settings.voice.wakeWordPhraseHint": "La phrase que vous dites pour activer les commandes vocales en mains libres.",
-  "settings.voice.pttHint": "Raccourci Push-to-Talk : Ctrl+Maj+V (Windows/Linux) ou ⌘+Maj+V (Mac). Maintenir pour enregistrer.",
-  "settings.voice.privacy.revokeConsentHint": "Révoque votre consentement Web Speech. La voix reviendra au WASM local ou sera désactivée.",
   "help.advanced.adaptiveAi.content": "Le moteur d’IA adaptatif détecte votre matériel à l’exécution — WebGPU, WebNN, DirectML, cœurs CPU, niveau de VRAM et batterie — et achemine l’inférence locale vers le backend le plus rapide disponible. Activez-le dans Paramètres → Fonctionnalités en accès anticipé et consultez l’affichage en direct dans Paramètres → IA. Le mode Éco choisit le plus petit modèle viable pour économiser batterie et mémoire, et une élection de l’onglet meneur garantit qu’un seul onglet du navigateur utilise le GPU à la fois pour éviter les collisions de VRAM.",
   "help.advanced.adaptiveAi.title": "IA adaptative, GPU et mode Éco",
   "help.advanced.cloudSync.content": "La synchronisation chiffrée de bout en bout, optionnelle, maintient votre bibliothèque à jour sur tous vos appareils via un backend Cloudflare R2. Activez « Synchronisation cloud » dans Paramètres → Fonctionnalités en accès anticipé et configurez-la dans Paramètres → Connexions. Les données du projet sont chiffrées localement en AES-256-GCM avant l’envoi, et vos clés d’API IA ne sont jamais transmises au cloud. La synchronisation est entièrement optionnelle ; sans elle, StoryCraft reste une application entièrement « offline-first ».",
diff --git a/public/locales/he/bundle.json b/public/locales/he/bundle.json
index 0b9efad7b..ef89a5509 100644
--- a/public/locales/he/bundle.json
+++ b/public/locales/he/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "מתבונן במבנה הסיפור שלכם…",
   "proforge.pipeline.noneActive": "אין צינור פעיל",
   "proforge.pipeline.title": "צינור הסופר האולטימטיבי",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "מילים שנכתבו בהפעלות",
   "progress.allTime.title": "כל הזמן",
   "progress.bestDay.title": "היום הטוב ביותר",
diff --git a/public/locales/it/bundle.json b/public/locales/it/bundle.json
index 982073abe..3c9130e2d 100644
--- a/public/locales/it/bundle.json
+++ b/public/locales/it/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Osservando la forma della tua storia…",
   "proforge.pipeline.noneActive": "Nessuna pipeline attiva",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "parole scritte nelle sessioni",
   "progress.allTime.title": "Totale storico",
   "progress.bestDay.title": "Giorno migliore",
@@ -1865,11 +1920,11 @@
   "settings.privacy.encryptionUnlockButton": "Sblocca",
   "settings.privacy.encryptionWarning": "Se dimentichi la passphrase, i tuoi dati non potranno essere recuperati.",
   "settings.privacy.encryptionWrongPassphrase": "Passphrase errata — decifratura non riuscita",
+  "settings.privacy.euDataResidency": "Residenza dati UE",
+  "settings.privacy.euDataResidencyHint": "Mantenere tutti i dati su server dell'Unione Europea ove possibile.",
   "settings.privacy.localStorageOnly": "Solo archiviazione locale",
   "settings.privacy.migrationComplete": "Tutti i dati cifrati con successo",
   "settings.privacy.migrationProgress": "Cifratura dei dati esistenti in corso…",
-  "settings.privacy.euDataResidency": "Residenza dati UE",
-  "settings.privacy.euDataResidencyHint": "Mantenere tutti i dati su server dell'Unione Europea ove possibile.",
   "settings.privacy.shareUsageData": "Condividi dati di utilizzo",
   "settings.privacy.title": "Privacy e sicurezza",
   "settings.projectAi.activeIndicator": "IA progetto attiva",
@@ -1946,6 +2001,9 @@
   "settings.voice.downloadTtsModel": "Download TTS (Kokoro)",
   "settings.voice.enableHint": "Attiva comandi vocali, dettatura e feedback audio.",
   "settings.voice.enableLabel": "Abilita controllo vocale",
+  "settings.voice.engine.auto": "Automatico (consigliato)",
+  "settings.voice.engine.wasm": "Modello WASM locale (offline)",
+  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
   "settings.voice.feedbackLevel": "Livello di feedback audio",
   "settings.voice.intro": "Controlla StoryCraft Studio con la tua voce. Tutto l’elaborazione vocale avviene localmente per la massima privacy.",
   "settings.voice.level.minimal": "Minimo — solo errori",
@@ -1960,12 +2018,24 @@
   "settings.voice.privacy.consentDecline": "Rifiuta",
   "settings.voice.privacy.consentTitle": "Consenso Web Speech",
   "settings.voice.privacy.revokeConsent": "Revoca il consenso",
+  "settings.voice.privacy.revokeConsentHint": "Revoca il tuo consenso Web Speech. La voce passerà al WASM locale o verrà disattivata.",
   "settings.voice.privacy.statusExternal": "Esterno",
   "settings.voice.privacy.statusLocal": "Locale",
   "settings.voice.privacyNotice": "I dati vocali vengono elaborati localmente sul tuo dispositivo. Nessun audio viene inviato al cloud a meno che tu non abiliti esplicitamente il fallback cloud.",
+  "settings.voice.pttHint": "Scorciatoia Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Tenere premuto per registrare.",
+  "settings.voice.speechRate": "Velocità del parlato",
+  "settings.voice.speechRateHint": "Quanto velocemente parla la voce di sintesi. 1.0 = velocità normale.",
+  "settings.voice.speechVolume": "Volume del parlato",
+  "settings.voice.speechVolumeHint": "Volume della voce di sintesi. Non influisce sull'ingresso del microfono.",
+  "settings.voice.sttEngine": "Motore di riconoscimento vocale (STT)",
+  "settings.voice.sttEngineHint": "Scegli il motore che converte il tuo parlato in testo. 'Auto' sceglie la migliore opzione disponibile.",
   "settings.voice.title": "Controllo vocale",
+  "settings.voice.ttsEngine": "Motore di sintesi vocale (TTS)",
+  "settings.voice.ttsEngineHint": "Scegli la voce che legge le risposte ad alta voce. 'Auto' sceglie la migliore opzione disponibile.",
   "settings.voice.ttsMuted": "Silenzia tutta l’uscita vocale",
   "settings.voice.ttsMutedHint": "Disabilita il text-to-speech mantenendo gli indicatori visivi.",
+  "settings.voice.wakeWordPhrase": "Frase di attivazione",
+  "settings.voice.wakeWordPhraseHint": "La frase che dici per attivare i comandi vocali a mani libere.",
   "settings.voice.wasmModels": "Modelli vocali WASM",
   "settings.voice.wasmModelsNotReady": "Modelli Whisper STT (~42 MB) e Kokoro TTS (~15 MB) non scaricati",
   "settings.voice.wasmModelsReady": "Scaricati e pronti per il riconoscimento vocale locale",
@@ -1975,21 +2045,6 @@
   "voice.modelDownload.progress": "{{percent}}% completato",
   "voice.modelDownload.retry": "Riprova",
   "voice.modelDownload.title": "Download modello vocale",
-  "settings.voice.sttEngine": "Motore di riconoscimento vocale (STT)",
-  "settings.voice.sttEngineHint": "Scegli il motore che converte il tuo parlato in testo. 'Auto' sceglie la migliore opzione disponibile.",
-  "settings.voice.ttsEngine": "Motore di sintesi vocale (TTS)",
-  "settings.voice.ttsEngineHint": "Scegli la voce che legge le risposte ad alta voce. 'Auto' sceglie la migliore opzione disponibile.",
-  "settings.voice.engine.auto": "Automatico (consigliato)",
-  "settings.voice.engine.webSpeech": "Web Speech API (cloud)",
-  "settings.voice.engine.wasm": "Modello WASM locale (offline)",
-  "settings.voice.speechRate": "Velocità del parlato",
-  "settings.voice.speechRateHint": "Quanto velocemente parla la voce di sintesi. 1.0 = velocità normale.",
-  "settings.voice.speechVolume": "Volume del parlato",
-  "settings.voice.speechVolumeHint": "Volume della voce di sintesi. Non influisce sull'ingresso del microfono.",
-  "settings.voice.wakeWordPhrase": "Frase di attivazione",
-  "settings.voice.wakeWordPhraseHint": "La frase che dici per attivare i comandi vocali a mani libere.",
-  "settings.voice.pttHint": "Scorciatoia Push-to-Talk: Ctrl+Shift+V (Windows/Linux) o ⌘+Shift+V (Mac). Tenere premuto per registrare.",
-  "settings.voice.privacy.revokeConsentHint": "Revoca il tuo consenso Web Speech. La voce passerà al WASM locale o verrà disattivata.",
   "help.advanced.adaptiveAi.content": "Il motore IA adattivo rileva l’hardware in fase di esecuzione — WebGPU, WebNN, DirectML, core CPU, livello VRAM e batteria — e instrada l’inferenza locale al backend più veloce disponibile. Attivalo in Impostazioni → Funzioni ad accesso anticipato e controlla la lettura in tempo reale in Impostazioni → IA. La modalità Eco sceglie il modello più piccolo utilizzabile per risparmiare batteria e memoria, e un’elezione della scheda capofila garantisce che solo una scheda del browser usi la GPU alla volta per evitare collisioni di VRAM.",
   "help.advanced.adaptiveAi.title": "IA adattiva, GPU e modalità Eco",
   "help.advanced.cloudSync.content": "La sincronizzazione opzionale crittografata end-to-end mantiene allineata la tua libreria su più dispositivi tramite un backend Cloudflare R2. Attiva «Sincronizzazione cloud» in Impostazioni → Funzioni ad accesso anticipato e configurala in Impostazioni → Connessioni. I dati del progetto vengono crittografati localmente con AES-256-GCM prima del caricamento e le tue chiavi API dell’IA non vengono mai inviate al cloud. La sincronizzazione è del tutto facoltativa; senza di essa StoryCraft resta un’app completamente offline-first.",
diff --git a/public/locales/ja/bundle.json b/public/locales/ja/bundle.json
index 0ae7b5ea3..74de4559d 100644
--- a/public/locales/ja/bundle.json
+++ b/public/locales/ja/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "あなたの物語の形を見てみると…",
   "proforge.pipeline.noneActive": "アクティブなパイプラインがありません",
   "proforge.pipeline.title": "究極の著者パイプライン",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "セッションで書いた言葉",
   "progress.allTime.title": "オールタイム",
   "progress.bestDay.title": "最高の一日",
diff --git a/public/locales/pt/bundle.json b/public/locales/pt/bundle.json
index 1200bae9e..59adde272 100644
--- a/public/locales/pt/bundle.json
+++ b/public/locales/pt/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "Olhando para a forma da sua história…",
   "proforge.pipeline.noneActive": "Nenhum pipeline ativo",
   "proforge.pipeline.title": "Pipeline final do autor",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "palavras escritas em sessões",
   "progress.allTime.title": "Todos os tempos",
   "progress.bestDay.title": "Melhor dia",
diff --git a/public/locales/zh/bundle.json b/public/locales/zh/bundle.json
index 15a219e1c..318a44f74 100644
--- a/public/locales/zh/bundle.json
+++ b/public/locales/zh/bundle.json
@@ -255,6 +255,61 @@
   "proforge.loading.structural": "看看你的故事的形状……",
   "proforge.pipeline.noneActive": "无活动管道",
   "proforge.pipeline.title": "最终作者管道",
+  "proforge.review.accept": "Accept",
+  "proforge.review.acceptAll": "Accept All",
+  "proforge.review.acceptAllCritical": "Accept All Critical",
+  "proforge.review.announceSubmitted": "{{stage}} review submitted",
+  "proforge.review.confidence": "{{percent}}% confidence",
+  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
+  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
+  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
+  "proforge.review.filter.accepted": "Accepted",
+  "proforge.review.filter.all": "All",
+  "proforge.review.filter.pending": "Pending",
+  "proforge.review.filter.rejected": "Rejected",
+  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.group.info": "Suggestions",
+  "proforge.review.group.warning": "Warnings",
+  "proforge.review.heading": "Review: {{stage}}",
+  "proforge.review.ignore": "Ignore",
+  "proforge.review.moreCritical": "+{{count}} more critical issues below",
+  "proforge.review.noItems": "No review items available.",
+  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.original": "Original:",
+  "proforge.review.pendingCount": "{{count}} Pending",
+  "proforge.review.proposed": "Proposed:",
+  "proforge.review.quickAccept": "Quick Accept ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
+  "proforge.review.reject": "Reject",
+  "proforge.review.rejectAll": "Reject All",
+  "proforge.review.severity.critical": "Critical",
+  "proforge.review.severity.info": "Info",
+  "proforge.review.severity.warning": "Warning",
+  "proforge.review.showDetails": "Show details",
+  "proforge.review.showLess": "Show less",
+  "proforge.review.skipStage": "Skip Stage",
+  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.type.arcIssue": "Arc",
+  "proforge.review.type.consistencyIssue": "Consistency",
+  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.type.legalWarning": "Legal",
+  "proforge.review.type.pacingIssue": "Pacing",
+  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.proseEdit": "Prose",
+  "proforge.review.type.repetitionHit": "Repetition",
+  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.styleEdit": "Style",
+  "proforge.review.type.technicalIssue": "Technical",
+  "proforge.stageName.analytics": "Analytics",
+  "proforge.stageName.archived": "Archived",
+  "proforge.stageName.copyEdit": "Copy Edit",
+  "proforge.stageName.idle": "Idle",
+  "proforge.stageName.intake": "Intake & Diagnostic",
+  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.stageName.production": "Production",
+  "proforge.stageName.proof": "Proofreading",
+  "proforge.stageName.publishing": "Publishing",
+  "proforge.stageName.structural": "Structural Edit",
   "progress.allTime.label": "会话中写下的单词",
   "progress.allTime.title": "所有时间",
   "progress.bestDay.title": "最好的一天",
diff --git a/services/proForge/applyReviewEdits.ts b/services/proForge/applyReviewEdits.ts
new file mode 100644
index 000000000..545dbeaf6
--- /dev/null
+++ b/services/proForge/applyReviewEdits.ts
@@ -0,0 +1,133 @@
+/**
+ * ProForge — apply accepted review edits back into manuscript text.
+ * QNBS-v3: Closes the core gap where "accept" recorded a status but never changed the
+ * manuscript. Offset-safe (back-to-front), with a text-match fallback for stale offsets,
+ * so an AI that returned slightly-off ranges still anchors correctly or is skipped — never
+ * corrupts the section. Applied edits are natively undoable (project slice is redux-undo wrapped).
+ */
+
+import type { ReviewItem } from '../../features/proForge/types';
+
+export interface SectionContentUpdate {
+  id: string;
+  content: string;
+}
+
+export interface ApplyEditsResult {
+  /** New content for the section after all applicable edits. */
+  content: string;
+  /** Number of edits successfully applied. */
+  applied: number;
+  /** Number of edits that could not be anchored (stale/overlapping) and were skipped. */
+  skipped: number;
+}
+
+interface PlannedEdit {
+  start: number;
+  end: number;
+  proposed: string;
+}
+
+function isValidRange(range: { start: number; end: number }, len: number): boolean {
+  return (
+    Number.isFinite(range.start) &&
+    Number.isFinite(range.end) &&
+    range.start >= 0 &&
+    range.end >= range.start &&
+    range.end <= len
+  );
+}
+
+/**
+ * Apply the accepted `ReviewItem`s that target a single section's content.
+ * Only items carrying a `proposed` replacement are considered text edits; advisory items
+ * (no `proposed`) are ignored and not counted as skipped.
+ */
+export function applyReviewEditsToSection(content: string, items: ReviewItem[]): ApplyEditsResult {
+  let skipped = 0;
+  const planned: PlannedEdit[] = [];
+
+  for (const item of items) {
+    // Advisory-only item (pacing note, quality score, plot-hole hint) — nothing to apply.
+    if (item.proposed === undefined) continue;
+    const proposed = item.proposed;
+
+    // Strategy 1: trust offsets only if they still resolve to the expected original text
+    // (or no original was provided to verify against).
+    if (item.range && isValidRange(item.range, content.length)) {
+      const slice = content.slice(item.range.start, item.range.end);
+      if (item.original === undefined || slice === item.original) {
+        planned.push({ start: item.range.start, end: item.range.end, proposed });
+        continue;
+      }
+    }
+
+    // Strategy 2: offsets missing or stale — locate the original text directly.
+    if (item.original) {
+      const idx = content.indexOf(item.original);
+      if (idx !== -1) {
+        planned.push({ start: idx, end: idx + item.original.length, proposed });
+        continue;
+      }
+    }
+
+    // Could not anchor the edit — skip rather than guess.
+    skipped++;
+  }
+
+  // Apply back-to-front so earlier edits never invalidate the offsets of later (already-applied) ones.
+  planned.sort((a, b) => b.start - a.start);
+  let next = content;
+  let lowerBound = content.length + 1; // start of the most recently applied (right-most) edit
+  let applied = 0;
+  for (const edit of planned) {
+    // Drop edits that overlap one already applied to its right.
+    if (edit.end > lowerBound) {
+      skipped++;
+      continue;
+    }
+    next = next.slice(0, edit.start) + edit.proposed + next.slice(edit.end);
+    lowerBound = edit.start;
+    applied++;
+  }
+
+  return { content: next, applied, skipped };
+}
+
+/**
+ * Plan content updates for a whole manuscript from a flat list of accepted review items.
+ * Returns only the sections whose content actually changed, plus aggregate counts.
+ */
+export function planAcceptedManuscriptEdits(
+  manuscript: ReadonlyArray<{ id: string; content?: string }>,
+  acceptedItems: ReadonlyArray<ReviewItem>,
+): { updates: SectionContentUpdate[]; applied: number; skipped: number } {
+  const bySection = new Map<string, ReviewItem[]>();
+  for (const item of acceptedItems) {
+    if (!item.sectionId || item.proposed === undefined) continue;
+    const list = bySection.get(item.sectionId);
+    if (list) list.push(item);
+    else bySection.set(item.sectionId, [item]);
+  }
+
+  const updates: SectionContentUpdate[] = [];
+  let applied = 0;
+  let skipped = 0;
+
+  for (const [sectionId, items] of bySection) {
+    const section = manuscript.find((s) => s.id === sectionId);
+    if (!section) {
+      // Section was deleted between analysis and acceptance.
+      skipped += items.length;
+      continue;
+    }
+    const result = applyReviewEditsToSection(section.content ?? '', items);
+    applied += result.applied;
+    skipped += result.skipped;
+    if (result.content !== (section.content ?? '')) {
+      updates.push({ id: sectionId, content: result.content });
+    }
+  }
+
+  return { updates, applied, skipped };
+}
diff --git a/services/proForge/pipelineAgents/baseAgent.ts b/services/proForge/pipelineAgents/baseAgent.ts
index 8334c0ff4..b30589d62 100644
--- a/services/proForge/pipelineAgents/baseAgent.ts
+++ b/services/proForge/pipelineAgents/baseAgent.ts
@@ -15,12 +15,20 @@ export abstract class BaseAgent {
   protected readonly context: OrchestratorContext;
   // QNBS-v3: Gateway injected from context or falls back to module singleton — keeps agents testable.
   protected readonly gateway: InferenceGateway;
+  // QNBS-v3: Supervisor feedback from the previous failed attempt; prepended to the next prompt
+  // so a retry is materially different instead of re-rolling the identical request.
+  private retryFeedback = '';
 
   constructor(context: OrchestratorContext) {
     this.context = context;
     this.gateway = context.gateway ?? inferenceGateway;
   }
 
+  /** Orchestrator-only: seed corrective feedback for a retry attempt. */
+  setRetryFeedback(feedback: string): void {
+    this.retryFeedback = feedback;
+  }
+
   abstract execute(
     signal: AbortSignal,
   ): Promise<Pick<StageResult, 'reviewItems' | 'metrics' | 'agentOutput'>>;
@@ -44,13 +52,25 @@ export abstract class BaseAgent {
   protected async generate(prompt: string, maxTokens?: number): Promise<string> {
     // QNBS-v3: exactOptionalPropertyTypes — only pass maxTokens when it's defined.
     const result = await this.gateway.generate({
-      prompt,
+      prompt: this.withRetryPreamble(prompt),
       creativity: this.context.config.creativity,
       options: this.buildAiOpts(maxTokens !== undefined ? { maxTokens } : undefined),
     });
     return result.text;
   }
 
+  // QNBS-v3: Prepend corrective guidance from the prior failed attempt so the model addresses
+  // the supervisor's concerns rather than repeating the same output.
+  private withRetryPreamble(prompt: string): string {
+    if (!this.retryFeedback) return prompt;
+    return `IMPORTANT — your previous attempt was rejected by the quality reviewer for these reasons:
+${this.retryFeedback}
+
+Produce a corrected response that resolves the issues above. Do not repeat the prior output.
+
+${prompt}`;
+  }
+
   // QNBS-v3: Builds AIRequestOptions from context.config — provider/model defaulting for pipeline agents.
   protected buildAiOpts(overrides?: {
     maxTokens?: number;
diff --git a/services/proForge/pipelineAgents/styleAdaptedWriterAgent.ts b/services/proForge/pipelineAgents/styleAdaptedWriterAgent.ts
deleted file mode 100644
index 2b2bcd33d..000000000
--- a/services/proForge/pipelineAgents/styleAdaptedWriterAgent.ts
+++ /dev/null
@@ -1,119 +0,0 @@
-/**
- * StyleAdaptedWriterAgent — Optional ProForge agent using an active LoRA adapter.
- * QNBS-v3: Extends BaseAgent; inserts after 'lineProse' via PipelineStageExtension.
- *          Produces proseEdit review items generated by the user's personal style model.
- *          Returns a fallback (isFallback: true) when no adapter is active.
- */
-
-import { v4 as uuid } from 'uuid';
-import type { ReviewItem, StageResult } from '../../../features/proForge/types';
-import { generateLocalText } from '../../localAiFacade';
-import { logger } from '../../logger';
-import { getActiveAdapter } from '../../loraAdapterService';
-import { BaseAgent } from './baseAgent';
-
-export class StyleAdaptedWriterAgent extends BaseAgent {
-  async execute(
-    signal: AbortSignal,
-  ): Promise<Pick<StageResult, 'reviewItems' | 'metrics' | 'agentOutput'>> {
-    const startTime = performance.now();
-
-    // Resolve active adapter from IDB (not Redux — agents run outside React lifecycle)
-    const activeAdapter = await getActiveAdapter();
-
-    if (!activeAdapter) {
-      logger.info('StyleAdaptedWriterAgent: no active LoRA adapter — returning fallback');
-      return {
-        reviewItems: [],
-        metrics: {
-          aiCalls: 0,
-          tokensConsumed: 0,
-          durationMs: this.elapsed(startTime),
-          itemsFound: 0,
-          itemsAccepted: 0,
-          itemsRejected: 0,
-        },
-        agentOutput: { isFallback: true, reason: 'no_active_adapter' },
-      };
-    }
-
-    if (signal.aborted) {
-      return {
-        reviewItems: [],
-        metrics: {
-          aiCalls: 0,
-          tokensConsumed: 0,
-          durationMs: 0,
-          itemsFound: 0,
-          itemsAccepted: 0,
-          itemsRejected: 0,
-        },
-        agentOutput: { isFallback: true, reason: 'aborted' },
-      };
-    }
-
-    const project = this.requireProject();
-    const sections = project.manuscript ?? [];
-    const reviewItems: ReviewItem[] = [];
-    let aiCalls = 0;
-    let tokensConsumed = 0;
-
-    for (const section of sections.slice(0, 3)) {
-      if (signal.aborted) break;
-      const text = section.content ?? '';
-      if (text.length < 200) continue;
-
-      const prompt = `You are a style-adapted writing assistant trained on this author's voice. Suggest one improvement to this passage that preserves the author's style while enhancing prose quality:\n\n${text.slice(0, 800)}`;
-
-      try {
-        const response = await generateLocalText(
-          prompt,
-          activeAdapter.modelCompatibility,
-          undefined,
-          activeAdapter.id,
-        );
-        const suggestion = response?.text ?? '';
-        if (suggestion.length > 50) {
-          reviewItems.push({
-            id: uuid(),
-            stage: 'lineProse' as const,
-            type: 'proseEdit',
-            severity: 'info',
-            status: 'pending',
-            sectionId: section.id,
-            description: `Style-adapted suggestion from ${activeAdapter.name}`,
-            original: text.slice(0, 200),
-            proposed: suggestion,
-            rationale: `Adapter: ${activeAdapter.name}`,
-            confidence: 0.7,
-            createdAt: new Date().toISOString(),
-          });
-          aiCalls++;
-          tokensConsumed += suggestion.length;
-        }
-      } catch (err) {
-        logger.warn('StyleAdaptedWriterAgent: generation failed for section', {
-          sectionId: section.id,
-          err,
-        });
-      }
-    }
-
-    return {
-      reviewItems,
-      metrics: {
-        aiCalls,
-        tokensConsumed,
-        durationMs: this.elapsed(startTime),
-        itemsFound: reviewItems.length,
-        itemsAccepted: 0,
-        itemsRejected: 0,
-      },
-      agentOutput: {
-        adapterId: activeAdapter.id,
-        adapterName: activeAdapter.name,
-        suggestionsGenerated: reviewItems.length,
-      },
-    };
-  }
-}
diff --git a/services/proForge/pipelineAgents/supervisorAgent.ts b/services/proForge/pipelineAgents/supervisorAgent.ts
index 412284514..b2a5def5c 100644
--- a/services/proForge/pipelineAgents/supervisorAgent.ts
+++ b/services/proForge/pipelineAgents/supervisorAgent.ts
@@ -5,8 +5,12 @@
  */
 
 import type {
+  CopyEditPlan,
   DiagnosticReport,
+  PipelineAnalyticsReport,
   PipelineStage,
+  ProductionManifest,
+  PublishingPackage,
   QualityGateReport,
   StageResult,
   StructuralEditPlan,
@@ -30,8 +34,18 @@ export class SupervisorAgent {
         return this.evaluateIntake(result);
       case 'structural':
         return this.evaluateStructural(result);
+      case 'lineProse':
+        return this.evaluateLineProse(result);
+      case 'copyEdit':
+        return this.evaluateCopyEdit(result);
       case 'proof':
         return this.evaluateProof(result);
+      case 'production':
+        return this.evaluateProduction(result);
+      case 'publishing':
+        return this.evaluatePublishing(result);
+      case 'analytics':
+        return this.evaluateAnalytics(result);
       default:
         return { pass: true, retryRecommended: false, qualityScore: 100, reasons: [] };
     }
@@ -43,12 +57,9 @@ export class SupervisorAgent {
     const output = result.agentOutput as DiagnosticReport | undefined;
     const reasons: string[] = [];
 
-    // QNBS-v3: qualityScore.overall===50 across all 6 dimensions is the fallback sentinel.
-    const isFallback =
-      output?.qualityScore !== undefined &&
-      output.qualityScore.overall === 50 &&
-      output.qualityScore.prose === 50 &&
-      output.qualityScore.structure === 50;
+    // QNBS-v3: Agents mark synthetic reports with isFallback:true + zeroed scores
+    // (see createFallbackReport). The previous `=== 50` heuristic never matched.
+    const isFallback = output?.isFallback === true;
 
     if (isFallback) {
       reasons.push(
@@ -77,6 +88,12 @@ export class SupervisorAgent {
     const output = result.agentOutput as StructuralEditPlan | undefined;
     const reasons: string[] = [];
 
+    // QNBS-v3: Honour the explicit fallback marker before heuristics.
+    if (output?.isFallback === true) {
+      reasons.push('Structural analysis returned a fallback plan — the AI call did not complete.');
+      return { pass: false, retryRecommended: true, qualityScore: 0, reasons };
+    }
+
     const wordCount = this.estimateManuscriptWordCount();
     const hasEdits = (output?.edits?.length ?? 0) > 0;
     const hasReviewItems = result.reviewItems.length > 0;
@@ -98,6 +115,12 @@ export class SupervisorAgent {
     const output = result.agentOutput as QualityGateReport | undefined;
     const reasons: string[] = [];
 
+    // QNBS-v3: Honour the explicit fallback marker before heuristics.
+    if (output?.isFallback === true) {
+      reasons.push('Proof/quality gate returned a fallback report — the AI call did not complete.');
+      return { pass: false, retryRecommended: true, qualityScore: 0, reasons };
+    }
+
     const wordCount = this.estimateManuscriptWordCount();
     const seemsFallback =
       output?.overallPass === true &&
@@ -114,6 +137,103 @@ export class SupervisorAgent {
     return { pass: true, retryRecommended: false, qualityScore: 90, reasons };
   }
 
+  private evaluateLineProse(
+    result: Pick<StageResult, 'reviewItems' | 'agentOutput'>,
+  ): SupervisionDecision {
+    const output = result.agentOutput as { edits?: unknown[] } | undefined;
+    const wordCount = this.estimateManuscriptWordCount();
+    const hasEdits = (output?.edits?.length ?? 0) > 0;
+
+    // QNBS-v3: A substantial manuscript with zero prose edits suggests the AI call didn't land.
+    if (!hasEdits && result.reviewItems.length === 0 && wordCount > 1000) {
+      return {
+        pass: false,
+        retryRecommended: true,
+        qualityScore: 45,
+        reasons: [
+          `No prose edits for a ${wordCount}-word manuscript — verify the AI provider responded.`,
+        ],
+      };
+    }
+    return { pass: true, retryRecommended: false, qualityScore: 85, reasons: [] };
+  }
+
+  private evaluateCopyEdit(
+    result: Pick<StageResult, 'reviewItems' | 'agentOutput'>,
+  ): SupervisionDecision {
+    const output = result.agentOutput as CopyEditPlan | undefined;
+    const total =
+      (output?.grammarEdits?.length ?? 0) +
+      (output?.styleEdits?.length ?? 0) +
+      (output?.repetitionHits?.length ?? 0) +
+      (output?.formatIssues?.length ?? 0);
+    const wordCount = this.estimateManuscriptWordCount();
+
+    // QNBS-v3: Zero grammar/style/repetition/format findings on a long manuscript is suspicious.
+    if (total === 0 && result.reviewItems.length === 0 && wordCount > 1500) {
+      return {
+        pass: false,
+        retryRecommended: true,
+        qualityScore: 50,
+        reasons: [
+          'Copy-edit found zero grammar/style/repetition issues on a long manuscript — verify the AI ran.',
+        ],
+      };
+    }
+    return { pass: true, retryRecommended: false, qualityScore: 88, reasons: [] };
+  }
+
+  private evaluateProduction(
+    result: Pick<StageResult, 'reviewItems' | 'agentOutput'>,
+  ): SupervisionDecision {
+    const output = result.agentOutput as ProductionManifest | undefined;
+    const artifactCount = output?.artifacts?.length ?? 0;
+
+    // QNBS-v3: Production must emit at least one artifact; none means the build failed.
+    if (artifactCount === 0) {
+      return {
+        pass: false,
+        retryRecommended: true,
+        qualityScore: 0,
+        reasons: ['Production produced no export artifacts.'],
+      };
+    }
+    return { pass: true, retryRecommended: false, qualityScore: 95, reasons: [] };
+  }
+
+  private evaluatePublishing(
+    result: Pick<StageResult, 'reviewItems' | 'agentOutput'>,
+  ): SupervisionDecision {
+    const output = result.agentOutput as PublishingPackage | undefined;
+    const reasons: string[] = [];
+    const hasTitle = (output?.metadata?.title ?? '').trim().length > 0;
+    const hasBlurb = (output?.blurbs?.backCover ?? '').trim().length > 0;
+
+    if (!hasTitle) reasons.push('Publishing package is missing book-title metadata.');
+    if (!hasBlurb) reasons.push('Publishing package is missing a back-cover blurb.');
+    if (reasons.length > 0) {
+      return { pass: false, retryRecommended: true, qualityScore: 40, reasons };
+    }
+    return { pass: true, retryRecommended: false, qualityScore: 90, reasons: [] };
+  }
+
+  private evaluateAnalytics(
+    result: Pick<StageResult, 'reviewItems' | 'agentOutput'>,
+  ): SupervisionDecision {
+    const output = result.agentOutput as PipelineAnalyticsReport | undefined;
+    // QNBS-v3: Analytics is the terminal, informational stage — never block the pipeline on it,
+    // but flag a missing metrics block so the trace records it.
+    if (!output?.metrics) {
+      return {
+        pass: true,
+        retryRecommended: false,
+        qualityScore: 70,
+        reasons: ['Analytics report has no metrics block (non-blocking).'],
+      };
+    }
+    return { pass: true, retryRecommended: false, qualityScore: 100, reasons: [] };
+  }
+
   private estimateManuscriptWordCount(): number {
     const project = this.context.getState().project.present?.data;
     if (!project) return 0;
diff --git a/services/proForge/pipelineTools/toolRegistry.ts b/services/proForge/pipelineTools/toolRegistry.ts
deleted file mode 100644
index fcff2fd9d..000000000
--- a/services/proForge/pipelineTools/toolRegistry.ts
+++ /dev/null
@@ -1,303 +0,0 @@
-/**
- * ProForge Tool Registry — Tool-calling infrastructure for agentic AI.
- * QNBS-v3: Vercel AI SDK compatible tool schemas with sandboxed execution.
- */
-
-import { z } from 'zod';
-import { logger } from '../../logger';
-
-// ---------------------------------------------------------------------------
-// Tool Schema Types
-// ---------------------------------------------------------------------------
-
-export interface ToolDefinition {
-  name: string;
-  description: string;
-  parameters: z.ZodType<unknown>;
-  handler: (args: unknown, context: ToolContext) => Promise<unknown>;
-  /** Which pipeline stages this tool is available in */
-  stages: string[];
-  /** Whether this tool modifies project data */
-  isWrite: boolean;
-}
-
-export interface ToolContext {
-  projectId: string;
-  dispatch: import('../../../app/store').AppDispatch;
-  getState: () => import('../../../app/store').RootState;
-  memoryBank: import('../proForgeMemoryBank').ProForgeMemoryBank;
-  signal: AbortSignal;
-}
-
-// ---------------------------------------------------------------------------
-// Registry
-// ---------------------------------------------------------------------------
-
-class ToolRegistry {
-  private tools = new Map<string, ToolDefinition>();
-
-  register(tool: ToolDefinition): void {
-    if (this.tools.has(tool.name)) {
-      logger.warn(`ToolRegistry: Overwriting existing tool "${tool.name}"`);
-    }
-    this.tools.set(tool.name, tool);
-  }
-
-  unregister(name: string): boolean {
-    return this.tools.delete(name);
-  }
-
-  get(name: string): ToolDefinition | undefined {
-    return this.tools.get(name);
-  }
-
-  list(): ToolDefinition[] {
-    return Array.from(this.tools.values());
-  }
-
-  listForStage(stage: string): ToolDefinition[] {
-    return this.list().filter((t) => t.stages.includes(stage) || t.stages.includes('*'));
-  }
-
-  async execute(name: string, args: unknown, context: ToolContext): Promise<unknown> {
-    const tool = this.tools.get(name);
-    if (!tool) {
-      throw new Error(`Tool "${name}" not found in registry`);
-    }
-    // Validate args
-    const parsed = tool.parameters.safeParse(args);
-    if (!parsed.success) {
-      throw new Error(`Invalid arguments for tool "${name}": ${parsed.error.message}`);
-    }
-    return tool.handler(parsed.data, context);
-  }
-
-  /** Convert to Vercel AI SDK tool format */
-  toAiSdkTools(
-    stage: string,
-  ): Record<string, { description: string; parameters: z.ZodType<unknown> }> {
-    const result: Record<string, { description: string; parameters: z.ZodType<unknown> }> = {};
-    for (const tool of this.listForStage(stage)) {
-      result[tool.name] = {
-        description: tool.description,
-        parameters: tool.parameters,
-      };
-    }
-    return result;
-  }
-}
-
-export const toolRegistry = new ToolRegistry();
-
-// ---------------------------------------------------------------------------
-// Built-in Tools
-// ---------------------------------------------------------------------------
-
-// -- Manuscript Tools --
-
-toolRegistry.register({
-  name: 'readSection',
-  description: 'Read the full content of a manuscript section by ID.',
-  parameters: z.object({
-    sectionId: z.string().describe('The section ID to read'),
-  }),
-  stages: ['*'],
-  isWrite: false,
-  handler: async (args, context) => {
-    const { sectionId } = args as { sectionId: string };
-    const state = context.getState();
-    const sections = state.project.present?.data?.manuscript ?? [];
-    const section = sections.find((s) => s.id === sectionId);
-    if (!section) {
-      return { error: `Section "${sectionId}" not found` };
-    }
-    return {
-      id: section.id,
-      title: section.title,
-      content: section.content,
-      status: section.status,
-    };
-  },
-});
-
-toolRegistry.register({
-  name: 'readAllSections',
-  description:
-    'Read all manuscript sections (metadata only, no content). Use readSection for full content.',
-  parameters: z.object({
-    includeContent: z
-      .boolean()
-      .optional()
-      .describe('If true, includes full content (warning: large payload)'),
-  }),
-  stages: ['*'],
-  isWrite: false,
-  handler: async (args, context) => {
-    const { includeContent } = (args as { includeContent?: boolean }) ?? {};
-    const state = context.getState();
-    const sections = state.project.present?.data?.manuscript ?? [];
-    return sections.map((s) => ({
-      id: s.id,
-      title: s.title,
-      wordCount: s.content ? s.content.trim().split(/\s+/).length : 0,
-      status: s.status,
-      act: s.act,
-      ...(includeContent ? { content: s.content } : {}),
-    }));
-  },
-});
-
-toolRegistry.register({
-  name: 'readProjectMeta',
-  description: 'Read project metadata: title, logline, characters, worlds.',
-  parameters: z.object({}),
-  stages: ['*'],
-  isWrite: false,
-  handler: async (_args, context) => {
-    const state = context.getState();
-    const data = state.project.present?.data;
-    if (!data) return { error: 'No project data' };
-    return {
-      title: data.title,
-      logline: data.logline,
-      characters: Object.values(data.characters?.entities ?? {}),
-      worlds: Object.values(data.worlds?.entities ?? {}),
-    };
-  },
-});
-
-toolRegistry.register({
-  name: 'searchLore',
-  description: 'Search the story lore (characters, worlds, codex) for a query.',
-  parameters: z.object({
-    query: z.string().describe('Search query'),
-    limit: z.number().optional().describe('Max results'),
-  }),
-  stages: ['*'],
-  isWrite: false,
-  handler: async (args, context) => {
-    const { query, limit = 5 } = args as { query: string; limit?: number };
-    const entries = await context.memoryBank.search(query, limit);
-    return entries.map((e) => ({
-      category: e.category,
-      key: e.key,
-      content: e.content,
-    }));
-  },
-});
-
-// -- Analysis Tools --
-
-toolRegistry.register({
-  name: 'analyzePacing',
-  description: 'Analyze manuscript pacing per section. Returns tension scores and recommendations.',
-  parameters: z.object({}),
-  stages: ['intake', 'structural'],
-  isWrite: false,
-  handler: async (_args, context) => {
-    const state = context.getState();
-    const sections = state.project.present?.data?.manuscript ?? [];
-    // Simple heuristic: longer sections = slower pacing, shorter = faster
-    const wordCounts = sections.map((s) => s.content?.trim().split(/\s+/).length ?? 0);
-    const avg = wordCounts.reduce((a, b) => a + b, 0) / (wordCounts.length || 1);
-    const analysis = sections.map((s, i) => {
-      const wc = wordCounts[i] ?? 0;
-      const ratio = avg > 0 ? wc / avg : 1;
-      let tensionScore = 5;
-      if (ratio > 1.5) tensionScore = 3; // slow
-      if (ratio < 0.7) tensionScore = 7; // fast
-      return {
-        sectionId: s.id,
-        title: s.title,
-        wordCount: wc,
-        tensionScore,
-        recommendedAction: ratio > 1.5 ? 'compress' : ratio < 0.7 ? 'expand' : 'keep',
-      };
-    });
-    return { sections: analysis, overallPacing: 'uneven' };
-  },
-});
-
-toolRegistry.register({
-  name: 'countWords',
-  description: 'Count total words in the manuscript.',
-  parameters: z.object({}),
-  stages: ['*'],
-  isWrite: false,
-  handler: async (_args, context) => {
-    const state = context.getState();
-    const sections = state.project.present?.data?.manuscript ?? [];
-    const counts = sections.map((s) => s.content?.trim().split(/\s+/).length ?? 0);
-    const total = counts.reduce((a, b) => a + b, 0);
-    return { total, perSection: counts };
-  },
-});
-
-// -- Report Tools --
-
-toolRegistry.register({
-  name: 'generateReport',
-  description: 'Save a structured report to the memory bank for later retrieval.',
-  parameters: z.object({
-    category: z.enum(['lore', 'character', 'style', 'feedback', 'edit', 'meta']),
-    key: z.string(),
-    content: z.string(),
-  }),
-  stages: ['*'],
-  isWrite: true,
-  handler: async (args, context) => {
-    const { category, key, content } = args as {
-      category: import('../../../features/proForge/types').MemoryBankEntry['category'];
-      key: string;
-      content: string;
-    };
-    const run = context.getState().proForge.currentRun;
-    const stage = run?.activeStage ?? 'intake';
-    await context.memoryBank.remember(category, key, content, stage);
-    return { success: true };
-  },
-});
-
-toolRegistry.register({
-  name: 'getMemoryContext',
-  description: 'Retrieve relevant memory context for the current stage.',
-  parameters: z.object({
-    query: z.string().optional(),
-    maxChars: z.number().optional(),
-  }),
-  stages: ['*'],
-  isWrite: false,
-  handler: async (args, context) => {
-    const { query, maxChars = 4000 } = args as { query?: string; maxChars?: number };
-    const run = context.getState().proForge.currentRun;
-    const stage = run?.activeStage ?? 'intake';
-    const contextStr = await context.memoryBank.buildContextString(stage, query, maxChars);
-    return { context: contextStr };
-  },
-});
-
-// -- Edit Proposal Tools (write operations gated) --
-
-toolRegistry.register({
-  name: 'proposeEdit',
-  description:
-    'Propose an edit to a manuscript section. The edit is NOT applied automatically; it goes to the review queue.',
-  parameters: z.object({
-    sectionId: z.string(),
-    startOffset: z.number(),
-    endOffset: z.number(),
-    original: z.string(),
-    proposed: z.string(),
-    rationale: z.string(),
-    category: z.string(),
-    confidence: z.number().min(0).max(1),
-  }),
-  stages: ['structural', 'lineProse', 'copyEdit'],
-  isWrite: true,
-  handler: async (args) => {
-    // Returns the proposal for the agent to collect into review items
-    return { proposal: args, status: 'queued_for_review' };
-  },
-});
-
-export { ToolRegistry };
diff --git a/services/proForge/proForgeHistoryStore.ts b/services/proForge/proForgeHistoryStore.ts
new file mode 100644
index 000000000..abc13d057
--- /dev/null
+++ b/services/proForge/proForgeHistoryStore.ts
@@ -0,0 +1,65 @@
+/**
+ * ProForge Run-History Store — persists completed/aborted pipeline runs per project.
+ * QNBS-v3: The proForge slice is ephemeral (cleared on reload), so analytics comparisons across
+ * runs were always empty after a refresh. This IDB-backed store survives reloads. Capped per
+ * project to keep storage bounded.
+ */
+
+import type { PipelineRun } from '../../features/proForge/types';
+
+const HISTORY_DB = 'proforge-run-history';
+const HISTORY_VERSION = 1;
+const STORE = 'history';
+/** Keep at most this many runs per project (most recent first). */
+export const MAX_RUN_HISTORY = 20;
+
+let dbPromise: Promise<IDBDatabase> | null = null;
+
+function openHistoryDb(): Promise<IDBDatabase> {
+  if (dbPromise) return dbPromise;
+  dbPromise = new Promise((resolve, reject) => {
+    const request = indexedDB.open(HISTORY_DB, HISTORY_VERSION);
+    request.onerror = () => reject(new Error('Failed to open ProForge history DB'));
+    request.onsuccess = () => resolve(request.result);
+    request.onupgradeneeded = (event) => {
+      const db = (event.target as IDBOpenDBRequest).result;
+      if (!db.objectStoreNames.contains(STORE)) {
+        db.createObjectStore(STORE, { keyPath: 'projectId' });
+      }
+    };
+  });
+  return dbPromise;
+}
+
+interface HistoryRecord {
+  projectId: string;
+  runs: PipelineRun[];
+}
+
+/** Persist the run history for a project (capped to MAX_RUN_HISTORY, most-recent-first). */
+export async function saveRunHistory(projectId: string, runs: PipelineRun[]): Promise<void> {
+  const db = await openHistoryDb();
+  const record: HistoryRecord = { projectId, runs: runs.slice(0, MAX_RUN_HISTORY) };
+  await new Promise<void>((resolve, reject) => {
+    const tx = db.transaction(STORE, 'readwrite');
+    const request = tx.objectStore(STORE).put(record);
+    request.onsuccess = () => resolve();
+    request.onerror = () => reject(new Error('Failed to save ProForge run history'));
+  });
+}
+
+/** Load the persisted run history for a project (empty array if none). */
+export async function loadRunHistory(projectId: string): Promise<PipelineRun[]> {
+  const db = await openHistoryDb();
+  return new Promise((resolve, reject) => {
+    const tx = db.transaction(STORE, 'readonly');
+    const request = tx.objectStore(STORE).get(projectId);
+    request.onsuccess = () => resolve((request.result as HistoryRecord | undefined)?.runs ?? []);
+    request.onerror = () => reject(new Error('Failed to load ProForge run history'));
+  });
+}
+
+/** Reset the DB connection — test-only. */
+export function _resetHistoryDbForTest(): void {
+  dbPromise = null;
+}
diff --git a/services/proForge/proForgeMemoryBank.ts b/services/proForge/proForgeMemoryBank.ts
index 0e1cfe988..d6c276644 100644
--- a/services/proForge/proForgeMemoryBank.ts
+++ b/services/proForge/proForgeMemoryBank.ts
@@ -1,6 +1,7 @@
 /**
  * ProForge Memory Bank — Persistent project-specific memory for agentic context.
- * QNBS-v3: IndexedDB-backed storage with hybrid RAG retrieval for agents.
+ * QNBS-v3: IndexedDB-backed storage. Retrieval honours the run's ragMode — keyword by default,
+ * or semantic/hybrid via the local MiniLM embedding service (best-effort, keyword fallback).
  */
 
 import type { MemoryBankEntry, PipelineStage } from '../../features/proForge/types';
@@ -88,29 +89,72 @@ export async function getMemoryEntries(
   });
 }
 
+export type MemoryRagMode = 'lexical' | 'semantic' | 'hybrid';
+
+function keywordScorer(query: string): (entry: MemoryBankEntry) => number {
+  const keywords = query
+    .toLowerCase()
+    .split(/\s+/)
+    .filter((k) => k.length > 2);
+  return (entry) => {
+    const text = `${entry.key} ${entry.content}`.toLowerCase();
+    let score = 0;
+    for (const kw of keywords) if (text.includes(kw)) score += 1;
+    return score;
+  };
+}
+
+function rankByKeyword(
+  entries: MemoryBankEntry[],
+  query: string,
+  limit: number,
+): MemoryBankEntry[] {
+  const score = keywordScorer(query);
+  return entries
+    .map((entry) => ({ entry, score: score(entry) }))
+    .filter((s) => s.score > 0)
+    .sort((a, b) => b.score - a.score)
+    .slice(0, limit)
+    .map((s) => s.entry);
+}
+
 export async function searchMemoryEntries(
   projectId: string,
   query: string,
   limit = 10,
+  mode: MemoryRagMode = 'lexical',
 ): Promise<MemoryBankEntry[]> {
   const entries = await getMemoryEntries(projectId);
-  // Simple keyword scoring for now; can be upgraded to semantic search
-  const queryLower = query.toLowerCase();
-  const scored = entries
-    .map((entry) => {
-      const text = `${entry.key} ${entry.content}`.toLowerCase();
-      let score = 0;
-      const keywords = queryLower.split(/\s+/).filter((k) => k.length > 2);
-      for (const kw of keywords) {
-        if (text.includes(kw)) score += 1;
-      }
-      return { entry, score };
-    })
-    .filter((s) => s.score > 0)
-    .sort((a, b) => b.score - a.score)
-    .slice(0, limit);
+  if (entries.length === 0) return [];
 
-  return scored.map((s) => s.entry);
+  if (mode === 'lexical') {
+    return rankByKeyword(entries, query, limit);
+  }
+
+  // QNBS-v3: Semantic/hybrid recall — best-effort. The embedding service is dynamically imported
+  // so the memory bank stays light in node tests, and ANY failure (no model, offline) falls back
+  // to keyword ranking rather than breaking the pipeline.
+  try {
+    const { embedText, cosineSimilarity } = await import('../ai/localEmbeddingService');
+    const scoreKeyword = keywordScorer(query);
+    const maxKw = Math.max(1, ...entries.map(scoreKeyword));
+    const qVec = await embedText(query);
+    const scored = await Promise.all(
+      entries.map(async (entry) => {
+        const eVec = entry.embedding ?? (await embedText(`${entry.key} ${entry.content}`));
+        const sim = cosineSimilarity(qVec, eVec);
+        // QNBS-v3: hybrid blends semantic (0.7) with normalised keyword overlap (0.3).
+        const score = mode === 'semantic' ? sim : 0.7 * sim + 0.3 * (scoreKeyword(entry) / maxKw);
+        return { entry, score };
+      }),
+    );
+    return scored
+      .sort((a, b) => b.score - a.score)
+      .slice(0, limit)
+      .map((s) => s.entry);
+  } catch {
+    return rankByKeyword(entries, query, limit);
+  }
 }
 
 export async function deleteMemoryEntry(id: string): Promise<void> {
@@ -159,8 +203,12 @@ export class ProForgeMemoryBank {
     return getMemoryEntries(this.projectId, category);
   }
 
-  async search(query: string, limit = 10): Promise<MemoryBankEntry[]> {
-    return searchMemoryEntries(this.projectId, query, limit);
+  async search(
+    query: string,
+    limit = 10,
+    mode: MemoryRagMode = 'lexical',
+  ): Promise<MemoryBankEntry[]> {
+    return searchMemoryEntries(this.projectId, query, limit, mode);
   }
 
   async recallForStage(stage: PipelineStage): Promise<MemoryBankEntry[]> {
@@ -192,10 +240,15 @@ export class ProForgeMemoryBank {
   /**
    * Build a context string for prompt injection from relevant memory entries.
    */
-  async buildContextString(stage: PipelineStage, query?: string, maxChars = 4000): Promise<string> {
+  async buildContextString(
+    stage: PipelineStage,
+    query?: string,
+    maxChars = 4000,
+    mode: MemoryRagMode = 'lexical',
+  ): Promise<string> {
     let entries: MemoryBankEntry[];
     if (query) {
-      entries = await this.search(query, 20);
+      entries = await this.search(query, 20, mode);
     } else {
       entries = await this.recallForStage(stage);
     }
diff --git a/services/proForge/proForgeOrchestrator.ts b/services/proForge/proForgeOrchestrator.ts
index 02a3cc3fe..2985e648c 100644
--- a/services/proForge/proForgeOrchestrator.ts
+++ b/services/proForge/proForgeOrchestrator.ts
@@ -20,8 +20,9 @@ import type {
   ReviewItemStatus,
   SupervisionDecision,
 } from '../../features/proForge/types';
-import { nextStage } from '../../features/proForge/types';
+import { isEditingStage, nextStage } from '../../features/proForge/types';
 import { logger } from '../logger';
+import { planAcceptedManuscriptEdits } from './applyReviewEdits';
 
 // ---------------------------------------------------------------------------
 // Agent imports (lazy to avoid circular deps at module level)
@@ -76,6 +77,28 @@ export class ProForgeOrchestrator {
     this.context = context;
   }
 
+  // -------------------------------------------------------------------------
+  // Internal helpers
+  // -------------------------------------------------------------------------
+
+  /** QNBS-v3: HEAD snapshot of the current branch — deduplicates the 4× inline lookup. */
+  private headSnapshotId(): string | undefined {
+    const vcState = this.context.getState().versionControl;
+    return vcState.branches.find((b) => b.id === vcState.currentBranchId)?.headSnapshotId;
+  }
+
+  /** QNBS-v3: Best-effort persist of run history to IDB (survives reload). Never throws. */
+  private async persistHistory(): Promise<void> {
+    try {
+      const runHistory = this.context.getState().proForge.runHistory;
+      if (!runHistory?.length) return;
+      const { saveRunHistory } = await import('./proForgeHistoryStore');
+      await saveRunHistory(this.context.projectId, runHistory);
+    } catch (err) {
+      logger.warn('ProForge: failed to persist run history', err);
+    }
+  }
+
   // -------------------------------------------------------------------------
   // Public API
   // -------------------------------------------------------------------------
@@ -86,6 +109,9 @@ export class ProForgeOrchestrator {
    */
   async startPipeline(label: string, config: PipelineConfig): Promise<void> {
     const { dispatch, getState, projectId } = this.context;
+    // QNBS-v3: Reset the abort signal — the orchestrator instance is reused across runs
+    // (cached in a hook ref); a prior abort would otherwise leave every agent pre-aborted.
+    this.abortController = new AbortController();
     const state = getState();
 
     // Create pre-pipeline snapshot via version control
@@ -106,9 +132,7 @@ export class ProForgeOrchestrator {
     );
 
     // Retrieve the snapshot ID (it's the last one created on current branch)
-    const vcState = getState().versionControl;
-    const currentBranch = vcState.branches.find((b) => b.id === vcState.currentBranchId);
-    const preSnapshotId = currentBranch?.headSnapshotId ?? 'unknown';
+    const preSnapshotId = this.headSnapshotId() ?? 'unknown';
 
     dispatch(
       (await import('../../features/proForge/proForgeSlice')).startPipeline({
@@ -150,9 +174,7 @@ export class ProForgeOrchestrator {
         }),
       );
     }
-    const vcState = getState().versionControl;
-    const currentBranch = vcState.branches.find((b) => b.id === vcState.currentBranchId);
-    const snapshotId = currentBranch?.headSnapshotId;
+    const snapshotId = this.headSnapshotId();
 
     dispatch(stageStarted({ stage, ...(snapshotId !== undefined && { snapshotId }) }));
 
@@ -166,11 +188,14 @@ export class ProForgeOrchestrator {
     maxRetries: number,
   ): Promise<void> {
     const { dispatch } = this.context;
+    // QNBS-v3: Carries the prior attempt's rejection reasons into the next prompt.
+    let retryFeedback = '';
 
     for (let attempt = 0; attempt <= maxRetries; attempt++) {
       try {
         const AgentClass = await loadAgent(stage);
         const agent = new AgentClass(this.context);
+        if (retryFeedback) agent.setRetryFeedback(retryFeedback);
         const result = await agent.execute(this.abortController.signal);
 
         if (this.abortController.signal.aborted) {
@@ -187,6 +212,13 @@ export class ProForgeOrchestrator {
             `SupervisorAgent: Stage ${stage} flagged (retry ${attempt + 1}):`,
             decision.reasons,
           );
+          // QNBS-v3: Feed the supervisor's reasons + any self-reflection note into the retry.
+          const reflectionNote = (result.agentOutput as { reflectionNotes?: string } | undefined)
+            ?.reflectionNotes;
+          retryFeedback = [...decision.reasons, reflectionNote]
+            .filter((r): r is string => Boolean(r))
+            .map((r) => `- ${r}`)
+            .join('\n');
           continue;
         }
 
@@ -235,6 +267,7 @@ export class ProForgeOrchestrator {
     if (!next || next === 'archived') {
       // Pipeline complete
       dispatch(pipelineCompleted());
+      await this.persistHistory();
       return;
     }
 
@@ -259,7 +292,41 @@ export class ProForgeOrchestrator {
   ): Promise<void> {
     const { dispatch, getState } = this.context;
 
-    // Create post-stage snapshot after applying accepted edits
+    // QNBS-v3: Apply accepted edits to the manuscript BEFORE snapshotting, so the post-stage
+    // snapshot captures the edited text. Only editing stages mutate prose; production/publishing/
+    // analytics are advisory. Stale/unanchorable edits are skipped, never force-applied.
+    if (isEditingStage(stage)) {
+      const stageResult = getState().proForge.currentRun?.stages.find((s) => s.stage === stage);
+      const project = getState().project.present?.data;
+      if (stageResult && project) {
+        const acceptedIds = new Set(
+          decisions.filter((d) => d.status === 'accepted').map((d) => d.itemId),
+        );
+        const acceptedItems = stageResult.reviewItems.filter((ri) => acceptedIds.has(ri.id));
+        const { updates, applied, skipped } = planAcceptedManuscriptEdits(
+          project.manuscript,
+          acceptedItems,
+        );
+        if (updates.length > 0) {
+          const { projectActions } = await import('../../features/project/projectSlice');
+          for (const update of updates) {
+            dispatch(
+              projectActions.updateManuscriptSection({
+                id: update.id,
+                changes: { content: update.content },
+              }),
+            );
+          }
+        }
+        if (skipped > 0) {
+          logger.warn(
+            `ProForge submitReview: stage ${stage} applied ${applied} edit(s), skipped ${skipped} stale/unanchorable edit(s).`,
+          );
+        }
+      }
+    }
+
+    // Create post-stage snapshot (now reflecting any applied edits).
     const project = getState().project.present?.data;
     if (project) {
       const { versionControlActions } = await import(
@@ -272,9 +339,7 @@ export class ProForgeOrchestrator {
         }),
       );
     }
-    const vcState = getState().versionControl;
-    const currentBranch = vcState.branches.find((b) => b.id === vcState.currentBranchId);
-    const postSnapshotId = currentBranch?.headSnapshotId;
+    const postSnapshotId = this.headSnapshotId();
 
     dispatch(
       submitStageReview({
@@ -336,6 +401,7 @@ export class ProForgeOrchestrator {
     }
 
     dispatch(pipelineAborted());
+    await this.persistHistory();
   }
 
   /**
diff --git a/tests/unit/lora/styleAdaptedWriterAgent.test.ts b/tests/unit/lora/styleAdaptedWriterAgent.test.ts
deleted file mode 100644
index 8bb300c4e..000000000
--- a/tests/unit/lora/styleAdaptedWriterAgent.test.ts
+++ /dev/null
@@ -1,173 +0,0 @@
-/**
- * Tests for services/proForge/pipelineAgents/styleAdaptedWriterAgent.ts
- * QNBS-v3: Fallback path (no active adapter) + suggestion path with mocked localAiFacade.
- */
-
-import { beforeEach, describe, expect, it, vi } from 'vitest';
-
-// ---------------------------------------------------------------------------
-// Mocks
-// ---------------------------------------------------------------------------
-
-vi.mock('../../../services/loraAdapterService', () => ({
-  getActiveAdapter: vi.fn(),
-}));
-
-vi.mock('../../../services/localAiFacade', () => ({
-  generateLocalText: vi.fn(),
-}));
-
-vi.mock('../../../services/proForge/proForgeMemoryBank', () => ({
-  getMemoryBank: vi.fn(() => ({
-    buildContextString: vi.fn().mockResolvedValue(''),
-    remember: vi.fn(),
-    recall: vi.fn().mockResolvedValue([]),
-  })),
-}));
-
-vi.mock('../../../services/logger', () => ({
-  logger: { debug: vi.fn(), info: vi.fn(), warn: vi.fn(), error: vi.fn() },
-}));
-
-vi.mock('../../../services/aiProviderService', () => ({
-  aiProviderService: { generateText: vi.fn() },
-}));
-
-// ---------------------------------------------------------------------------
-// Imports
-// ---------------------------------------------------------------------------
-
-import { generateLocalText } from '../../../services/localAiFacade';
-import { getActiveAdapter } from '../../../services/loraAdapterService';
-import { StyleAdaptedWriterAgent } from '../../../services/proForge/pipelineAgents/styleAdaptedWriterAgent';
-import type { OrchestratorContext } from '../../../services/proForge/proForgeOrchestrator';
-
-// ---------------------------------------------------------------------------
-// Fixtures
-// ---------------------------------------------------------------------------
-
-const ACTIVE_ADAPTER = {
-  id: 'lora-1',
-  name: 'HemingwayStyle',
-  description: '',
-  modelCompatibility: 'llama-3.2-7b',
-  scale: 1,
-  fileSizeBytes: 100,
-  createdAt: 1,
-  isActive: true,
-};
-
-/** Long enough content to pass the 200-char threshold in the agent (≥ 201 chars). */
-const LONG_CONTENT =
-  'The old lighthouse keeper had not left the island in forty years. ' +
-  'Each morning he climbed the spiral staircase and polished the great lens until it blazed. ' +
-  'Tonight a storm was drawing close across the black water and the fog horns wailed.';
-
-function makeContext(
-  sections = [{ id: 's1', title: 'Ch 1', content: LONG_CONTENT }],
-): OrchestratorContext {
-  return {
-    projectId: 'proj-1',
-    // biome-ignore lint/suspicious/noExplicitAny: test mock
-    dispatch: vi.fn() as any,
-    getState: vi.fn().mockReturnValue({
-      project: {
-        present: {
-          data: {
-            title: 'Test Novel',
-            logline: '',
-            manuscript: sections,
-            characters: { ids: [], entities: {} },
-            worlds: { ids: [], entities: {} },
-            outline: [],
-          },
-        },
-      },
-      // biome-ignore lint/suspicious/noExplicitAny: partial test state
-    } as any),
-    manuscript: sections,
-    characters: [],
-    worlds: [],
-    config: {
-      genrePreset: 'literary-fiction',
-      selectedStages: ['lineProse'],
-      aiProvider: 'webllm',
-      ragMode: 'hybrid',
-      maxTokens: 2000,
-      creativity: 'Balanced',
-      useDuckDb: false,
-      autoAcceptThreshold: 0,
-      language: 'en',
-    },
-  };
-}
-
-// ---------------------------------------------------------------------------
-// Tests
-// ---------------------------------------------------------------------------
-
-describe('StyleAdaptedWriterAgent — no active adapter', () => {
-  beforeEach(() => {
-    vi.mocked(getActiveAdapter).mockResolvedValue(null);
-  });
-
-  it('returns isFallback: true and empty reviewItems', async () => {
-    const agent = new StyleAdaptedWriterAgent(makeContext());
-    const result = await agent.execute(new AbortController().signal);
-
-    expect(result.reviewItems).toHaveLength(0);
-    expect(result.agentOutput).toMatchObject({ isFallback: true, reason: 'no_active_adapter' });
-    expect(result.metrics.aiCalls).toBe(0);
-  });
-});
-
-describe('StyleAdaptedWriterAgent — with active adapter', () => {
-  beforeEach(() => {
-    vi.clearAllMocks();
-    vi.mocked(getActiveAdapter).mockResolvedValue(ACTIVE_ADAPTER);
-    vi.mocked(generateLocalText).mockResolvedValue({
-      layer: 'webllm' as const,
-      text: 'The keeper gripped the railing as the storm surged closer, each wave a drumbeat of inevitability.',
-    });
-  });
-
-  it('produces at least one proseEdit reviewItem', async () => {
-    const agent = new StyleAdaptedWriterAgent(makeContext());
-    const result = await agent.execute(new AbortController().signal);
-
-    // Verify the mock was actually invoked via the dynamic import path
-    expect(vi.mocked(generateLocalText)).toHaveBeenCalled();
-    expect(result.reviewItems.length).toBeGreaterThan(0);
-    expect(result.reviewItems[0]!.type).toBe('proseEdit');
-    expect(result.reviewItems[0]!.sectionId).toBe('s1');
-    expect(result.agentOutput).not.toHaveProperty('isFallback');
-  });
-
-  it('reports adapter id and name in agentOutput', async () => {
-    const agent = new StyleAdaptedWriterAgent(makeContext());
-    const result = await agent.execute(new AbortController().signal);
-
-    expect(result.agentOutput).toMatchObject({
-      adapterId: 'lora-1',
-      adapterName: 'HemingwayStyle',
-    });
-  });
-
-  it('skips sections shorter than 200 characters', async () => {
-    const agent = new StyleAdaptedWriterAgent(
-      makeContext([{ id: 's1', title: 'Short', content: 'Too short.' }]),
-    );
-    const result = await agent.execute(new AbortController().signal);
-    expect(result.reviewItems).toHaveLength(0);
-    expect(vi.mocked(generateLocalText)).not.toHaveBeenCalled();
-  });
-
-  it('returns isFallback: true when signal is already aborted', async () => {
-    const ctrl = new AbortController();
-    ctrl.abort();
-    const agent = new StyleAdaptedWriterAgent(makeContext());
-    const result = await agent.execute(ctrl.signal);
-
-    expect(result.agentOutput).toMatchObject({ isFallback: true, reason: 'aborted' });
-  });
-});
diff --git a/tests/unit/proForge/applyReviewEdits.test.ts b/tests/unit/proForge/applyReviewEdits.test.ts
new file mode 100644
index 000000000..edadf685e
--- /dev/null
+++ b/tests/unit/proForge/applyReviewEdits.test.ts
@@ -0,0 +1,127 @@
+import { describe, expect, it } from 'vitest';
+import type { ReviewItem } from '../../../features/proForge/types';
+import {
+  applyReviewEditsToSection,
+  planAcceptedManuscriptEdits,
+} from '../../../services/proForge/applyReviewEdits';
+
+// QNBS-v3: Minimal ReviewItem factory — only the fields the applier reads.
+function item(partial: Partial<ReviewItem>): ReviewItem {
+  return {
+    id: partial.id ?? 'r1',
+    stage: 'lineProse',
+    type: 'proseEdit',
+    severity: 'info',
+    description: 'd',
+    confidence: 0.9,
+    status: 'accepted',
+    createdAt: '2026-01-01T00:00:00.000Z',
+    ...partial,
+  };
+}
+
+describe('applyReviewEditsToSection', () => {
+  it('applies a single range-based edit verified against original', () => {
+    const content = 'The quick brown fox.';
+    const result = applyReviewEditsToSection(content, [
+      item({ range: { start: 4, end: 9 }, original: 'quick', proposed: 'slow' }),
+    ]);
+    expect(result.content).toBe('The slow brown fox.');
+    expect(result.applied).toBe(1);
+    expect(result.skipped).toBe(0);
+  });
+
+  it('applies multiple edits back-to-front so offsets stay valid', () => {
+    const content = 'aaa bbb ccc';
+    const result = applyReviewEditsToSection(content, [
+      item({ id: '1', range: { start: 0, end: 3 }, original: 'aaa', proposed: 'X' }),
+      item({ id: '2', range: { start: 8, end: 11 }, original: 'ccc', proposed: 'YYYYY' }),
+    ]);
+    expect(result.content).toBe('X bbb YYYYY');
+    expect(result.applied).toBe(2);
+  });
+
+  it('falls back to text match when the offset is stale', () => {
+    const content = 'hello wonderful world';
+    // Range points at the wrong place; original text still locatable.
+    const result = applyReviewEditsToSection(content, [
+      item({ range: { start: 0, end: 5 }, original: 'wonderful', proposed: 'cruel' }),
+    ]);
+    expect(result.content).toBe('hello cruel world');
+    expect(result.applied).toBe(1);
+    expect(result.skipped).toBe(0);
+  });
+
+  it('skips an edit whose original text cannot be found', () => {
+    const content = 'nothing to see here';
+    const result = applyReviewEditsToSection(content, [
+      item({ original: 'absent phrase', proposed: 'x' }),
+    ]);
+    expect(result.content).toBe(content);
+    expect(result.applied).toBe(0);
+    expect(result.skipped).toBe(1);
+  });
+
+  it('ignores advisory items with no proposed replacement (not counted as skipped)', () => {
+    const content = 'unchanged';
+    const result = applyReviewEditsToSection(content, [
+      item({ type: 'pacingIssue', description: 'advisory only' }),
+    ]);
+    expect(result.content).toBe('unchanged');
+    expect(result.applied).toBe(0);
+    expect(result.skipped).toBe(0);
+  });
+
+  it('drops overlapping edits rather than corrupting the text', () => {
+    const content = 'abcdef';
+    const result = applyReviewEditsToSection(content, [
+      item({ id: '1', range: { start: 1, end: 4 }, proposed: 'XX' }),
+      item({ id: '2', range: { start: 2, end: 5 }, proposed: 'YY' }),
+    ]);
+    // Right-most (start=2) applies first; the start=1 edit overlaps it and is skipped.
+    expect(result.applied).toBe(1);
+    expect(result.skipped).toBe(1);
+  });
+});
+
+describe('planAcceptedManuscriptEdits', () => {
+  const manuscript = [
+    { id: 's1', content: 'The quick brown fox.' },
+    { id: 's2', content: 'Jumped over the lazy dog.' },
+  ];
+
+  it('returns only sections that changed, with aggregate counts', () => {
+    const { updates, applied, skipped } = planAcceptedManuscriptEdits(manuscript, [
+      item({
+        id: 'a',
+        sectionId: 's1',
+        range: { start: 4, end: 9 },
+        original: 'quick',
+        proposed: 'sly',
+      }),
+      item({ id: 'b', sectionId: 's2', original: 'lazy', proposed: 'sleepy' }),
+    ]);
+    expect(updates).toHaveLength(2);
+    expect(updates.find((u) => u.id === 's1')?.content).toBe('The sly brown fox.');
+    expect(updates.find((u) => u.id === 's2')?.content).toBe('Jumped over the sleepy dog.');
+    expect(applied).toBe(2);
+    expect(skipped).toBe(0);
+  });
+
+  it('counts edits for a deleted section as skipped', () => {
+    const { updates, applied, skipped } = planAcceptedManuscriptEdits(manuscript, [
+      item({ id: 'a', sectionId: 'ghost', original: 'x', proposed: 'y' }),
+    ]);
+    expect(updates).toHaveLength(0);
+    expect(applied).toBe(0);
+    expect(skipped).toBe(1);
+  });
+
+  it('ignores items without a sectionId or proposed text', () => {
+    const { updates } = planAcceptedManuscriptEdits(manuscript, [
+      item({ id: 'a', proposed: 'orphan' }),
+      item({ id: 'b', sectionId: 's1', type: 'pacingIssue' }),
+    ]);
+    expect(updates).toHaveLength(0);
+  });
+});
diff --git a/tests/unit/proForge/components/PipelineReviewPanel.test.tsx b/tests/unit/proForge/components/PipelineReviewPanel.test.tsx
index 7a230b05e..bc542700c 100644
--- a/tests/unit/proForge/components/PipelineReviewPanel.test.tsx
+++ b/tests/unit/proForge/components/PipelineReviewPanel.test.tsx
@@ -12,6 +12,8 @@ import type {
   ReviewItemType,
   StageStatus,
 } from '../../../../features/proForge/types';
+// QNBS-v3: Real English strings so assertions read naturally; t() resolves keys + {{vars}}.
+import mockEnCommon from '../../../../locales/en/common.json';
 
 // ---------------------------------------------------------------------------
 // Mocks
@@ -144,6 +146,24 @@ vi.mock('../../../../features/proForge/proForgeSlice', () => ({
   },
 }));
 
+vi.mock('../../../../hooks/useTranslation', () => ({
+  useTranslation: () => ({
+    t: (key: string, vars?: Record<string, string | number>) => {
+      const dict = mockEnCommon as Record<string, string>;
+      let s = dict[key] ?? key;
+      if (vars) {
+        for (const [k, v] of Object.entries(vars)) s = s.replace(`{{${k}}}`, String(v));
+      }
+      return s;
+    },
+    language: 'en',
+  }),
+}));
+
+vi.mock('../../../../contexts/LiveRegionContext', () => ({
+  useAnnounce: () => vi.fn(),
+}));
+
 // ---------------------------------------------------------------------------
 // Imports after mocks
 // ---------------------------------------------------------------------------
@@ -189,7 +209,8 @@ describe('PipelineReviewPanel', () => {
   describe('header', () => {
     it('shows stage name in header', () => {
       render(<PipelineReviewPanel />);
-      expect(screen.getByText(/Review: intake/i)).toBeInTheDocument();
+      // Stage id 'intake' is localised to its display name in the header.
+      expect(screen.getByText(/Review: Intake/i)).toBeInTheDocument();
     });
 
     it('shows correct pending/accepted/rejected counts', () => {
@@ -425,8 +446,8 @@ describe('PipelineReviewPanel', () => {
         currentStageReviewItems: [makeItem('item-1', 'pending', { severity: 'critical' })],
       });
       render(<PipelineReviewPanel />);
-      // SEVERITY_ICONS.critical = '🔴' in ReviewItemCard; summary card shows longer string
-      expect(screen.getByText('🔴')).toBeInTheDocument();
+      // 🔴 appears in the item card and (for critical items) the summary card; both are aria-hidden.
+      expect(screen.getAllByText('🔴').length).toBeGreaterThanOrEqual(1);
     });
 
     it('shows sectionTitle when present', () => {
diff --git a/tests/unit/proForge/pipelineAgents/baseAgent.test.ts b/tests/unit/proForge/pipelineAgents/baseAgent.test.ts
index 7cf0de228..f5da84252 100644
--- a/tests/unit/proForge/pipelineAgents/baseAgent.test.ts
+++ b/tests/unit/proForge/pipelineAgents/baseAgent.test.ts
@@ -275,6 +275,24 @@ describe('BaseAgent', () => {
         expect.objectContaining({ options: expect.objectContaining({ maxTokens: 4000 }) }),
       );
     });
+
+    it('does not alter the prompt when no retry feedback is set', async () => {
+      mockGenerate.mockResolvedValueOnce({ text: 'OK', usage: {} });
+      await agent.publicGenerate('Original prompt.');
+      expect(mockGenerate).toHaveBeenCalledWith(
+        expect.objectContaining({ prompt: 'Original prompt.' }),
+      );
+    });
+
+    it('prepends supervisor feedback to the prompt after setRetryFeedback', async () => {
+      mockGenerate.mockResolvedValueOnce({ text: 'Corrected.', usage: {} });
+      agent.setRetryFeedback('- No structural edits found.');
+      await agent.publicGenerate('Original prompt.');
+      const callArg = mockGenerate.mock.calls.at(-1)?.[0] as { prompt: string };
+      expect(callArg.prompt).toContain('previous attempt was rejected');
+      expect(callArg.prompt).toContain('No structural edits found.');
+      expect(callArg.prompt).toContain('Original prompt.');
+    });
   });
 
   describe('selfReflect()', () => {
diff --git a/tests/unit/proForge/pipelineAgents/supervisorAgent.test.ts b/tests/unit/proForge/pipelineAgents/supervisorAgent.test.ts
index 3247e431e..4d9d0ea74 100644
--- a/tests/unit/proForge/pipelineAgents/supervisorAgent.test.ts
+++ b/tests/unit/proForge/pipelineAgents/supervisorAgent.test.ts
@@ -66,7 +66,7 @@ const REAL_QUALITY_SCORE = {
   marketability: 77,
 };
 
-// Fallback sentinel: all 3 checked dimensions are exactly 50.
+// Uniform 50 scores — a plausible-but-mediocre real analysis, NOT a fallback by itself.
 const FALLBACK_QUALITY_SCORE = {
   overall: 50,
   prose: 50,
@@ -77,6 +77,17 @@ const FALLBACK_QUALITY_SCORE = {
   marketability: 50,
 };
 
+// Zeroed scores accompany a genuine fallback report (isFallback: true).
+const ZERO_QUALITY_SCORE = {
+  overall: 0,
+  prose: 0,
+  structure: 0,
+  consistency: 0,
+  pacing: 0,
+  dialogue: 0,
+  marketability: 0,
+};
+
 // ---------------------------------------------------------------------------
 // Tests: evaluate() dispatch
 // ---------------------------------------------------------------------------
@@ -88,21 +99,109 @@ describe('SupervisorAgent', () => {
     agent = new SupervisorAgent(makeContext());
   });
 
-  describe('evaluate() — unknown/default stages', () => {
-    it('passes with score 100 for stages without specific logic', () => {
-      const result = agent.evaluate('lineProse', { reviewItems: [], agentOutput: undefined });
+  describe('evaluate() — default/terminal stages', () => {
+    it('passes with score 100 for idle/archived stages (no specific logic)', () => {
+      const idle = agent.evaluate('idle', { reviewItems: [], agentOutput: undefined });
+      expect(idle.pass).toBe(true);
+      expect(idle.qualityScore).toBe(100);
+      expect(idle.reasons).toHaveLength(0);
+      expect(agent.evaluate('archived', { reviewItems: [], agentOutput: undefined }).pass).toBe(
+        true,
+      );
+    });
+  });
+
+  describe('evaluateLineProse', () => {
+    it('passes with score 85 when prose edits are present', () => {
+      const result = agent.evaluate('lineProse', { reviewItems: [], agentOutput: { edits: [{}] } });
       expect(result.pass).toBe(true);
-      expect(result.qualityScore).toBe(100);
-      expect(result.reasons).toHaveLength(0);
+      expect(result.qualityScore).toBe(85);
+    });
+
+    it('passes for a short manuscript even with zero edits', () => {
+      const result = agent.evaluate('lineProse', { reviewItems: [], agentOutput: { edits: [] } });
+      expect(result.pass).toBe(true);
+    });
+
+    it('fails for a long manuscript with zero prose edits and no review items', () => {
+      const longAgent = new SupervisorAgent(makeContext('word '.repeat(1100)));
+      const result = longAgent.evaluate('lineProse', {
+        reviewItems: [],
+        agentOutput: { edits: [] },
+      });
+      expect(result.pass).toBe(false);
+      expect(result.retryRecommended).toBe(true);
+    });
+  });
+
+  describe('evaluateCopyEdit', () => {
+    it('passes when copy-edit findings exist', () => {
+      const result = agent.evaluate('copyEdit', {
+        reviewItems: [],
+        agentOutput: { grammarEdits: [{}], styleEdits: [], repetitionHits: [], formatIssues: [] },
+      });
+      expect(result.pass).toBe(true);
+    });
+
+    it('fails for a long manuscript with zero findings', () => {
+      const longAgent = new SupervisorAgent(makeContext('word '.repeat(1600)));
+      const result = longAgent.evaluate('copyEdit', {
+        reviewItems: [],
+        agentOutput: { grammarEdits: [], styleEdits: [], repetitionHits: [], formatIssues: [] },
+      });
+      expect(result.pass).toBe(false);
+      expect(result.retryRecommended).toBe(true);
+    });
+  });
+
+  describe('evaluateProduction', () => {
+    it('passes with score 95 when at least one artifact exists', () => {
+      const result = agent.evaluate('production', {
+        reviewItems: [],
+        agentOutput: { artifacts: [{}] },
+      });
+      expect(result.pass).toBe(true);
+      expect(result.qualityScore).toBe(95);
+    });
+
+    it('fails when no artifacts were produced', () => {
+      const result = agent.evaluate('production', {
+        reviewItems: [],
+        agentOutput: { artifacts: [] },
+      });
+      expect(result.pass).toBe(false);
+      expect(result.reasons.some((r) => r.includes('artifact'))).toBe(true);
+    });
+  });
+
+  describe('evaluatePublishing', () => {
+    it('passes when title + back-cover blurb are present', () => {
+      const result = agent.evaluate('publishing', {
+        reviewItems: [],
+        agentOutput: { metadata: { title: 'My Book' }, blurbs: { backCover: 'A gripping tale.' } },
+      });
+      expect(result.pass).toBe(true);
+    });
+
+    it('fails when title/blurb are missing', () => {
+      const result = agent.evaluate('publishing', {
+        reviewItems: [],
+        agentOutput: { metadata: { title: '' }, blurbs: { backCover: '' } },
+      });
+      expect(result.pass).toBe(false);
+      expect(result.reasons.length).toBeGreaterThanOrEqual(1);
     });
+  });
 
-    it('passes for copyEdit stage', () => {
-      const result = agent.evaluate('copyEdit', { reviewItems: [], agentOutput: undefined });
+  describe('evaluateAnalytics', () => {
+    it('passes with score 100 when metrics exist', () => {
+      const result = agent.evaluate('analytics', { reviewItems: [], agentOutput: { metrics: {} } });
       expect(result.pass).toBe(true);
+      expect(result.qualityScore).toBe(100);
     });
 
-    it('passes for publishing stage', () => {
-      const result = agent.evaluate('publishing', { reviewItems: [], agentOutput: undefined });
+    it('never blocks the pipeline even without metrics', () => {
+      const result = agent.evaluate('analytics', { reviewItems: [], agentOutput: undefined });
       expect(result.pass).toBe(true);
     });
   });
@@ -126,11 +225,14 @@ describe('SupervisorAgent', () => {
       expect(result.qualityScore).toBe(72);
     });
 
-    it('fails and recommends retry when all 3 sentinel dimensions are 50', () => {
+    it('fails and recommends retry when the report is marked isFallback', () => {
+      // QNBS-v3: Real fallback reports carry isFallback:true with zeroed scores
+      // (see DiagnosticAgent.createFallbackReport). Detection keys off the flag, not score===50.
       const result = agent.evaluate('intake', {
         reviewItems: [],
         agentOutput: {
-          qualityScore: FALLBACK_QUALITY_SCORE,
+          isFallback: true,
+          qualityScore: ZERO_QUALITY_SCORE,
           consistencyIssues: [],
           structuralGaps: [],
         },
@@ -141,6 +243,19 @@ describe('SupervisorAgent', () => {
       expect(result.reasons.some((r) => r.includes('fallback'))).toBe(true);
     });
 
+    it('does NOT treat uniform 50 scores as fallback without the isFallback flag', () => {
+      const result = agent.evaluate('intake', {
+        reviewItems: [],
+        agentOutput: {
+          qualityScore: FALLBACK_QUALITY_SCORE,
+          consistencyIssues: [{ id: 'ci-1' }],
+          structuralGaps: [],
+        },
+      });
+      // No isFallback flag → treated as a (low but real) analysis, passes the gate.
+      expect(result.pass).toBe(true);
+    });
+
     it('does NOT treat partial 50-scores as fallback (only all-3-at-50 triggers)', () => {
       const result = agent.evaluate('intake', {
         reviewItems: [],
diff --git a/tests/unit/proForge/proForgeHistoryStore.test.ts b/tests/unit/proForge/proForgeHistoryStore.test.ts
new file mode 100644
index 000000000..18433cb96
--- /dev/null
+++ b/tests/unit/proForge/proForgeHistoryStore.test.ts
@@ -0,0 +1,70 @@
+// @vitest-environment node
+// QNBS-v3: node environment + fresh IDBFactory per test for full isolation (mirrors memory-bank test).
+
+import { IDBFactory, IDBKeyRange } from 'fake-indexeddb';
+import { afterEach, beforeEach, describe, expect, it } from 'vitest';
+import type { PipelineRun } from '../../../features/proForge/types';
+import {
+  _resetHistoryDbForTest,
+  loadRunHistory,
+  MAX_RUN_HISTORY,
+  saveRunHistory,
+} from '../../../services/proForge/proForgeHistoryStore';
+
+beforeEach(() => {
+  global.indexedDB = new IDBFactory();
+  global.IDBKeyRange = IDBKeyRange;
+  _resetHistoryDbForTest();
+});
+
+afterEach(() => {
+  _resetHistoryDbForTest();
+});
+
+function run(id: string): PipelineRun {
+  return {
+    id,
+    projectId: 'p1',
+    label: id,
+    config: {} as PipelineRun['config'],
+    status: 'completed',
+    activeStage: 'archived',
+    stages: [],
+    startedAt: '2026-01-01T00:00:00.000Z',
+    prePipelineSnapshotId: 'snap-1',
+    traceLog: [],
+  };
+}
+
+describe('proForgeHistoryStore', () => {
+  it('returns an empty array when nothing is persisted', async () => {
+    expect(await loadRunHistory('p1')).toEqual([]);
+  });
+
+  it('round-trips run history for a project', async () => {
+    await saveRunHistory('p1', [run('a'), run('b')]);
+    const loaded = await loadRunHistory('p1');
+    expect(loaded.map((r) => r.id)).toEqual(['a', 'b']);
+  });
+
+  it('scopes history by project id', async () => {
+    await saveRunHistory('p1', [run('a')]);
+    await saveRunHistory('p2', [run('z')]);
+    expect((await loadRunHistory('p2')).map((r) => r.id)).toEqual(['z']);
+    expect((await loadRunHistory('p1')).map((r) => r.id)).toEqual(['a']);
+  });
+
+  it('caps stored history to MAX_RUN_HISTORY (most-recent-first)', async () => {
+    const many = Array.from({ length: MAX_RUN_HISTORY + 5 }, (_, i) => run(`r${i}`));
+    await saveRunHistory('p1', many);
+    const loaded = await loadRunHistory('p1');
+    expect(loaded).toHaveLength(MAX_RUN_HISTORY);
+    expect(loaded[0]?.id).toBe('r0');
+  });
+
+  it('overwrites prior history for the same project', async () => {
+    await saveRunHistory('p1', [run('old')]);
+    await saveRunHistory('p1', [run('new')]);
+    expect((await loadRunHistory('p1')).map((r) => r.id)).toEqual(['new']);
+  });
+});
diff --git a/tests/unit/proForge/proForgeMemoryBank.test.ts b/tests/unit/proForge/proForgeMemoryBank.test.ts
index 944e45bac..7683d7f96 100644
--- a/tests/unit/proForge/proForgeMemoryBank.test.ts
+++ b/tests/unit/proForge/proForgeMemoryBank.test.ts
@@ -3,7 +3,25 @@
 //          Fresh IDBFactory + _resetDbForTest() ensures complete test isolation.
 
 import { IDBFactory, IDBKeyRange } from 'fake-indexeddb';
-import { afterEach, beforeEach, describe, expect, it } from 'vitest';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+// QNBS-v3: Deterministic 4-dim embeddings over a tiny vocabulary so semantic ranking is testable
+// without loading the real MiniLM model. searchMemoryEntries dynamically imports this module.
+vi.mock('../../../services/ai/localEmbeddingService', () => {
+  const VOCAB = ['dragon', 'sea', 'hero', 'magic'];
+  const embed = (text: string): number[] => {
+    const lower = text.toLowerCase();
+    const v = VOCAB.map((w) => (lower.match(new RegExp(w, 'g')) ?? []).length);
+    const norm = Math.sqrt(v.reduce((a, b) => a + b * b, 0)) || 1;
+    return v.map((x) => x / norm);
+  };
+  return {
+    embedText: vi.fn(async (t: string) => embed(t)),
+    cosineSimilarity: (a: number[], b: number[]) => a.reduce((s, x, i) => s + x * (b[i] ?? 0), 0),
+    embedBatch: vi.fn(),
+  };
+});
+
 import {
   _resetDbForTest,
   clearMemoryBankCache,
@@ -217,6 +235,42 @@ describe('searchMemoryEntries', () => {
   });
 });
 
+describe('searchMemoryEntries — semantic & hybrid modes', () => {
+  beforeEach(async () => {
+    await saveMemoryEntry({
+      projectId: 'p2',
+      category: 'lore',
+      key: 'k1',
+      content: 'The dragon guards the mountain.',
+      sourceStage: 'intake',
+    });
+    await saveMemoryEntry({
+      projectId: 'p2',
+      category: 'lore',
+      key: 'k2',
+      content: 'The hero sails the sea.',
+      sourceStage: 'intake',
+    });
+  });
+
+  it('ranks by embedding similarity in semantic mode', async () => {
+    const results = await searchMemoryEntries('p2', 'dragon', 10, 'semantic');
+    expect(results[0]?.key).toBe('k1');
+  });
+
+  it('blends keyword + semantic in hybrid mode', async () => {
+    const results = await searchMemoryEntries('p2', 'hero sea', 10, 'hybrid');
+    expect(results[0]?.key).toBe('k2');
+  });
+
+  it('falls back to keyword ranking when embedding fails', async () => {
+    const svc = await import('../../../services/ai/localEmbeddingService');
+    vi.mocked(svc.embedText).mockRejectedValueOnce(new Error('no model'));
+    const results = await searchMemoryEntries('p2', 'dragon', 10, 'hybrid');
+    expect(results.some((r) => r.key === 'k1')).toBe(true);
+  });
+});
+
 // ---------------------------------------------------------------------------
 // deleteMemoryEntry
 // ---------------------------------------------------------------------------
diff --git a/tests/unit/proForge/proForgeOrchestrator.test.ts b/tests/unit/proForge/proForgeOrchestrator.test.ts
index 8d0e92d82..dd5175731 100644
--- a/tests/unit/proForge/proForgeOrchestrator.test.ts
+++ b/tests/unit/proForge/proForgeOrchestrator.test.ts
@@ -33,48 +33,56 @@ vi.mock('../../../services/proForge/pipelineAgents/diagnosticAgent', () => ({
   DiagnosticAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/structuralAgent', () => ({
   StructuralAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/proseAgent', () => ({
   ProseAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/copyEditAgent', () => ({
   CopyEditAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/proofAgent', () => ({
   ProofAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/productionAgent', () => ({
   ProductionAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/publishingAgent', () => ({
   PublishingAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 vi.mock('../../../services/proForge/pipelineAgents/analyticsAgent', () => ({
   AnalyticsAgent: class {
     constructor(public ctx: unknown) {}
     execute = mockAgentExecute;
+    setRetryFeedback = vi.fn();
   },
 }));
 
@@ -100,6 +108,11 @@ vi.mock('../../../features/proForge/proForgeSlice', () => ({
   proForgeActions: {},
 }));
 
+// Mock run-history persistence dynamic import
+vi.mock('../../../services/proForge/proForgeHistoryStore', () => ({
+  saveRunHistory: vi.fn().mockResolvedValue(undefined),
+}));
+
 // ---------------------------------------------------------------------------
 // Import after mocks
 // ---------------------------------------------------------------------------
@@ -266,6 +279,35 @@ describe('ProForgeOrchestrator', () => {
         'No project data available',
       );
     });
+
+    it('resets the abort signal so a re-run after dispose still completes the stage', async () => {
+      const ctx = makeContext({
+        currentRun: {
+          id: 'run-1',
+          status: 'running',
+          stages: [],
+          config: { ...DEFAULT_CONFIG, selectedStages: ['intake'], maxRetries: 0 },
+          label: 'Re-run',
+        },
+        isRunning: true,
+      });
+      const orch = new ProForgeOrchestrator(ctx);
+      // Simulate a prior abort — the constructor-created controller is now aborted.
+      orch.dispose();
+
+      await orch.startPipeline('Re-run', {
+        ...DEFAULT_CONFIG,
+        selectedStages: ['intake'],
+        maxRetries: 0,
+      });
+
+      const { stageCompleted, stageFailed } = await import(
+        '../../../features/proForge/proForgeSlice'
+      );
+      // Without the reset, the stale aborted signal would trip the post-execute guard → stageFailed.
+      expect(vi.mocked(stageCompleted)).toHaveBeenCalled();
+      expect(vi.mocked(stageFailed)).not.toHaveBeenCalled();
+    });
   });
 
   describe('executeStage', () => {
@@ -449,6 +491,140 @@ describe('ProForgeOrchestrator', () => {
     });
   });
 
+  describe('submitReview — applying accepted edits to the manuscript', () => {
+    const editItem = {
+      id: 'e1',
+      stage: 'lineProse',
+      type: 'proseEdit',
+      severity: 'info',
+      sectionId: 's1',
+      range: { start: 0, end: 5 },
+      original: 'Hello',
+      proposed: 'Hi',
+      description: 'Tighten greeting',
+      confidence: 0.9,
+      status: 'accepted',
+      createdAt: '2026-01-01T00:00:00.000Z',
+    };
+
+    function findUpdate(ctx: OrchestratorContext) {
+      const calls = (ctx.dispatch as unknown as { mock: { calls: unknown[][] } }).mock.calls;
+      return calls
+        .map((c) => c[0] as { payload?: { id?: string; changes?: { content?: string } } })
+        .find((a) => a?.payload?.changes?.content !== undefined);
+    }
+
+    it('writes accepted edits into the section content (editing stage)', async () => {
+      const ctx = makeContext({
+        currentRun: {
+          id: 'run-1',
+          status: 'awaitingReview',
+          activeStage: 'lineProse',
+          label: 'Edit Test',
+          config: DEFAULT_CONFIG,
+          stages: [{ stage: 'lineProse', status: 'awaitingReview', reviewItems: [editItem] }],
+        },
+      });
+      const orch = new ProForgeOrchestrator(ctx);
+      await orch.submitReview('lineProse', [{ itemId: 'e1', status: 'accepted' }], {
+        advance: false,
+      });
+
+      const update = findUpdate(ctx);
+      expect(update?.payload?.id).toBe('s1');
+      expect(update?.payload?.changes?.content).toBe('Hi world.');
+    });
+
+    it('does not modify the manuscript for non-editing stages', async () => {
+      const ctx = makeContext({
+        currentRun: {
+          id: 'run-1',
+          status: 'awaitingReview',
+          activeStage: 'publishing',
+          label: 'Edit Test',
+          config: DEFAULT_CONFIG,
+          stages: [
+            {
+              stage: 'publishing',
+              status: 'awaitingReview',
+              reviewItems: [{ ...editItem, stage: 'publishing' }],
+            },
+          ],
+        },
+      });
+      const orch = new ProForgeOrchestrator(ctx);
+      await orch.submitReview('publishing', [{ itemId: 'e1', status: 'accepted' }], {
+        advance: false,
+      });
+
+      expect(findUpdate(ctx)).toBeUndefined();
+    });
+
+    it('does not apply rejected edits', async () => {
+      const ctx = makeContext({
+        currentRun: {
+          id: 'run-1',
+          status: 'awaitingReview',
+          activeStage: 'lineProse',
+          label: 'Edit Test',
+          config: DEFAULT_CONFIG,
+          stages: [{ stage: 'lineProse', status: 'awaitingReview', reviewItems: [editItem] }],
+        },
+      });
+      const orch = new ProForgeOrchestrator(ctx);
+      await orch.submitReview('lineProse', [{ itemId: 'e1', status: 'rejected' }], {
+        advance: false,
+      });
+
+      expect(findUpdate(ctx)).toBeUndefined();
+    });
+  });
+
+  describe('run-history persistence', () => {
+    it('persists run history to IDB on completion', async () => {
+      const ctx = makeContext({
+        currentRun: {
+          id: 'run-1',
+          status: 'running',
+          activeStage: 'analytics',
+          stages: [],
+          config: DEFAULT_CONFIG,
+          label: 'Done',
+        },
+        runHistory: [{ id: 'run-1', label: 'Done' }],
+      });
+      const orch = new ProForgeOrchestrator(ctx);
+      await orch.advanceToNextStage('analytics');
+
+      const { saveRunHistory } = await import('../../../services/proForge/proForgeHistoryStore');
+      expect(vi.mocked(saveRunHistory)).toHaveBeenCalledWith('p1', [
+        { id: 'run-1', label: 'Done' },
+      ]);
+    });
+
+    it('persists run history to IDB on abort', async () => {
+      const ctx = makeContext({
+        currentRun: {
+          id: 'run-1',
+          status: 'running',
+          activeStage: 'structural',
+          stages: [],
+          config: DEFAULT_CONFIG,
+          label: 'Aborted',
+          prePipelineSnapshotId: 'snap-pre',
+        },
+        runHistory: [{ id: 'run-1', label: 'Aborted' }],
+      });
+      const orch = new ProForgeOrchestrator(ctx);
+      await orch.abortPipeline();
+
+      const { saveRunHistory } = await import('../../../services/proForge/proForgeHistoryStore');
+      expect(vi.mocked(saveRunHistory)).toHaveBeenCalledWith('p1', [
+        { id: 'run-1', label: 'Aborted' },
+      ]);
+    });
+  });
+
   describe('rollbackTo', () => {
     it('dispatches rollbackToStage', async () => {
       const ctx = makeContext({
diff --git a/tests/unit/proForge/toolRegistry.test.ts b/tests/unit/proForge/toolRegistry.test.ts
deleted file mode 100644
index ddac2cc70..000000000
--- a/tests/unit/proForge/toolRegistry.test.ts
+++ /dev/null
@@ -1,333 +0,0 @@
-/**
- * Tests for ProForge ToolRegistry.
- * QNBS-v3: Pure class behaviour — mocks logger and memoryBank.
- */
-
-import { beforeEach, describe, expect, it, vi } from 'vitest';
-import { z } from 'zod';
-
-vi.mock('../../../services/logger', () => ({
-  logger: { debug: vi.fn(), info: vi.fn(), warn: vi.fn(), error: vi.fn() },
-}));
-
-// Import AFTER mocks are set up
-import {
-  type ToolContext,
-  type ToolDefinition,
-  ToolRegistry,
-  toolRegistry,
-} from '../../../services/proForge/pipelineTools/toolRegistry';
-
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-
-function makeTool(overrides: Partial<ToolDefinition> = {}): ToolDefinition {
-  return {
-    name: 'testTool',
-    description: 'A test tool',
-    parameters: z.object({ input: z.string() }),
-    handler: vi.fn().mockResolvedValue({ result: 'ok' }),
-    stages: ['intake'],
-    isWrite: false,
-    ...overrides,
-  };
-}
-
-function makeContext(): ToolContext {
-  return {
-    projectId: 'proj-1',
-    // biome-ignore lint/suspicious/noExplicitAny: test mock
-    dispatch: vi.fn() as any,
-    getState: vi.fn().mockReturnValue({
-      project: {
-        present: {
-          data: { manuscript: [], characters: { entities: {} }, worlds: { entities: {} } },
-        },
-      },
-      proForge: { currentRun: null },
-    }),
-    memoryBank: {
-      search: vi.fn().mockResolvedValue([]),
-      buildContextString: vi.fn().mockResolvedValue(''),
-      remember: vi.fn().mockResolvedValue({}),
-      recall: vi.fn().mockResolvedValue([]),
-      recallForStage: vi.fn().mockResolvedValue([]),
-      clear: vi.fn().mockResolvedValue(undefined),
-      // biome-ignore lint/suspicious/noExplicitAny: test mock
-    } as any,
-    signal: new AbortController().signal,
-  };
-}
-
-// ---------------------------------------------------------------------------
-// ToolRegistry class tests
-// ---------------------------------------------------------------------------
-
-describe('ToolRegistry', () => {
-  let registry: ToolRegistry;
-
-  beforeEach(() => {
-    registry = new ToolRegistry();
-  });
-
-  describe('register / get', () => {
-    it('registers a tool and retrieves it by name', () => {
-      const tool = makeTool({ name: 'myTool' });
-      registry.register(tool);
-      expect(registry.get('myTool')).toBe(tool);
-    });
-
-    it('returns undefined for unknown tool', () => {
-      expect(registry.get('nonExistent')).toBeUndefined();
-    });
-
-    it('overwrites an existing tool on re-register (warns)', async () => {
-      const { logger } = await import('../../../services/logger');
-      const tool1 = makeTool({ name: 'dup' });
-      const tool2 = makeTool({ name: 'dup', description: 'updated' });
-      registry.register(tool1);
-      registry.register(tool2);
-      expect(registry.get('dup')?.description).toBe('updated');
-      expect(vi.mocked(logger.warn)).toHaveBeenCalled();
-    });
-  });
-
-  describe('unregister', () => {
-    it('removes a registered tool and returns true', () => {
-      registry.register(makeTool({ name: 'toRemove' }));
-      expect(registry.unregister('toRemove')).toBe(true);
-      expect(registry.get('toRemove')).toBeUndefined();
-    });
-
-    it('returns false when tool does not exist', () => {
-      expect(registry.unregister('ghost')).toBe(false);
-    });
-  });
-
-  describe('list', () => {
-    it('returns empty array when no tools registered', () => {
-      expect(registry.list()).toHaveLength(0);
-    });
-
-    it('returns all registered tools', () => {
-      registry.register(makeTool({ name: 'a' }));
-      registry.register(makeTool({ name: 'b' }));
-      expect(registry.list()).toHaveLength(2);
-    });
-  });
-
-  describe('listForStage', () => {
-    beforeEach(() => {
-      registry.register(makeTool({ name: 'intakeTool', stages: ['intake'] }));
-      registry.register(makeTool({ name: 'structuralTool', stages: ['structural'] }));
-      registry.register(makeTool({ name: 'globalTool', stages: ['*'] }));
-    });
-
-    it('returns tools for the specified stage plus global tools', () => {
-      const tools = registry.listForStage('intake');
-      const names = tools.map((t) => t.name);
-      expect(names).toContain('intakeTool');
-      expect(names).toContain('globalTool');
-      expect(names).not.toContain('structuralTool');
-    });
-
-    it('returns only global tools for an unmatched stage', () => {
-      const tools = registry.listForStage('proof');
-      expect(tools.map((t) => t.name)).toEqual(['globalTool']);
-    });
-
-    it('returns nothing if no tools match', () => {
-      const emptyRegistry = new ToolRegistry();
-      emptyRegistry.register(makeTool({ name: 'intakeOnly', stages: ['intake'] }));
-      expect(emptyRegistry.listForStage('analytics')).toHaveLength(0);
-    });
-  });
-
-  describe('execute', () => {
-    it('calls the handler with validated args and returns the result', async () => {
-      const handler = vi.fn().mockResolvedValue({ done: true });
-      registry.register(makeTool({ name: 'execTool', handler }));
-      const ctx = makeContext();
-      const result = await registry.execute('execTool', { input: 'hello' }, ctx);
-      expect(result).toEqual({ done: true });
-      expect(handler).toHaveBeenCalledWith({ input: 'hello' }, ctx);
-    });
-
-    it('throws when tool is not registered', async () => {
-      await expect(registry.execute('missing', {}, makeContext())).rejects.toThrow(
-        'Tool "missing" not found',
-      );
-    });
-
-    it('throws on invalid arguments (schema mismatch)', async () => {
-      registry.register(makeTool({ name: 'strictTool' }));
-      await expect(registry.execute('strictTool', { input: 123 }, makeContext())).rejects.toThrow(
-        'Invalid arguments for tool',
-      );
-    });
-  });
-
-  describe('toAiSdkTools', () => {
-    it('returns tool descriptions and parameters for the stage', () => {
-      registry.register(
-        makeTool({ name: 'aiTool', stages: ['intake'], description: 'AI tool desc' }),
-      );
-      const sdkTools = registry.toAiSdkTools('intake');
-      expect(sdkTools['aiTool']).toBeDefined();
-      expect(sdkTools['aiTool']!.description).toBe('AI tool desc');
-      expect(sdkTools['aiTool']!.parameters).toBeDefined();
-    });
-
-    it('excludes tools not available in the stage', () => {
-      registry.register(makeTool({ name: 'proofOnly', stages: ['proof'] }));
-      const sdkTools = registry.toAiSdkTools('intake');
-      expect(sdkTools['proofOnly']).toBeUndefined();
-    });
-  });
-});
-
-// ---------------------------------------------------------------------------
-// Global toolRegistry singleton — built-in tools
-// ---------------------------------------------------------------------------
-
-describe('toolRegistry singleton (built-in tools)', () => {
-  it('has readSection registered', () => {
-    expect(toolRegistry.get('readSection')).toBeDefined();
-  });
-
-  it('has readAllSections registered', () => {
-    expect(toolRegistry.get('readAllSections')).toBeDefined();
-  });
-
-  it('has readProjectMeta registered', () => {
-    expect(toolRegistry.get('readProjectMeta')).toBeDefined();
-  });
-
-  it('has searchLore registered', () => {
-    expect(toolRegistry.get('searchLore')).toBeDefined();
-  });
-
-  it('has analyzePacing registered (intake/structural only)', () => {
-    const tool = toolRegistry.get('analyzePacing');
-    expect(tool).toBeDefined();
-    expect(tool?.stages).toContain('intake');
-    expect(tool?.stages).toContain('structural');
-  });
-
-  it('has countWords registered', () => {
-    expect(toolRegistry.get('countWords')).toBeDefined();
-  });
-
-  it('has generateReport registered', () => {
-    expect(toolRegistry.get('generateReport')).toBeDefined();
-  });
-
-  it('has getMemoryContext registered', () => {
-    expect(toolRegistry.get('getMemoryContext')).toBeDefined();
-  });
-
-  it('has proposeEdit registered (write tool)', () => {
-    const tool = toolRegistry.get('proposeEdit');
-    expect(tool).toBeDefined();
-    expect(tool?.isWrite).toBe(true);
-  });
-
-  it('readSection handler returns section data', async () => {
-    const ctx = makeContext();
-    ctx.getState = vi.fn().mockReturnValue({
-      project: {
-        present: {
-          data: {
-            manuscript: [
-              { id: 's1', title: 'Ch 1', content: 'Hello world', status: 'draft', act: 1 },
-            ],
-            characters: { entities: {} },
-            worlds: { entities: {} },
-          },
-        },
-      },
-      proForge: { currentRun: null },
-    });
-    const result = (await toolRegistry.execute('readSection', { sectionId: 's1' }, ctx)) as Record<
-      string,
-      unknown
-    >;
-    expect(result['id']).toBe('s1');
-    expect(result['title']).toBe('Ch 1');
-    expect(result['content']).toBe('Hello world');
-  });
-
-  it('readSection handler returns error for unknown section', async () => {
-    const ctx = makeContext();
-    const result = (await toolRegistry.execute(
-      'readSection',
-      { sectionId: 'ghost' },
-      ctx,
-    )) as Record<string, unknown>;
-    expect(result['error']).toContain('"ghost" not found');
-  });
-
-  it('countWords handler returns total word count', async () => {
-    const ctx = makeContext();
-    ctx.getState = vi.fn().mockReturnValue({
-      project: {
-        present: {
-          data: {
-            manuscript: [
-              { id: 's1', title: 'Ch 1', content: 'one two three' },
-              { id: 's2', title: 'Ch 2', content: 'four five' },
-            ],
-          },
-        },
-      },
-      proForge: { currentRun: null },
-    });
-    const result = (await toolRegistry.execute('countWords', {}, ctx)) as Record<string, unknown>;
-    expect(result['total']).toBe(5);
-  });
-
-  it('readAllSections returns metadata without content by default', async () => {
-    const ctx = makeContext();
-    ctx.getState = vi.fn().mockReturnValue({
-      project: {
-        present: {
-          data: {
-            manuscript: [
-              { id: 's1', title: 'Ch 1', content: 'alpha beta', status: 'draft', act: 1 },
-            ],
-          },
-        },
-      },
-      proForge: { currentRun: null },
-    });
-    const result = (await toolRegistry.execute('readAllSections', {}, ctx)) as Array<
-      Record<string, unknown>
-    >;
-    expect(Array.isArray(result)).toBe(true);
-    expect(result[0]!['content']).toBeUndefined();
-    expect(result[0]!['wordCount']).toBe(2);
-  });
-
-  it('analyzePacing categorizes sections by word count relative to average', async () => {
-    const ctx = makeContext();
-    // Section 1 has 3x the words of section 2 → should be flagged as slow
-    ctx.getState = vi.fn().mockReturnValue({
-      project: {
-        present: {
-          data: {
-            manuscript: [
-              { id: 's1', title: 'Long', content: 'word '.repeat(300) },
-              { id: 's2', title: 'Short', content: 'a b c d e' },
-            ],
-          },
-        },
-      },
-      proForge: { currentRun: null },
-    });
-    const result = (await toolRegistry.execute('analyzePacing', {}, ctx)) as {
-      sections: Array<{ recommendedAction: string }>;
-    };
-    expect(result.sections[0]!.recommendedAction).toBe('compress');
-  });
-});

From 50aefb84e586e346f4a5d3f5dc4ce97686af8e75 Mon Sep 17 00:00:00 2001
From: qnbs <155236708+qnbs@users.noreply.github.com>
Date: Wed, 10 Jun 2026 09:00:23 +0200
Subject: [PATCH 2/3] fix(proforge): coerce stored embedding to Float32Array;
 align local typecheck with CI
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

CI Quality Gate caught a type error the local typecheck missed:
proForgeMemoryBank.ts passed a `number[]` (persisted embedding) where `cosineSimilarity`
expects `EmbeddingVector` (Float32Array). Coerce stored embeddings via `new Float32Array(...)`.

Root cause of the local miss: `pnpm run typecheck` ran tsgo WITHOUT `--checkers 4`, while CI
runs `--checkers 4`; the single-checker path did not surface the dynamic-import type error.
Align the canonical `typecheck` script to the exact CI command so nothing slips past the gate:
- `typecheck` → `tsgo --project tsconfig.tsgo.json --noEmit --checkers 4` (matches ci.yml)
- add `typecheck:single` for the lighter single-checker variant (not gate-authoritative)
- document the exact command in CLAUDE.md

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 CLAUDE.md                               | 2 +-
 package.json                            | 3 ++-
 services/proForge/proForgeMemoryBank.ts | 7 ++++++-
 3 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/CLAUDE.md b/CLAUDE.md
index 23a82dad3..def38a27f 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -22,7 +22,7 @@ pnpm run build         # Production build to dist/
 pnpm run smoke:prod    # Headless mount check on dist/ (run AFTER build; catches prod-only crashes)
 pnpm run lint          # Biome lint (--error-on-warnings — warnings fail like CI)
 pnpm run lint:fix      # Biome auto-fix (lint + format)
-pnpm run typecheck     # TypeScript type check (tsgo --noEmit)
+pnpm run typecheck     # TypeScript type check — EXACT CI command (tsgo --project tsconfig.tsgo.json --noEmit --checkers 4). typecheck:single = lighter single-checker (may miss errors the gate catches; do not trust for the gate)
 pnpm run test          # Vitest watch mode
 pnpm run test:run      # Vitest single run (CI mode)
 pnpm run test:coverage # Vitest with V8 coverage (thresholds: lines 54%, branches 46%, functions 68%, statements 56%)
diff --git a/package.json b/package.json
index c04a6037c..52b1c83c8 100644
--- a/package.json
+++ b/package.json
@@ -37,7 +37,8 @@
     "build:pages": "vite build",
     "build:turbo": "turbo run build",
     "preview": "vite preview",
-    "typecheck": "tsgo --project tsconfig.tsgo.json --noEmit",
+    "typecheck": "tsgo --project tsconfig.tsgo.json --noEmit --checkers 4",
+    "typecheck:single": "tsgo --project tsconfig.tsgo.json --noEmit",
     "typecheck:parallel": "tsgo --project tsconfig.tsgo.json --noEmit --checkers 4",
     "typecheck:turbo": "turbo run typecheck",
     "lhci:assert": "lhci autorun --upload.target=temporary-public-storage",
diff --git a/services/proForge/proForgeMemoryBank.ts b/services/proForge/proForgeMemoryBank.ts
index d6c276644..76b9a4f64 100644
--- a/services/proForge/proForgeMemoryBank.ts
+++ b/services/proForge/proForgeMemoryBank.ts
@@ -141,7 +141,12 @@ export async function searchMemoryEntries(
     const qVec = await embedText(query);
     const scored = await Promise.all(
       entries.map(async (entry) => {
-        const eVec = entry.embedding ?? (await embedText(`${entry.key} ${entry.content}`));
+        // QNBS-v3: stored embeddings are persisted as number[]; coerce to the Float32Array the
+        // similarity fn expects. Falls back to computing the embedding when none is stored.
+        const eVec =
+          entry.embedding != null
+            ? new Float32Array(entry.embedding)
+            : await embedText(`${entry.key} ${entry.content}`);
         const sim = cosineSimilarity(qVec, eVec);
         // QNBS-v3: hybrid blends semantic (0.7) with normalised keyword overlap (0.3).
         const score = mode === 'semantic' ? sim : 0.7 * sim + 0.3 * (scoreKeyword(entry) / maxKw);

From adedb31a9e34034010a663970ee9b41ac71f59d6 Mon Sep 17 00:00:00 2001
From: qnbs <155236708+qnbs@users.noreply.github.com>
Date: Wed, 10 Jun 2026 12:11:26 +0200
Subject: [PATCH 3/3] fix(proforge): resolve all CodeAnt review comments +
 finish es/fr/it i18n

Addresses every inline review comment on PR #106:

- #2 proForgeHistoryStore: reset cached dbPromise on open error so a
  failed IDB open can be retried instead of poisoning the singleton.
- #3 proForgeHistoryStore: resolve/reject on tx.oncomplete/onerror/onabort
  (not request.onsuccess) so a save only resolves once durably committed.
- #4 applyReviewEdits: anchor stale-offset edits to the occurrence nearest
  the original range and skip already-claimed positions, so duplicate
  phrases map to distinct occurrences deterministically.
- #5 baseAgent.gatherMemoryContext: feed a project-derived query
  (title/logline/genre) + ragMode into memory-bank retrieval instead of
  an undefined query; all 6 AI agents now route through it.
- #6 PipelineReviewPanel: await submitReview before announce/navigate.
- #7 useProForgeOrchestrator: always dispatch loadRunHistory (even []) so
  switching projects clears stale history.
- #1 i18n: translate ~55 proforge.review.*/stageName.* keys to es/fr/it
  (were English fallback); rebuild bundles.

Tests updated for the 4-arg gatherMemoryContext signature
(diagnostic/structural/publishing) + duplicate-phrase edit cases.

Local gates green: typecheck (--checkers 4), lint, full proForge suite.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 components/proForge/PipelineReviewPanel.tsx   |  10 +-
 hooks/useProForgeOrchestrator.ts              |   4 +-
 locales/es/common.json                        | 106 ++++++++---------
 locales/fr/common.json                        |  98 ++++++++--------
 locales/it/common.json                        | 108 +++++++++---------
 public/locales/es/bundle.json                 | 106 ++++++++---------
 public/locales/fr/bundle.json                 |  98 ++++++++--------
 public/locales/it/bundle.json                 | 108 +++++++++---------
 services/proForge/applyReviewEdits.ts         |  40 ++++++-
 services/proForge/pipelineAgents/baseAgent.ts |  21 +++-
 .../proForge/pipelineAgents/copyEditAgent.ts  |   2 +-
 .../pipelineAgents/diagnosticAgent.ts         |   2 +-
 .../proForge/pipelineAgents/proofAgent.ts     |   2 +-
 .../proForge/pipelineAgents/proseAgent.ts     |   2 +-
 .../pipelineAgents/publishingAgent.ts         |   2 +-
 .../pipelineAgents/structuralAgent.ts         |   2 +-
 services/proForge/proForgeHistoryStore.ts     |  17 ++-
 tests/unit/proForge/applyReviewEdits.test.ts  |  21 ++++
 .../pipelineAgents/diagnosticAgent.test.ts    |   8 +-
 .../pipelineAgents/publishingAgent.test.ts    |   8 +-
 .../pipelineAgents/structuralAgent.test.ts    |   8 +-
 21 files changed, 438 insertions(+), 335 deletions(-)

diff --git a/components/proForge/PipelineReviewPanel.tsx b/components/proForge/PipelineReviewPanel.tsx
index 0318a53c7..1e03b84d8 100644
--- a/components/proForge/PipelineReviewPanel.tsx
+++ b/components/proForge/PipelineReviewPanel.tsx
@@ -125,12 +125,14 @@ export const PipelineReviewPanel: React.FC = () => {
     (i) => i.confidence >= 0.85 && i.severity !== 'critical' && i.status === 'pending',
   ).length;
 
-  const handleSubmit = useCallback(() => {
+  const handleSubmit = useCallback(async () => {
     if (!stage) return;
     const items = activeStageResult?.reviewItems ?? [];
     const decisions = items.map((item) => ({ itemId: item.id, status: item.status }));
-    void submitReview(stage, decisions);
-    // QNBS-v3: WCAG live-region announce so screen-reader users hear the stage was submitted.
+    // QNBS-v3: Await completion before announcing/navigating — submitReview is async (applies
+    // edits + snapshots), so announcing success before it resolves could lie if it rejects.
+    await submitReview(stage, decisions);
+    // WCAG live-region announce so screen-reader users hear the stage was submitted.
     announce(t('proforge.review.announceSubmitted', { stage: stageLabel }), 'polite');
     setActiveView('dashboard');
   }, [stage, stageLabel, activeStageResult, submitReview, setActiveView, announce, t]);
@@ -297,7 +299,7 @@ export const PipelineReviewPanel: React.FC = () => {
         </button>
         <button
           type="button"
-          onClick={handleSubmit}
+          onClick={() => void handleSubmit()}
           disabled={pendingCount > 0}
           className="px-4 py-1.5 text-xs font-medium rounded-sc-md text-white disabled:opacity-50 transition-opacity"
           style={{ backgroundColor: 'var(--sc-accent)' }}
diff --git a/hooks/useProForgeOrchestrator.ts b/hooks/useProForgeOrchestrator.ts
index 1869b8c34..2f31aa667 100644
--- a/hooks/useProForgeOrchestrator.ts
+++ b/hooks/useProForgeOrchestrator.ts
@@ -65,7 +65,9 @@ export function useProForgeOrchestrator() {
       try {
         const { loadRunHistory } = await import('../services/proForge/proForgeHistoryStore');
         const runs = await loadRunHistory(projectId);
-        if (!cancelled && runs.length > 0) {
+        // QNBS-v3: Always dispatch (even []) so switching to a project with no saved history
+        // clears the previous project's run history from Redux instead of leaving it stale.
+        if (!cancelled) {
           dispatch(proForgeActions.loadRunHistory(runs));
         }
       } catch {
diff --git a/locales/es/common.json b/locales/es/common.json
index 87b3ec96d..0087b72f1 100644
--- a/locales/es/common.json
+++ b/locales/es/common.json
@@ -255,61 +255,61 @@
   "proforge.loading.structural": "Observando la forma de tu historia…",
   "proforge.pipeline.noneActive": "Sin pipeline activa",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
-  "proforge.review.accept": "Accept",
-  "proforge.review.acceptAll": "Accept All",
-  "proforge.review.acceptAllCritical": "Accept All Critical",
-  "proforge.review.announceSubmitted": "{{stage}} review submitted",
-  "proforge.review.confidence": "{{percent}}% confidence",
-  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
-  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
-  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
-  "proforge.review.filter.accepted": "Accepted",
-  "proforge.review.filter.all": "All",
-  "proforge.review.filter.pending": "Pending",
-  "proforge.review.filter.rejected": "Rejected",
-  "proforge.review.group.critical": "Critical Actions",
-  "proforge.review.group.info": "Suggestions",
-  "proforge.review.group.warning": "Warnings",
-  "proforge.review.heading": "Review: {{stage}}",
-  "proforge.review.ignore": "Ignore",
-  "proforge.review.moreCritical": "+{{count}} more critical issues below",
-  "proforge.review.noItems": "No review items available.",
-  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.accept": "Aceptar",
+  "proforge.review.acceptAll": "Aceptar todo",
+  "proforge.review.acceptAllCritical": "Aceptar todo lo crítico",
+  "proforge.review.announceSubmitted": "Revisión de {{stage}} enviada",
+  "proforge.review.confidence": "{{percent}} % de confianza",
+  "proforge.review.counts": "{{pending}} pendientes · {{accepted}} aceptados · {{rejected}} rechazados",
+  "proforge.review.criticalHeadingOne": "{{count}} problema crítico requiere atención",
+  "proforge.review.criticalHeadingOther": "{{count}} problemas críticos requieren atención",
+  "proforge.review.filter.accepted": "Aceptados",
+  "proforge.review.filter.all": "Todos",
+  "proforge.review.filter.pending": "Pendientes",
+  "proforge.review.filter.rejected": "Rechazados",
+  "proforge.review.group.critical": "Acciones críticas",
+  "proforge.review.group.info": "Sugerencias",
+  "proforge.review.group.warning": "Advertencias",
+  "proforge.review.heading": "Revisión: {{stage}}",
+  "proforge.review.ignore": "Ignorar",
+  "proforge.review.moreCritical": "+{{count}} más problemas críticos abajo",
+  "proforge.review.noItems": "No hay elementos de revisión disponibles.",
+  "proforge.review.noItemsForFilter": "Ningún elemento coincide con este filtro.",
   "proforge.review.original": "Original:",
-  "proforge.review.pendingCount": "{{count}} Pending",
-  "proforge.review.proposed": "Proposed:",
-  "proforge.review.quickAccept": "Quick Accept ({{count}})",
-  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
-  "proforge.review.reject": "Reject",
-  "proforge.review.rejectAll": "Reject All",
-  "proforge.review.severity.critical": "Critical",
-  "proforge.review.severity.info": "Info",
-  "proforge.review.severity.warning": "Warning",
-  "proforge.review.showDetails": "Show details",
-  "proforge.review.showLess": "Show less",
-  "proforge.review.skipStage": "Skip Stage",
-  "proforge.review.submitContinue": "Submit & Continue",
-  "proforge.review.type.arcIssue": "Arc",
-  "proforge.review.type.consistencyIssue": "Consistency",
-  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.pendingCount": "{{count}} pendientes",
+  "proforge.review.proposed": "Propuesta:",
+  "proforge.review.quickAccept": "Aceptar rápido ({{count}})",
+  "proforge.review.quickAcceptTitle": "Aceptar todas las sugerencias no críticas de alta confianza",
+  "proforge.review.reject": "Rechazar",
+  "proforge.review.rejectAll": "Rechazar todo",
+  "proforge.review.severity.critical": "Crítico",
+  "proforge.review.severity.info": "Información",
+  "proforge.review.severity.warning": "Advertencia",
+  "proforge.review.showDetails": "Mostrar detalles",
+  "proforge.review.showLess": "Mostrar menos",
+  "proforge.review.skipStage": "Omitir fase",
+  "proforge.review.submitContinue": "Enviar y continuar",
+  "proforge.review.type.arcIssue": "Arco",
+  "proforge.review.type.consistencyIssue": "Coherencia",
+  "proforge.review.type.grammarEdit": "Gramática",
   "proforge.review.type.legalWarning": "Legal",
-  "proforge.review.type.pacingIssue": "Pacing",
-  "proforge.review.type.plotHole": "Plot Hole",
-  "proforge.review.type.proseEdit": "Prose",
-  "proforge.review.type.repetitionHit": "Repetition",
-  "proforge.review.type.structuralEdit": "Structural",
-  "proforge.review.type.styleEdit": "Style",
-  "proforge.review.type.technicalIssue": "Technical",
-  "proforge.stageName.analytics": "Analytics",
-  "proforge.stageName.archived": "Archived",
-  "proforge.stageName.copyEdit": "Copy Edit",
-  "proforge.stageName.idle": "Idle",
-  "proforge.stageName.intake": "Intake & Diagnostic",
-  "proforge.stageName.lineProse": "Line & Prose",
-  "proforge.stageName.production": "Production",
-  "proforge.stageName.proof": "Proofreading",
-  "proforge.stageName.publishing": "Publishing",
-  "proforge.stageName.structural": "Structural Edit",
+  "proforge.review.type.pacingIssue": "Ritmo",
+  "proforge.review.type.plotHole": "Vacío argumental",
+  "proforge.review.type.proseEdit": "Prosa",
+  "proforge.review.type.repetitionHit": "Repetición",
+  "proforge.review.type.structuralEdit": "Estructura",
+  "proforge.review.type.styleEdit": "Estilo",
+  "proforge.review.type.technicalIssue": "Técnico",
+  "proforge.stageName.analytics": "Análisis",
+  "proforge.stageName.archived": "Archivado",
+  "proforge.stageName.copyEdit": "Corrección de estilo",
+  "proforge.stageName.idle": "Inactivo",
+  "proforge.stageName.intake": "Recepción y diagnóstico",
+  "proforge.stageName.lineProse": "Línea y prosa",
+  "proforge.stageName.production": "Producción",
+  "proforge.stageName.proof": "Corrección de pruebas",
+  "proforge.stageName.publishing": "Publicación",
+  "proforge.stageName.structural": "Edición estructural",
   "progress.allTime.label": "palabras escritas en sesiones",
   "progress.allTime.title": "Total histórico",
   "progress.bestDay.title": "Mejor día",
diff --git a/locales/fr/common.json b/locales/fr/common.json
index a34bb87b4..f699dd75a 100644
--- a/locales/fr/common.json
+++ b/locales/fr/common.json
@@ -255,61 +255,61 @@
   "proforge.loading.structural": "Examen de la structure de votre histoire…",
   "proforge.pipeline.noneActive": "Aucune pipeline active",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
-  "proforge.review.accept": "Accept",
-  "proforge.review.acceptAll": "Accept All",
-  "proforge.review.acceptAllCritical": "Accept All Critical",
-  "proforge.review.announceSubmitted": "{{stage}} review submitted",
-  "proforge.review.confidence": "{{percent}}% confidence",
-  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
-  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
-  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
-  "proforge.review.filter.accepted": "Accepted",
-  "proforge.review.filter.all": "All",
-  "proforge.review.filter.pending": "Pending",
-  "proforge.review.filter.rejected": "Rejected",
-  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.accept": "Accepter",
+  "proforge.review.acceptAll": "Tout accepter",
+  "proforge.review.acceptAllCritical": "Tout accepter (critiques)",
+  "proforge.review.announceSubmitted": "Révision de {{stage}} envoyée",
+  "proforge.review.confidence": "{{percent}} % de confiance",
+  "proforge.review.counts": "{{pending}} en attente · {{accepted}} acceptés · {{rejected}} rejetés",
+  "proforge.review.criticalHeadingOne": "{{count}} problème critique requiert votre attention",
+  "proforge.review.criticalHeadingOther": "{{count}} problèmes critiques requièrent votre attention",
+  "proforge.review.filter.accepted": "Acceptés",
+  "proforge.review.filter.all": "Tous",
+  "proforge.review.filter.pending": "En attente",
+  "proforge.review.filter.rejected": "Rejetés",
+  "proforge.review.group.critical": "Actions critiques",
   "proforge.review.group.info": "Suggestions",
-  "proforge.review.group.warning": "Warnings",
-  "proforge.review.heading": "Review: {{stage}}",
-  "proforge.review.ignore": "Ignore",
-  "proforge.review.moreCritical": "+{{count}} more critical issues below",
-  "proforge.review.noItems": "No review items available.",
-  "proforge.review.noItemsForFilter": "No items match this filter.",
-  "proforge.review.original": "Original:",
-  "proforge.review.pendingCount": "{{count}} Pending",
-  "proforge.review.proposed": "Proposed:",
-  "proforge.review.quickAccept": "Quick Accept ({{count}})",
-  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
-  "proforge.review.reject": "Reject",
-  "proforge.review.rejectAll": "Reject All",
-  "proforge.review.severity.critical": "Critical",
+  "proforge.review.group.warning": "Avertissements",
+  "proforge.review.heading": "Révision : {{stage}}",
+  "proforge.review.ignore": "Ignorer",
+  "proforge.review.moreCritical": "+{{count}} autres problèmes critiques ci-dessous",
+  "proforge.review.noItems": "Aucun élément à réviser.",
+  "proforge.review.noItemsForFilter": "Aucun élément ne correspond à ce filtre.",
+  "proforge.review.original": "Original :",
+  "proforge.review.pendingCount": "{{count}} en attente",
+  "proforge.review.proposed": "Proposition :",
+  "proforge.review.quickAccept": "Acceptation rapide ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accepter toutes les suggestions non critiques à haute confiance",
+  "proforge.review.reject": "Rejeter",
+  "proforge.review.rejectAll": "Tout rejeter",
+  "proforge.review.severity.critical": "Critique",
   "proforge.review.severity.info": "Info",
-  "proforge.review.severity.warning": "Warning",
-  "proforge.review.showDetails": "Show details",
-  "proforge.review.showLess": "Show less",
-  "proforge.review.skipStage": "Skip Stage",
-  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.severity.warning": "Avertissement",
+  "proforge.review.showDetails": "Afficher les détails",
+  "proforge.review.showLess": "Afficher moins",
+  "proforge.review.skipStage": "Ignorer l'étape",
+  "proforge.review.submitContinue": "Envoyer et continuer",
   "proforge.review.type.arcIssue": "Arc",
-  "proforge.review.type.consistencyIssue": "Consistency",
-  "proforge.review.type.grammarEdit": "Grammar",
-  "proforge.review.type.legalWarning": "Legal",
-  "proforge.review.type.pacingIssue": "Pacing",
-  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.consistencyIssue": "Cohérence",
+  "proforge.review.type.grammarEdit": "Grammaire",
+  "proforge.review.type.legalWarning": "Juridique",
+  "proforge.review.type.pacingIssue": "Rythme",
+  "proforge.review.type.plotHole": "Incohérence narrative",
   "proforge.review.type.proseEdit": "Prose",
-  "proforge.review.type.repetitionHit": "Repetition",
-  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.repetitionHit": "Répétition",
+  "proforge.review.type.structuralEdit": "Structure",
   "proforge.review.type.styleEdit": "Style",
-  "proforge.review.type.technicalIssue": "Technical",
-  "proforge.stageName.analytics": "Analytics",
-  "proforge.stageName.archived": "Archived",
-  "proforge.stageName.copyEdit": "Copy Edit",
-  "proforge.stageName.idle": "Idle",
-  "proforge.stageName.intake": "Intake & Diagnostic",
-  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.review.type.technicalIssue": "Technique",
+  "proforge.stageName.analytics": "Analyse",
+  "proforge.stageName.archived": "Archivé",
+  "proforge.stageName.copyEdit": "Correction stylistique",
+  "proforge.stageName.idle": "Inactif",
+  "proforge.stageName.intake": "Réception et diagnostic",
+  "proforge.stageName.lineProse": "Ligne et prose",
   "proforge.stageName.production": "Production",
-  "proforge.stageName.proof": "Proofreading",
-  "proforge.stageName.publishing": "Publishing",
-  "proforge.stageName.structural": "Structural Edit",
+  "proforge.stageName.proof": "Relecture",
+  "proforge.stageName.publishing": "Publication",
+  "proforge.stageName.structural": "Édition structurelle",
   "progress.allTime.label": "mots écrits en sessions",
   "progress.allTime.title": "Total cumulé",
   "progress.bestDay.title": "Meilleur jour",
diff --git a/locales/it/common.json b/locales/it/common.json
index b8ce9d005..8d35b5eb4 100644
--- a/locales/it/common.json
+++ b/locales/it/common.json
@@ -255,61 +255,61 @@
   "proforge.loading.structural": "Osservando la forma della tua storia…",
   "proforge.pipeline.noneActive": "Nessuna pipeline attiva",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
-  "proforge.review.accept": "Accept",
-  "proforge.review.acceptAll": "Accept All",
-  "proforge.review.acceptAllCritical": "Accept All Critical",
-  "proforge.review.announceSubmitted": "{{stage}} review submitted",
-  "proforge.review.confidence": "{{percent}}% confidence",
-  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
-  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
-  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
-  "proforge.review.filter.accepted": "Accepted",
-  "proforge.review.filter.all": "All",
-  "proforge.review.filter.pending": "Pending",
-  "proforge.review.filter.rejected": "Rejected",
-  "proforge.review.group.critical": "Critical Actions",
-  "proforge.review.group.info": "Suggestions",
-  "proforge.review.group.warning": "Warnings",
-  "proforge.review.heading": "Review: {{stage}}",
-  "proforge.review.ignore": "Ignore",
-  "proforge.review.moreCritical": "+{{count}} more critical issues below",
-  "proforge.review.noItems": "No review items available.",
-  "proforge.review.noItemsForFilter": "No items match this filter.",
-  "proforge.review.original": "Original:",
-  "proforge.review.pendingCount": "{{count}} Pending",
-  "proforge.review.proposed": "Proposed:",
-  "proforge.review.quickAccept": "Quick Accept ({{count}})",
-  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
-  "proforge.review.reject": "Reject",
-  "proforge.review.rejectAll": "Reject All",
-  "proforge.review.severity.critical": "Critical",
+  "proforge.review.accept": "Accetta",
+  "proforge.review.acceptAll": "Accetta tutto",
+  "proforge.review.acceptAllCritical": "Accetta tutti i critici",
+  "proforge.review.announceSubmitted": "Revisione di {{stage}} inviata",
+  "proforge.review.confidence": "{{percent}}% di affidabilità",
+  "proforge.review.counts": "{{pending}} in sospeso · {{accepted}} accettati · {{rejected}} rifiutati",
+  "proforge.review.criticalHeadingOne": "{{count}} problema critico richiede attenzione",
+  "proforge.review.criticalHeadingOther": "{{count}} problemi critici richiedono attenzione",
+  "proforge.review.filter.accepted": "Accettati",
+  "proforge.review.filter.all": "Tutti",
+  "proforge.review.filter.pending": "In sospeso",
+  "proforge.review.filter.rejected": "Rifiutati",
+  "proforge.review.group.critical": "Azioni critiche",
+  "proforge.review.group.info": "Suggerimenti",
+  "proforge.review.group.warning": "Avvisi",
+  "proforge.review.heading": "Revisione: {{stage}}",
+  "proforge.review.ignore": "Ignora",
+  "proforge.review.moreCritical": "+{{count}} altri problemi critici sotto",
+  "proforge.review.noItems": "Nessun elemento da revisionare disponibile.",
+  "proforge.review.noItemsForFilter": "Nessun elemento corrisponde a questo filtro.",
+  "proforge.review.original": "Originale:",
+  "proforge.review.pendingCount": "{{count}} in sospeso",
+  "proforge.review.proposed": "Proposta:",
+  "proforge.review.quickAccept": "Accetta rapidamente ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accetta tutti i suggerimenti non critici ad alta affidabilità",
+  "proforge.review.reject": "Rifiuta",
+  "proforge.review.rejectAll": "Rifiuta tutto",
+  "proforge.review.severity.critical": "Critico",
   "proforge.review.severity.info": "Info",
-  "proforge.review.severity.warning": "Warning",
-  "proforge.review.showDetails": "Show details",
-  "proforge.review.showLess": "Show less",
-  "proforge.review.skipStage": "Skip Stage",
-  "proforge.review.submitContinue": "Submit & Continue",
-  "proforge.review.type.arcIssue": "Arc",
-  "proforge.review.type.consistencyIssue": "Consistency",
-  "proforge.review.type.grammarEdit": "Grammar",
-  "proforge.review.type.legalWarning": "Legal",
-  "proforge.review.type.pacingIssue": "Pacing",
-  "proforge.review.type.plotHole": "Plot Hole",
-  "proforge.review.type.proseEdit": "Prose",
-  "proforge.review.type.repetitionHit": "Repetition",
-  "proforge.review.type.structuralEdit": "Structural",
-  "proforge.review.type.styleEdit": "Style",
-  "proforge.review.type.technicalIssue": "Technical",
-  "proforge.stageName.analytics": "Analytics",
-  "proforge.stageName.archived": "Archived",
-  "proforge.stageName.copyEdit": "Copy Edit",
-  "proforge.stageName.idle": "Idle",
-  "proforge.stageName.intake": "Intake & Diagnostic",
-  "proforge.stageName.lineProse": "Line & Prose",
-  "proforge.stageName.production": "Production",
-  "proforge.stageName.proof": "Proofreading",
-  "proforge.stageName.publishing": "Publishing",
-  "proforge.stageName.structural": "Structural Edit",
+  "proforge.review.severity.warning": "Avviso",
+  "proforge.review.showDetails": "Mostra dettagli",
+  "proforge.review.showLess": "Mostra meno",
+  "proforge.review.skipStage": "Salta fase",
+  "proforge.review.submitContinue": "Invia e continua",
+  "proforge.review.type.arcIssue": "Arco",
+  "proforge.review.type.consistencyIssue": "Coerenza",
+  "proforge.review.type.grammarEdit": "Grammatica",
+  "proforge.review.type.legalWarning": "Legale",
+  "proforge.review.type.pacingIssue": "Ritmo",
+  "proforge.review.type.plotHole": "Buco di trama",
+  "proforge.review.type.proseEdit": "Prosa",
+  "proforge.review.type.repetitionHit": "Ripetizione",
+  "proforge.review.type.structuralEdit": "Struttura",
+  "proforge.review.type.styleEdit": "Stile",
+  "proforge.review.type.technicalIssue": "Tecnico",
+  "proforge.stageName.analytics": "Analisi",
+  "proforge.stageName.archived": "Archiviato",
+  "proforge.stageName.copyEdit": "Revisione di copia",
+  "proforge.stageName.idle": "Inattivo",
+  "proforge.stageName.intake": "Acquisizione e diagnosi",
+  "proforge.stageName.lineProse": "Riga e prosa",
+  "proforge.stageName.production": "Produzione",
+  "proforge.stageName.proof": "Correzione di bozze",
+  "proforge.stageName.publishing": "Pubblicazione",
+  "proforge.stageName.structural": "Modifica strutturale",
   "progress.allTime.label": "parole scritte nelle sessioni",
   "progress.allTime.title": "Totale storico",
   "progress.bestDay.title": "Giorno migliore",
diff --git a/public/locales/es/bundle.json b/public/locales/es/bundle.json
index aa73bf86a..974887504 100644
--- a/public/locales/es/bundle.json
+++ b/public/locales/es/bundle.json
@@ -255,61 +255,61 @@
   "proforge.loading.structural": "Observando la forma de tu historia…",
   "proforge.pipeline.noneActive": "Sin pipeline activa",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
-  "proforge.review.accept": "Accept",
-  "proforge.review.acceptAll": "Accept All",
-  "proforge.review.acceptAllCritical": "Accept All Critical",
-  "proforge.review.announceSubmitted": "{{stage}} review submitted",
-  "proforge.review.confidence": "{{percent}}% confidence",
-  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
-  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
-  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
-  "proforge.review.filter.accepted": "Accepted",
-  "proforge.review.filter.all": "All",
-  "proforge.review.filter.pending": "Pending",
-  "proforge.review.filter.rejected": "Rejected",
-  "proforge.review.group.critical": "Critical Actions",
-  "proforge.review.group.info": "Suggestions",
-  "proforge.review.group.warning": "Warnings",
-  "proforge.review.heading": "Review: {{stage}}",
-  "proforge.review.ignore": "Ignore",
-  "proforge.review.moreCritical": "+{{count}} more critical issues below",
-  "proforge.review.noItems": "No review items available.",
-  "proforge.review.noItemsForFilter": "No items match this filter.",
+  "proforge.review.accept": "Aceptar",
+  "proforge.review.acceptAll": "Aceptar todo",
+  "proforge.review.acceptAllCritical": "Aceptar todo lo crítico",
+  "proforge.review.announceSubmitted": "Revisión de {{stage}} enviada",
+  "proforge.review.confidence": "{{percent}} % de confianza",
+  "proforge.review.counts": "{{pending}} pendientes · {{accepted}} aceptados · {{rejected}} rechazados",
+  "proforge.review.criticalHeadingOne": "{{count}} problema crítico requiere atención",
+  "proforge.review.criticalHeadingOther": "{{count}} problemas críticos requieren atención",
+  "proforge.review.filter.accepted": "Aceptados",
+  "proforge.review.filter.all": "Todos",
+  "proforge.review.filter.pending": "Pendientes",
+  "proforge.review.filter.rejected": "Rechazados",
+  "proforge.review.group.critical": "Acciones críticas",
+  "proforge.review.group.info": "Sugerencias",
+  "proforge.review.group.warning": "Advertencias",
+  "proforge.review.heading": "Revisión: {{stage}}",
+  "proforge.review.ignore": "Ignorar",
+  "proforge.review.moreCritical": "+{{count}} más problemas críticos abajo",
+  "proforge.review.noItems": "No hay elementos de revisión disponibles.",
+  "proforge.review.noItemsForFilter": "Ningún elemento coincide con este filtro.",
   "proforge.review.original": "Original:",
-  "proforge.review.pendingCount": "{{count}} Pending",
-  "proforge.review.proposed": "Proposed:",
-  "proforge.review.quickAccept": "Quick Accept ({{count}})",
-  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
-  "proforge.review.reject": "Reject",
-  "proforge.review.rejectAll": "Reject All",
-  "proforge.review.severity.critical": "Critical",
-  "proforge.review.severity.info": "Info",
-  "proforge.review.severity.warning": "Warning",
-  "proforge.review.showDetails": "Show details",
-  "proforge.review.showLess": "Show less",
-  "proforge.review.skipStage": "Skip Stage",
-  "proforge.review.submitContinue": "Submit & Continue",
-  "proforge.review.type.arcIssue": "Arc",
-  "proforge.review.type.consistencyIssue": "Consistency",
-  "proforge.review.type.grammarEdit": "Grammar",
+  "proforge.review.pendingCount": "{{count}} pendientes",
+  "proforge.review.proposed": "Propuesta:",
+  "proforge.review.quickAccept": "Aceptar rápido ({{count}})",
+  "proforge.review.quickAcceptTitle": "Aceptar todas las sugerencias no críticas de alta confianza",
+  "proforge.review.reject": "Rechazar",
+  "proforge.review.rejectAll": "Rechazar todo",
+  "proforge.review.severity.critical": "Crítico",
+  "proforge.review.severity.info": "Información",
+  "proforge.review.severity.warning": "Advertencia",
+  "proforge.review.showDetails": "Mostrar detalles",
+  "proforge.review.showLess": "Mostrar menos",
+  "proforge.review.skipStage": "Omitir fase",
+  "proforge.review.submitContinue": "Enviar y continuar",
+  "proforge.review.type.arcIssue": "Arco",
+  "proforge.review.type.consistencyIssue": "Coherencia",
+  "proforge.review.type.grammarEdit": "Gramática",
   "proforge.review.type.legalWarning": "Legal",
-  "proforge.review.type.pacingIssue": "Pacing",
-  "proforge.review.type.plotHole": "Plot Hole",
-  "proforge.review.type.proseEdit": "Prose",
-  "proforge.review.type.repetitionHit": "Repetition",
-  "proforge.review.type.structuralEdit": "Structural",
-  "proforge.review.type.styleEdit": "Style",
-  "proforge.review.type.technicalIssue": "Technical",
-  "proforge.stageName.analytics": "Analytics",
-  "proforge.stageName.archived": "Archived",
-  "proforge.stageName.copyEdit": "Copy Edit",
-  "proforge.stageName.idle": "Idle",
-  "proforge.stageName.intake": "Intake & Diagnostic",
-  "proforge.stageName.lineProse": "Line & Prose",
-  "proforge.stageName.production": "Production",
-  "proforge.stageName.proof": "Proofreading",
-  "proforge.stageName.publishing": "Publishing",
-  "proforge.stageName.structural": "Structural Edit",
+  "proforge.review.type.pacingIssue": "Ritmo",
+  "proforge.review.type.plotHole": "Vacío argumental",
+  "proforge.review.type.proseEdit": "Prosa",
+  "proforge.review.type.repetitionHit": "Repetición",
+  "proforge.review.type.structuralEdit": "Estructura",
+  "proforge.review.type.styleEdit": "Estilo",
+  "proforge.review.type.technicalIssue": "Técnico",
+  "proforge.stageName.analytics": "Análisis",
+  "proforge.stageName.archived": "Archivado",
+  "proforge.stageName.copyEdit": "Corrección de estilo",
+  "proforge.stageName.idle": "Inactivo",
+  "proforge.stageName.intake": "Recepción y diagnóstico",
+  "proforge.stageName.lineProse": "Línea y prosa",
+  "proforge.stageName.production": "Producción",
+  "proforge.stageName.proof": "Corrección de pruebas",
+  "proforge.stageName.publishing": "Publicación",
+  "proforge.stageName.structural": "Edición estructural",
   "progress.allTime.label": "palabras escritas en sesiones",
   "progress.allTime.title": "Total histórico",
   "progress.bestDay.title": "Mejor día",
diff --git a/public/locales/fr/bundle.json b/public/locales/fr/bundle.json
index 3f5257e7f..0d70cf58d 100644
--- a/public/locales/fr/bundle.json
+++ b/public/locales/fr/bundle.json
@@ -255,61 +255,61 @@
   "proforge.loading.structural": "Examen de la structure de votre histoire…",
   "proforge.pipeline.noneActive": "Aucune pipeline active",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
-  "proforge.review.accept": "Accept",
-  "proforge.review.acceptAll": "Accept All",
-  "proforge.review.acceptAllCritical": "Accept All Critical",
-  "proforge.review.announceSubmitted": "{{stage}} review submitted",
-  "proforge.review.confidence": "{{percent}}% confidence",
-  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
-  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
-  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
-  "proforge.review.filter.accepted": "Accepted",
-  "proforge.review.filter.all": "All",
-  "proforge.review.filter.pending": "Pending",
-  "proforge.review.filter.rejected": "Rejected",
-  "proforge.review.group.critical": "Critical Actions",
+  "proforge.review.accept": "Accepter",
+  "proforge.review.acceptAll": "Tout accepter",
+  "proforge.review.acceptAllCritical": "Tout accepter (critiques)",
+  "proforge.review.announceSubmitted": "Révision de {{stage}} envoyée",
+  "proforge.review.confidence": "{{percent}} % de confiance",
+  "proforge.review.counts": "{{pending}} en attente · {{accepted}} acceptés · {{rejected}} rejetés",
+  "proforge.review.criticalHeadingOne": "{{count}} problème critique requiert votre attention",
+  "proforge.review.criticalHeadingOther": "{{count}} problèmes critiques requièrent votre attention",
+  "proforge.review.filter.accepted": "Acceptés",
+  "proforge.review.filter.all": "Tous",
+  "proforge.review.filter.pending": "En attente",
+  "proforge.review.filter.rejected": "Rejetés",
+  "proforge.review.group.critical": "Actions critiques",
   "proforge.review.group.info": "Suggestions",
-  "proforge.review.group.warning": "Warnings",
-  "proforge.review.heading": "Review: {{stage}}",
-  "proforge.review.ignore": "Ignore",
-  "proforge.review.moreCritical": "+{{count}} more critical issues below",
-  "proforge.review.noItems": "No review items available.",
-  "proforge.review.noItemsForFilter": "No items match this filter.",
-  "proforge.review.original": "Original:",
-  "proforge.review.pendingCount": "{{count}} Pending",
-  "proforge.review.proposed": "Proposed:",
-  "proforge.review.quickAccept": "Quick Accept ({{count}})",
-  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
-  "proforge.review.reject": "Reject",
-  "proforge.review.rejectAll": "Reject All",
-  "proforge.review.severity.critical": "Critical",
+  "proforge.review.group.warning": "Avertissements",
+  "proforge.review.heading": "Révision : {{stage}}",
+  "proforge.review.ignore": "Ignorer",
+  "proforge.review.moreCritical": "+{{count}} autres problèmes critiques ci-dessous",
+  "proforge.review.noItems": "Aucun élément à réviser.",
+  "proforge.review.noItemsForFilter": "Aucun élément ne correspond à ce filtre.",
+  "proforge.review.original": "Original :",
+  "proforge.review.pendingCount": "{{count}} en attente",
+  "proforge.review.proposed": "Proposition :",
+  "proforge.review.quickAccept": "Acceptation rapide ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accepter toutes les suggestions non critiques à haute confiance",
+  "proforge.review.reject": "Rejeter",
+  "proforge.review.rejectAll": "Tout rejeter",
+  "proforge.review.severity.critical": "Critique",
   "proforge.review.severity.info": "Info",
-  "proforge.review.severity.warning": "Warning",
-  "proforge.review.showDetails": "Show details",
-  "proforge.review.showLess": "Show less",
-  "proforge.review.skipStage": "Skip Stage",
-  "proforge.review.submitContinue": "Submit & Continue",
+  "proforge.review.severity.warning": "Avertissement",
+  "proforge.review.showDetails": "Afficher les détails",
+  "proforge.review.showLess": "Afficher moins",
+  "proforge.review.skipStage": "Ignorer l'étape",
+  "proforge.review.submitContinue": "Envoyer et continuer",
   "proforge.review.type.arcIssue": "Arc",
-  "proforge.review.type.consistencyIssue": "Consistency",
-  "proforge.review.type.grammarEdit": "Grammar",
-  "proforge.review.type.legalWarning": "Legal",
-  "proforge.review.type.pacingIssue": "Pacing",
-  "proforge.review.type.plotHole": "Plot Hole",
+  "proforge.review.type.consistencyIssue": "Cohérence",
+  "proforge.review.type.grammarEdit": "Grammaire",
+  "proforge.review.type.legalWarning": "Juridique",
+  "proforge.review.type.pacingIssue": "Rythme",
+  "proforge.review.type.plotHole": "Incohérence narrative",
   "proforge.review.type.proseEdit": "Prose",
-  "proforge.review.type.repetitionHit": "Repetition",
-  "proforge.review.type.structuralEdit": "Structural",
+  "proforge.review.type.repetitionHit": "Répétition",
+  "proforge.review.type.structuralEdit": "Structure",
   "proforge.review.type.styleEdit": "Style",
-  "proforge.review.type.technicalIssue": "Technical",
-  "proforge.stageName.analytics": "Analytics",
-  "proforge.stageName.archived": "Archived",
-  "proforge.stageName.copyEdit": "Copy Edit",
-  "proforge.stageName.idle": "Idle",
-  "proforge.stageName.intake": "Intake & Diagnostic",
-  "proforge.stageName.lineProse": "Line & Prose",
+  "proforge.review.type.technicalIssue": "Technique",
+  "proforge.stageName.analytics": "Analyse",
+  "proforge.stageName.archived": "Archivé",
+  "proforge.stageName.copyEdit": "Correction stylistique",
+  "proforge.stageName.idle": "Inactif",
+  "proforge.stageName.intake": "Réception et diagnostic",
+  "proforge.stageName.lineProse": "Ligne et prose",
   "proforge.stageName.production": "Production",
-  "proforge.stageName.proof": "Proofreading",
-  "proforge.stageName.publishing": "Publishing",
-  "proforge.stageName.structural": "Structural Edit",
+  "proforge.stageName.proof": "Relecture",
+  "proforge.stageName.publishing": "Publication",
+  "proforge.stageName.structural": "Édition structurelle",
   "progress.allTime.label": "mots écrits en sessions",
   "progress.allTime.title": "Total cumulé",
   "progress.bestDay.title": "Meilleur jour",
diff --git a/public/locales/it/bundle.json b/public/locales/it/bundle.json
index 3c9130e2d..7f8f91543 100644
--- a/public/locales/it/bundle.json
+++ b/public/locales/it/bundle.json
@@ -255,61 +255,61 @@
   "proforge.loading.structural": "Osservando la forma della tua storia…",
   "proforge.pipeline.noneActive": "Nessuna pipeline attiva",
   "proforge.pipeline.title": "Ultimate Author Pipeline",
-  "proforge.review.accept": "Accept",
-  "proforge.review.acceptAll": "Accept All",
-  "proforge.review.acceptAllCritical": "Accept All Critical",
-  "proforge.review.announceSubmitted": "{{stage}} review submitted",
-  "proforge.review.confidence": "{{percent}}% confidence",
-  "proforge.review.counts": "{{pending}} pending · {{accepted}} accepted · {{rejected}} rejected",
-  "proforge.review.criticalHeadingOne": "{{count}} critical issue needs attention",
-  "proforge.review.criticalHeadingOther": "{{count}} critical issues need attention",
-  "proforge.review.filter.accepted": "Accepted",
-  "proforge.review.filter.all": "All",
-  "proforge.review.filter.pending": "Pending",
-  "proforge.review.filter.rejected": "Rejected",
-  "proforge.review.group.critical": "Critical Actions",
-  "proforge.review.group.info": "Suggestions",
-  "proforge.review.group.warning": "Warnings",
-  "proforge.review.heading": "Review: {{stage}}",
-  "proforge.review.ignore": "Ignore",
-  "proforge.review.moreCritical": "+{{count}} more critical issues below",
-  "proforge.review.noItems": "No review items available.",
-  "proforge.review.noItemsForFilter": "No items match this filter.",
-  "proforge.review.original": "Original:",
-  "proforge.review.pendingCount": "{{count}} Pending",
-  "proforge.review.proposed": "Proposed:",
-  "proforge.review.quickAccept": "Quick Accept ({{count}})",
-  "proforge.review.quickAcceptTitle": "Accept all high-confidence non-critical suggestions",
-  "proforge.review.reject": "Reject",
-  "proforge.review.rejectAll": "Reject All",
-  "proforge.review.severity.critical": "Critical",
+  "proforge.review.accept": "Accetta",
+  "proforge.review.acceptAll": "Accetta tutto",
+  "proforge.review.acceptAllCritical": "Accetta tutti i critici",
+  "proforge.review.announceSubmitted": "Revisione di {{stage}} inviata",
+  "proforge.review.confidence": "{{percent}}% di affidabilità",
+  "proforge.review.counts": "{{pending}} in sospeso · {{accepted}} accettati · {{rejected}} rifiutati",
+  "proforge.review.criticalHeadingOne": "{{count}} problema critico richiede attenzione",
+  "proforge.review.criticalHeadingOther": "{{count}} problemi critici richiedono attenzione",
+  "proforge.review.filter.accepted": "Accettati",
+  "proforge.review.filter.all": "Tutti",
+  "proforge.review.filter.pending": "In sospeso",
+  "proforge.review.filter.rejected": "Rifiutati",
+  "proforge.review.group.critical": "Azioni critiche",
+  "proforge.review.group.info": "Suggerimenti",
+  "proforge.review.group.warning": "Avvisi",
+  "proforge.review.heading": "Revisione: {{stage}}",
+  "proforge.review.ignore": "Ignora",
+  "proforge.review.moreCritical": "+{{count}} altri problemi critici sotto",
+  "proforge.review.noItems": "Nessun elemento da revisionare disponibile.",
+  "proforge.review.noItemsForFilter": "Nessun elemento corrisponde a questo filtro.",
+  "proforge.review.original": "Originale:",
+  "proforge.review.pendingCount": "{{count}} in sospeso",
+  "proforge.review.proposed": "Proposta:",
+  "proforge.review.quickAccept": "Accetta rapidamente ({{count}})",
+  "proforge.review.quickAcceptTitle": "Accetta tutti i suggerimenti non critici ad alta affidabilità",
+  "proforge.review.reject": "Rifiuta",
+  "proforge.review.rejectAll": "Rifiuta tutto",
+  "proforge.review.severity.critical": "Critico",
   "proforge.review.severity.info": "Info",
-  "proforge.review.severity.warning": "Warning",
-  "proforge.review.showDetails": "Show details",
-  "proforge.review.showLess": "Show less",
-  "proforge.review.skipStage": "Skip Stage",
-  "proforge.review.submitContinue": "Submit & Continue",
-  "proforge.review.type.arcIssue": "Arc",
-  "proforge.review.type.consistencyIssue": "Consistency",
-  "proforge.review.type.grammarEdit": "Grammar",
-  "proforge.review.type.legalWarning": "Legal",
-  "proforge.review.type.pacingIssue": "Pacing",
-  "proforge.review.type.plotHole": "Plot Hole",
-  "proforge.review.type.proseEdit": "Prose",
-  "proforge.review.type.repetitionHit": "Repetition",
-  "proforge.review.type.structuralEdit": "Structural",
-  "proforge.review.type.styleEdit": "Style",
-  "proforge.review.type.technicalIssue": "Technical",
-  "proforge.stageName.analytics": "Analytics",
-  "proforge.stageName.archived": "Archived",
-  "proforge.stageName.copyEdit": "Copy Edit",
-  "proforge.stageName.idle": "Idle",
-  "proforge.stageName.intake": "Intake & Diagnostic",
-  "proforge.stageName.lineProse": "Line & Prose",
-  "proforge.stageName.production": "Production",
-  "proforge.stageName.proof": "Proofreading",
-  "proforge.stageName.publishing": "Publishing",
-  "proforge.stageName.structural": "Structural Edit",
+  "proforge.review.severity.warning": "Avviso",
+  "proforge.review.showDetails": "Mostra dettagli",
+  "proforge.review.showLess": "Mostra meno",
+  "proforge.review.skipStage": "Salta fase",
+  "proforge.review.submitContinue": "Invia e continua",
+  "proforge.review.type.arcIssue": "Arco",
+  "proforge.review.type.consistencyIssue": "Coerenza",
+  "proforge.review.type.grammarEdit": "Grammatica",
+  "proforge.review.type.legalWarning": "Legale",
+  "proforge.review.type.pacingIssue": "Ritmo",
+  "proforge.review.type.plotHole": "Buco di trama",
+  "proforge.review.type.proseEdit": "Prosa",
+  "proforge.review.type.repetitionHit": "Ripetizione",
+  "proforge.review.type.structuralEdit": "Struttura",
+  "proforge.review.type.styleEdit": "Stile",
+  "proforge.review.type.technicalIssue": "Tecnico",
+  "proforge.stageName.analytics": "Analisi",
+  "proforge.stageName.archived": "Archiviato",
+  "proforge.stageName.copyEdit": "Revisione di copia",
+  "proforge.stageName.idle": "Inattivo",
+  "proforge.stageName.intake": "Acquisizione e diagnosi",
+  "proforge.stageName.lineProse": "Riga e prosa",
+  "proforge.stageName.production": "Produzione",
+  "proforge.stageName.proof": "Correzione di bozze",
+  "proforge.stageName.publishing": "Pubblicazione",
+  "proforge.stageName.structural": "Modifica strutturale",
   "progress.allTime.label": "parole scritte nelle sessioni",
   "progress.allTime.title": "Totale storico",
   "progress.bestDay.title": "Giorno migliore",
diff --git a/services/proForge/applyReviewEdits.ts b/services/proForge/applyReviewEdits.ts
index 545dbeaf6..2ba49d13a 100644
--- a/services/proForge/applyReviewEdits.ts
+++ b/services/proForge/applyReviewEdits.ts
@@ -38,6 +38,36 @@ function isValidRange(range: { start: number; end: number }, len: number): boole
   );
 }
 
+/**
+ * Find the start offset of the occurrence of `needle` in `content` that is closest to `preferNear`
+ * (the stale original range start, when known) and does not overlap an already-planned edit.
+ * Returns -1 if no free occurrence exists. Keeps duplicate-phrase application deterministic.
+ */
+function nearestFreeOccurrence(
+  content: string,
+  needle: string,
+  preferNear: number | undefined,
+  planned: PlannedEdit[],
+): number {
+  let best = -1;
+  let bestDistance = Number.POSITIVE_INFINITY;
+  let idx = content.indexOf(needle);
+  while (idx !== -1) {
+    const end = idx + needle.length;
+    const overlaps = planned.some((p) => idx < p.end && end > p.start);
+    if (!overlaps) {
+      // No range hint → first free occurrence; otherwise the one nearest the original position.
+      const distance = preferNear === undefined ? idx : Math.abs(idx - preferNear);
+      if (distance < bestDistance) {
+        best = idx;
+        bestDistance = distance;
+      }
+    }
+    idx = content.indexOf(needle, idx + 1);
+  }
+  return best;
+}
+
 /**
  * Apply the accepted `ReviewItem`s that target a single section's content.
  * Only items carrying a `proposed` replacement are considered text edits; advisory items
@@ -62,11 +92,13 @@ export function applyReviewEditsToSection(content: string, items: ReviewItem[]):
       }
     }
 
-    // Strategy 2: offsets missing or stale — locate the original text directly.
+    // Strategy 2: offsets missing or stale — locate the original text directly. To stay
+    // deterministic when the same phrase appears multiple times, pick the occurrence nearest the
+    // (stale) original range and skip occurrences already claimed by another planned edit.
     if (item.original) {
-      const idx = content.indexOf(item.original);
-      if (idx !== -1) {
-        planned.push({ start: idx, end: idx + item.original.length, proposed });
+      const target = nearestFreeOccurrence(content, item.original, item.range?.start, planned);
+      if (target !== -1) {
+        planned.push({ start: target, end: target + item.original.length, proposed });
         continue;
       }
     }
diff --git a/services/proForge/pipelineAgents/baseAgent.ts b/services/proForge/pipelineAgents/baseAgent.ts
index b30589d62..b603d9bfa 100644
--- a/services/proForge/pipelineAgents/baseAgent.ts
+++ b/services/proForge/pipelineAgents/baseAgent.ts
@@ -3,7 +3,7 @@
  * QNBS-v3: Eliminates the ~30-line constructor/setup skeleton duplicated across 8 agents.
  */
 
-import type { StageResult } from '../../../features/proForge/types';
+import type { PipelineStage, StageResult } from '../../../features/proForge/types';
 import type { AIProvider, AiModel } from '../../../types';
 import { type InferenceGateway, inferenceGateway } from '../../ai/inferenceGateway';
 import type { AIRequestOptions } from '../../aiProviderService';
@@ -43,6 +43,25 @@ export abstract class BaseAgent {
     return getMemoryBank(this.context.projectId);
   }
 
+  // QNBS-v3: Assemble memory context for a stage. Honours the run's ragMode — a story-anchored
+  // query (title/logline/genre) drives semantic/hybrid relevance ranking of lore/character/prior
+  // entries; with ragMode 'lexical' (or no project) it degrades to keyword/chronological recall.
+  protected async gatherMemoryContext(stage: PipelineStage, maxChars: number): Promise<string> {
+    const project = this.context.getState().project.present?.data;
+    const query = project
+      ? [project.title, project.logline, this.context.config.genrePreset]
+          .filter(Boolean)
+          .join(' ')
+          .trim()
+      : '';
+    return this.getMemoryBank().buildContextString(
+      stage,
+      query || undefined,
+      maxChars,
+      this.context.config.ragMode,
+    );
+  }
+
   protected elapsed(startTime: number): number {
     return Math.round(performance.now() - startTime);
   }
diff --git a/services/proForge/pipelineAgents/copyEditAgent.ts b/services/proForge/pipelineAgents/copyEditAgent.ts
index 5dcfe1ec9..91a724e72 100644
--- a/services/proForge/pipelineAgents/copyEditAgent.ts
+++ b/services/proForge/pipelineAgents/copyEditAgent.ts
@@ -27,7 +27,7 @@ export class CopyEditAgent extends BaseAgent {
     const { config } = this.context;
     const project = this.requireProject();
     const memoryBank = this.getMemoryBank();
-    const memoryContext = await memoryBank.buildContextString('copyEdit', undefined, 2000);
+    const memoryContext = await this.gatherMemoryContext('copyEdit', 2000);
 
     const sections = project.manuscript;
     // Process up to 3 sections for MVP
diff --git a/services/proForge/pipelineAgents/diagnosticAgent.ts b/services/proForge/pipelineAgents/diagnosticAgent.ts
index 2cc0a0acf..d779aa5e7 100644
--- a/services/proForge/pipelineAgents/diagnosticAgent.ts
+++ b/services/proForge/pipelineAgents/diagnosticAgent.ts
@@ -36,7 +36,7 @@ export class DiagnosticAgent extends BaseAgent {
     const avgSectionLength = sections.length > 0 ? Math.round(totalWords / sections.length) : 0;
 
     // Build tool context from memory bank
-    const memoryContext = await memoryBank.buildContextString('intake', undefined, 3000);
+    const memoryContext = await this.gatherMemoryContext('intake', 3000);
 
     // Build the diagnostic prompt
     const manuscriptExcerpt = sections
diff --git a/services/proForge/pipelineAgents/proofAgent.ts b/services/proForge/pipelineAgents/proofAgent.ts
index a7f611193..ad84907b3 100644
--- a/services/proForge/pipelineAgents/proofAgent.ts
+++ b/services/proForge/pipelineAgents/proofAgent.ts
@@ -26,7 +26,7 @@ export class ProofAgent extends BaseAgent {
     const { config } = this.context;
     const project = this.requireProject();
     const memoryBank = this.getMemoryBank();
-    const memoryContext = await memoryBank.buildContextString('proof', undefined, 2000);
+    const memoryContext = await this.gatherMemoryContext('proof', 2000);
 
     // Build full manuscript text (truncated for token limits)
     const sections = project.manuscript;
diff --git a/services/proForge/pipelineAgents/proseAgent.ts b/services/proForge/pipelineAgents/proseAgent.ts
index a1f016d36..b00536157 100644
--- a/services/proForge/pipelineAgents/proseAgent.ts
+++ b/services/proForge/pipelineAgents/proseAgent.ts
@@ -51,7 +51,7 @@ export class ProseAgent extends BaseAgent {
     const { config } = this.context;
     const project = this.requireProject();
     const memoryBank = this.getMemoryBank();
-    const memoryContext = await memoryBank.buildContextString('lineProse', undefined, 2000);
+    const memoryContext = await this.gatherMemoryContext('lineProse', 2000);
 
     // Process sections in batches to avoid token limits
     const sections = project.manuscript;
diff --git a/services/proForge/pipelineAgents/publishingAgent.ts b/services/proForge/pipelineAgents/publishingAgent.ts
index 1454e2bc2..32900cafa 100644
--- a/services/proForge/pipelineAgents/publishingAgent.ts
+++ b/services/proForge/pipelineAgents/publishingAgent.ts
@@ -21,7 +21,7 @@ export class PublishingAgent extends BaseAgent {
     const { config } = this.context;
     const project = this.requireProject();
     const memoryBank = this.getMemoryBank();
-    const memoryContext = await memoryBank.buildContextString('publishing', undefined, 2000);
+    const memoryContext = await this.gatherMemoryContext('publishing', 2000);
 
     const totalWords = project.manuscript.reduce(
       (acc, s) => acc + (s.content?.trim().split(/\s+/).length ?? 0),
diff --git a/services/proForge/pipelineAgents/structuralAgent.ts b/services/proForge/pipelineAgents/structuralAgent.ts
index 1d0120442..6120910b8 100644
--- a/services/proForge/pipelineAgents/structuralAgent.ts
+++ b/services/proForge/pipelineAgents/structuralAgent.ts
@@ -26,7 +26,7 @@ export class StructuralAgent extends BaseAgent {
     const { config } = this.context;
     const project = this.requireProject();
     const memoryBank = this.getMemoryBank();
-    const memoryContext = await memoryBank.buildContextString('structural', undefined, 3000);
+    const memoryContext = await this.gatherMemoryContext('structural', 3000);
 
     // Get diagnostic report from memory
     const diagnosticEntries = await memoryBank.recall('meta');
diff --git a/services/proForge/proForgeHistoryStore.ts b/services/proForge/proForgeHistoryStore.ts
index abc13d057..b46deebda 100644
--- a/services/proForge/proForgeHistoryStore.ts
+++ b/services/proForge/proForgeHistoryStore.ts
@@ -19,7 +19,13 @@ function openHistoryDb(): Promise<IDBDatabase> {
   if (dbPromise) return dbPromise;
   dbPromise = new Promise((resolve, reject) => {
     const request = indexedDB.open(HISTORY_DB, HISTORY_VERSION);
-    request.onerror = () => reject(new Error('Failed to open ProForge history DB'));
+    request.onerror = () => {
+      // QNBS-v3: Don't memoize a rejected promise — a transient open failure (quota, locked DB)
+      // must not disable run-history for the rest of the session. Clear the cache so later
+      // calls retry the open.
+      dbPromise = null;
+      reject(new Error('Failed to open ProForge history DB'));
+    };
     request.onsuccess = () => resolve(request.result);
     request.onupgradeneeded = (event) => {
       const db = (event.target as IDBOpenDBRequest).result;
@@ -42,9 +48,12 @@ export async function saveRunHistory(projectId: string, runs: PipelineRun[]): Pr
   const record: HistoryRecord = { projectId, runs: runs.slice(0, MAX_RUN_HISTORY) };
   await new Promise<void>((resolve, reject) => {
     const tx = db.transaction(STORE, 'readwrite');
-    const request = tx.objectStore(STORE).put(record);
-    request.onsuccess = () => resolve();
-    request.onerror = () => reject(new Error('Failed to save ProForge run history'));
+    tx.objectStore(STORE).put(record);
+    // QNBS-v3: Resolve only once the transaction COMMITS — an IDB write is not durable on
+    // request.onsuccess; the tx can still abort (quota/commit failure) afterwards.
+    tx.oncomplete = () => resolve();
+    tx.onerror = () => reject(new Error('Failed to save ProForge run history'));
+    tx.onabort = () => reject(new Error('ProForge run-history transaction aborted'));
   });
 }
 
diff --git a/tests/unit/proForge/applyReviewEdits.test.ts b/tests/unit/proForge/applyReviewEdits.test.ts
index edadf685e..324816217 100644
--- a/tests/unit/proForge/applyReviewEdits.test.ts
+++ b/tests/unit/proForge/applyReviewEdits.test.ts
@@ -82,6 +82,27 @@ describe('applyReviewEditsToSection', () => {
     expect(result.applied).toBe(1);
     expect(result.skipped).toBe(1);
   });
+
+  it('anchors a stale-offset edit to the occurrence nearest the original range', () => {
+    const content = 'cat dog cat dog cat';
+    // 'cat' occurs at offsets 0, 8, 16; the stale range points near the middle one.
+    const result = applyReviewEditsToSection(content, [
+      item({ range: { start: 7, end: 10 }, original: 'cat', proposed: 'COW' }),
+    ]);
+    expect(result.content).toBe('cat dog COW dog cat');
+    expect(result.applied).toBe(1);
+    expect(result.skipped).toBe(0);
+  });
+
+  it('applies duplicate-phrase edits to distinct occurrences (no double-anchoring)', () => {
+    const content = 'go go go';
+    const result = applyReviewEditsToSection(content, [
+      item({ id: '1', original: 'go', proposed: 'A' }),
+      item({ id: '2', original: 'go', proposed: 'B' }),
+    ]);
+    expect(result.applied).toBe(2);
+    expect(result.content).toBe('A B go');
+  });
 });
 
 describe('planAcceptedManuscriptEdits', () => {
diff --git a/tests/unit/proForge/pipelineAgents/diagnosticAgent.test.ts b/tests/unit/proForge/pipelineAgents/diagnosticAgent.test.ts
index 0819e3899..2ef9ff0eb 100644
--- a/tests/unit/proForge/pipelineAgents/diagnosticAgent.test.ts
+++ b/tests/unit/proForge/pipelineAgents/diagnosticAgent.test.ts
@@ -237,7 +237,13 @@ describe('DiagnosticAgent', () => {
       const agent = new DiagnosticAgent(makeContext());
       await agent.execute(new AbortController().signal);
 
-      expect(mockMemoryBank.buildContextString).toHaveBeenCalledWith('intake', undefined, 3000);
+      // QNBS-v3: story-anchored query + the run's ragMode are threaded so semantic/hybrid recall is honoured.
+      expect(mockMemoryBank.buildContextString).toHaveBeenCalledWith(
+        'intake',
+        expect.stringContaining('Test Novel'),
+        3000,
+        'hybrid',
+      );
     });
 
     it('saves diagnosticReport to memoryBank under category "meta"', async () => {
diff --git a/tests/unit/proForge/pipelineAgents/publishingAgent.test.ts b/tests/unit/proForge/pipelineAgents/publishingAgent.test.ts
index a7af1b147..16434ad25 100644
--- a/tests/unit/proForge/pipelineAgents/publishingAgent.test.ts
+++ b/tests/unit/proForge/pipelineAgents/publishingAgent.test.ts
@@ -201,7 +201,13 @@ describe('PublishingAgent', () => {
       const agent = new PublishingAgent(makeContext());
       await agent.execute(new AbortController().signal);
 
-      expect(mockMemoryBank.buildContextString).toHaveBeenCalledWith('publishing', undefined, 2000);
+      // QNBS-v3: story-anchored query + the run's ragMode are threaded so semantic/hybrid recall is honoured.
+      expect(mockMemoryBank.buildContextString).toHaveBeenCalledWith(
+        'publishing',
+        expect.stringContaining('My Novel'),
+        2000,
+        'hybrid',
+      );
     });
 
     it('saves publishingPackage to memoryBank under "meta" category', async () => {
diff --git a/tests/unit/proForge/pipelineAgents/structuralAgent.test.ts b/tests/unit/proForge/pipelineAgents/structuralAgent.test.ts
index edcc193d8..a7007748f 100644
--- a/tests/unit/proForge/pipelineAgents/structuralAgent.test.ts
+++ b/tests/unit/proForge/pipelineAgents/structuralAgent.test.ts
@@ -220,7 +220,13 @@ describe('StructuralAgent', () => {
       const agent = new StructuralAgent(makeContext());
       await agent.execute(new AbortController().signal);
 
-      expect(mockMemoryBank.buildContextString).toHaveBeenCalledWith('structural', undefined, 3000);
+      // QNBS-v3: gatherMemoryContext now passes a project-derived query + ragMode (Comment #5).
+      expect(mockMemoryBank.buildContextString).toHaveBeenCalledWith(
+        'structural',
+        expect.stringContaining('Structural Test'),
+        3000,
+        'hybrid',
+      );
     });
 
     it('calls memoryBank.recall("meta") to retrieve diagnostic report', async () => {