diff --git a/.obsidian/graph.json b/.obsidian/graph.json
index 5e7dccc..e45bdf5 100644
--- a/.obsidian/graph.json
+++ b/.obsidian/graph.json
@@ -17,6 +17,6 @@
   "repelStrength": 10,
   "linkStrength": 1,
   "linkDistance": 250,
-  "scale": 0.29109721960434015,
+  "scale": 0.7932897629230251,
   "close": true
 }
\ No newline at end of file
diff --git a/01 Projects/semblance/DEVELOPER_MANUAL.md b/01 Projects/semblance/DEVELOPER_MANUAL.md
new file mode 100644
index 0000000..159f5a2
--- /dev/null
+++ b/01 Projects/semblance/DEVELOPER_MANUAL.md	
@@ -0,0 +1,180 @@
+---
+auto_generated: true
+manual_updated_at: 2026-05-18
+modified: 2026-05-18
+---
+
+# Semblance Synthetic Society — Developer Manual
+
+## Architecture Overview
+
+Semblance is a full-stack web application with a React frontend and a Node.js backend, communicating with a MongoDB database and external LLM providers. The application is containerized using Docker and Docker Compose for development and production environments.
+
+### High-Level Diagram
+```
+[Frontend (React/TS)] <--> [Backend (Node/Express)] <--> [MongoDB]
+       ^                           ^
+       |                           |
+[LLM Providers]          [Auth Middleware]
+```
+
+### Tech Stack
+- **Frontend:**
+  - React 18+ with TypeScript
+  - React Router DOM for routing
+  - TanStack Query for server state management
+  - Socket.IO Client for real-time WebSocket communication
+  - Sonner for toast notifications
+  - Azure MSAL React for Microsoft Authentication
+  - Tailwind CSS (implied by UI component structure)
+- **Backend:**
+  - Node.js
+  - Express (implied by API structure)
+  - Mongoose (MongoDB ODM)
+  - Socket.IO Server
+  - JWT for local authentication
+- **Infrastructure:**
+  - Docker & Docker Compose
+  - MongoDB 7
+
+## Local Setup
+
+### Prerequisites
+- Node.js 20+
+- Docker and Docker Compose
+- npm or yarn
+
+### Installation Steps
+
+1. **Clone the Repository**
+   ```bash
+   git clone <repo-url>
+   cd semblance
+   ```
+
+2. **Configure Environment Variables**
+   - Copy `.env.example` to `backend/.env`:
+     ```bash
+     cp backend/.env.example backend/.env
+     ```
+   - Edit `backend/.env` to set:
+     - `MONGO_URI`: Connection string to MongoDB (defaults to `mongodb://mongo:27017/semblance_db` in Docker).
+     - `JWT_SECRET`: A strong secret for JWT token signing.
+     - `LLM_API_KEYS`: API keys for Google Gemini and OpenAI.
+     - `MSAL_CONFIG`: Client ID and tenant ID for Microsoft Authentication.
+
+3. **Start Infrastructure**
+   ```bash
+   docker-compose up -d mongo
+   ```
+   This starts the MongoDB service with persistent volumes.
+
+4. **Build and Start Backend**
+   ```bash
+   docker-compose up --build backend
+   ```
+   The backend will connect to MongoDB and start on port `5137`.
+
+5. **Build and Start Frontend**
+   ```bash
+   docker-compose --profile build up --build frontend
+   ```
+   The frontend builds and serves the output to `./backend/dist-out` (or your configured dist directory). For local development with hot reload, you may need to run the frontend locally with `npm run dev` instead of the Docker build profile.
+
+6. **Access the Application**
+   - Frontend: `http://localhost:5173` (if running locally) or via the configured proxy.
+   - Backend API: `http://localhost:5137`.
+   - MongoDB: `127.0.0.1:27017`.
+
+## Environment Variables
+
+| Variable | Description | Required |
+|----------|-------------|----------|
+| `MONGO_URI` | MongoDB connection string | Yes |
+| `JWT_SECRET` | Secret key for JWT signing | Yes |
+| `GEMINI_API_KEY` | Google Gemini API key | Yes |
+| `OPENAI_API_KEY` | OpenAI API key | Yes |
+| `MSAL_CLIENT_ID` | Azure AD App Client ID | Yes |
+| `MSAL_TENANT_ID` | Azure AD Tenant ID | Yes |
+| `NODE_ENV` | Environment (development/production) | No |
+
+## Key Services & Entry Points
+
+### Frontend Entry Points
+- **`src/main.tsx`**: Root component initialization.
+- **`src/App.tsx`**: Defines routing structure, providers (Auth, WebSocket, QueryClient), and route definitions.
+  - **Routes:**
+    - `/`: Index page.
+    - `/login`: Authentication page.
+    - `/synthetic-users`: Persona management.
+    - `/focus-groups`: Session management.
+    - `/dashboard`: Overview analytics.
+    - `/admin`: Admin panel (protected).
+- **`src/contexts/`**:
+  - `AuthContext.tsx`: Handles local login and Microsoft SSO state. Uses `localStorage` for token persistence.
+  - `NavigationContext.tsx`: Manages app state for navigation history and focus group context, persisted in `localStorage`.
+  - `WebSocketContextNew.tsx`: Initializes a singleton Socket.IO instance. Uses a module-level `socketInitialized` flag to prevent re-initialization.
+
+### Backend Entry Points
+- **`backend/`**: Contains the Express server, routes, and controllers.
+- **Services:**
+  - `llm`: Handles interactions with Gemini and OpenAI models.
+  - `aiRunner`: Orchestrates autonomous focus group logic.
+  - `themeExtractor`: Processes session transcripts for key themes.
+
+### WebSocket Implementation
+- **Service:** `src/services/websocketServiceNew.ts` (referenced in context).
+- **Pattern:** Singleton pattern via `socketInitialized` flag.
+- **Authentication:** Socket connection is authenticated via JWT token passed in the `auth_token` from `localStorage` or the `AuthContext`.
+
+## API Reference
+
+### Authentication Endpoints
+- `POST /api/auth/login`: Local login. Expects `{ username, password }`. Returns JWT.
+- `POST /api/auth/microsoft`: Microsoft SSO callback. Expects ID token. Returns JWT and user details.
+
+### Persona Endpoints
+- `GET /api/personas`: List all personas.
+- `POST /api/personas`: Create a new persona.
+- `PUT /api/personas/:id`: Update a persona.
+- `GET /api/personas/:id`: Get persona details.
+
+### Focus Group Endpoints
+- `POST /api/focus-groups`: Create a session.
+- `GET /api/focus-groups/:id/stream`: WebSocket events for session data.
+
+### Admin Endpoints
+- `GET /api/admin/users`: List users.
+- `DELETE /api/admin/users/:id`: Remove a user.
+
+## Deployment
+
+### Production Build
+1. Ensure `MONGO_URI` points to the production MongoDB instance.
+2. Set `NODE_ENV=production`.
+3. Build the frontend using the Docker compose build profile:
+   ```bash
+   docker-compose --profile build up --build frontend
+   ```
+4. Deploy the backend container and frontend static assets to your web server (e.g., Nginx) or use a PaaS.
+
+### Docker Compose Notes
+- The `frontend` service in `docker-compose.yml` is configured for **build** only. It copies the built assets to `/app/dist-out`. The backend should serve these static files in production.
+- Volumes are used for MongoDB data persistence and backend uploads.
+
+## Known Gotchas
+
+1. **WebSocket Singleton Issue:**
+   The `WebSocketContextNew.tsx` uses a module-level `socketInitialized` flag. If the app is hot-reloaded in development, the flag may not reset, causing connection issues. In production, this is generally safe, but ensure server-side disconnects are handled if the client navigates away.
+
+2. **Navigation State Persistence:**
+   The `NavigationContext` persists state to `localStorage`. If a user switches browsers or clears storage, navigation history (like "Previous Route") will be lost. This may cause back-button issues if not handled gracefully in UI components.
+
+3. **Microsoft SSO Redirect:**
+   The MSAL `handleRedirectPromise` is called in `AuthContext` on mount. Ensure the redirect URI is correctly registered in Azure AD and matches the application's configuration.
+
+4. **Mention Parsing Regex:**
+   The `parseMentions` function uses a specific regex for `@mentions`. Be aware that complex names with special characters or punctuation may not be parsed correctly. The regex stops at conjunctions (`and`, `or`) and non-word boundaries.
+
+5. **Port Binding:**
+   Backend and MongoDB are bound to `127.0.0.1` in `docker-compose.yml`. This ensures they are not exposed to the public internet, but requires port forwarding or a reverse proxy for remote access.
\ No newline at end of file
diff --git a/01 Projects/semblance/USER_MANUAL.md b/01 Projects/semblance/USER_MANUAL.md
new file mode 100644
index 0000000..2320596
--- /dev/null
+++ b/01 Projects/semblance/USER_MANUAL.md	
@@ -0,0 +1,100 @@
+---
+auto_generated: true
+manual_updated_at: 2026-05-18
+modified: 2026-05-18
+---
+
+# Semblance Synthetic Society — User Manual
+
+## What This Tool Does
+
+Semblance is an AI-powered platform designed to simulate consumer behavior through synthetic personas. It allows researchers, marketers, and product teams to create realistic AI-driven user profiles and run autonomous or moderated focus group sessions. The platform extracts actionable insights, themes, and analytics from these simulated conversations in real-time.
+
+### Key Capabilities
+- **Synthetic Persona Management:** Create, edit, and organize AI-generated or manually built consumer profiles.
+- **Autonomous Focus Groups:** Run self-driving conversations where AI personas interact with each other based on research objectives.
+- **Moderated Sessions:** Act as a human moderator while AI assists with probes and follow-up questions.
+- **Real-Time Insights:** View live transcripts, theme extraction, and sentiment analysis during sessions.
+
+## Who Uses It
+
+- **Market Researchers:** To test concepts, messages, or products without recruiting real participants.
+- **Product Managers:** To validate feature ideas and user flows with synthetic user personas.
+- **UX Researchers:** To run rapid, low-cost qualitative research rounds.
+- **Marketing Teams:** To refine messaging and identify key themes from consumer perspectives.
+
+## How to Access
+
+1. Navigate to the application URL provided by your organization.
+2. Log in using one of the supported methods:
+   - **Local Login:** Enter your registered username and password.
+   - **Microsoft SSO:** Click the "Sign in with Microsoft" button to authenticate via your corporate Azure AD account.
+3. Once authenticated, you will be directed to the Dashboard or the last viewed page.
+
+## Main Workflows
+
+### Workflow 1: Creating Synthetic Personas
+
+1. **Navigate to Synthetic Users:**
+   - Click "Synthetic Users" in the left-hand navigation bar.
+
+2. **Create a New Persona:**
+   - Click the "Create Persona" button.
+   - **Option A: AI Generation:**
+     - Select "Generate with AI."
+     - Provide a brief description of the target audience (e.g., "Millennial parents interested in organic skincare").
+     - Choose an LLM model (Default: Google Gemini 3 Pro; Options: OpenAI GPT-4.1, GPT-5.2).
+     - Click "Generate." The AI will populate demographic, psychographic, and behavioral attributes.
+   - **Option B: Manual Creation:**
+     - Select "Manual Creation."
+     - Fill in attributes such as name, age, occupation, interests, and behavioral traits manually.
+
+3. **Refine and Save:**
+   - Adjust any AI-generated details if necessary.
+   - Save the persona.
+
+4. **Organize Personas:**
+   - Use drag-and-drop to move personas into folders for better organization.
+
+5. **Export Profiles:**
+   - Select one or more personas.
+   - Click "Export" and choose PDF or summary format to download the persona profiles.
+
+### Workflow 2: Running a Focus Group Session
+
+1. **Start a Session:**
+   - Navigate to "Focus Groups" and click "Create Session."
+   - Select the personas to participate.
+
+2. **Configure Session Settings:**
+   - **Mode:** Choose between **Manual Moderation** (you guide the chat) or **Autonomous AI Mode** (AI drives the conversation).
+   - **Discussion Guide:** Optionally, ask the AI to generate a structured discussion guide based on your research objectives.
+
+3. **Run the Session:**
+   - **Autonomous Mode:** Click "Start Session." Watch as AI personas converse in real-time via WebSocket. The AI moderator may intervene with probes or follow-up questions.
+   - **Manual Mode:** Type questions or prompts to guide the personas. Use the AI Assistant panel for suggested follow-ups.
+
+4. **Analyze Results:**
+   - During the session, observe real-time theme extraction and highlighting.
+   - After the session ends, review the session analytics, including summary generation, key themes, and sentiment scores.
+
+### Workflow 3: Managing Your Account
+
+1. **Check Usage:**
+   - Go to "My Usage" to view your token consumption and API limits.
+2. **Admin Tasks (Admins Only):**
+   - Access the "Admin" panel to manage users, system settings, and monitor system health.
+
+## FAQ
+
+**Q: How accurate are the synthetic personas?**
+A: Personas are generated using multi-model LLMs (Gemini, GPT-4.1, GPT-5.2). You can refine them manually after generation to better fit your specific research needs.
+
+**Q: Can I interrupt an autonomous session?**
+A: Yes. Even in Autonomous Mode, if you have moderator privileges, you can type prompts to steer the conversation.
+
+**Q: What data is stored?**
+A: Persona data, session transcripts, and analytics are stored in the MongoDB database. Ensure you comply with your organization's data privacy policies when creating personas.
+
+**Q: How do I export session insights?**
+A: Session summaries and theme extractions are available post-session. You can often export these as reports or view them in the analytics dashboard.
\ No newline at end of file
diff --git a/01 Projects/video-accessibility-old/DEVELOPER_MANUAL.md b/01 Projects/video-accessibility-old/DEVELOPER_MANUAL.md
new file mode 100644
index 0000000..579d052
--- /dev/null
+++ b/01 Projects/video-accessibility-old/DEVELOPER_MANUAL.md	
@@ -0,0 +1,176 @@
+---
+auto_generated: true
+manual_updated_at: 2026-05-18
+modified: 2026-05-18
+---
+
+# Developer Manual: Accessible Video Processing Platform
+
+## Architecture Overview
+
+The platform follows a microservices-oriented architecture deployed via Docker Compose, separated into frontend, API, and background workers.
+
+### High-Level Diagram
+
+```mermaid
+graph TD
+    User[User Browser] -->|HTTPS/WS| Frontend[Frontend (React)]
+    Frontend -->|REST API| API[FastAPI/Gunicorn]
+    Frontend -->|WebSocket| API
+    API -->|Task Publish| Redis[(Redis Broker)]
+    API -->|Cache/Session| Redis
+    API -->|Data Query| MongoDB[(MongoDB)]
+    Redis -->|Consume| Worker[Cellery Workers]
+    Worker -->|Process| FFmpeg[FFmpeg Worker]
+    Worker -->|Process| TTS[TTS Worker]
+    Worker -->|Process| Whisper[Whisper Worker]
+    Worker -->|Write| MongoDB
+```
+
+## Tech Stack
+
+-   **Frontend**: React 18+, TypeScript, Tailwind CSS, React Query, MSAL (Microsoft Authentication).
+-   **Backend**: Python, FastAPI, Gunicorn, Celery.
+-   **Database**: MongoDB 7.0.
+-   **Broker/Cache**: Redis 7.
+-   **AI/ML**: Gemini 2.5 Pro (Gemini API), Whisper (Speech-to-Text), Google Cloud TTS, ElevenLabs.
+-   **Monitoring**: Sentry (Error/Performance Tracking).
+-   **Infrastructure**: Docker, Docker Compose.
+
+## Local Setup
+
+### Prerequisites
+-   Docker and Docker Compose.
+-   Python 3.10+.
+-   Node.js 18+.
+-   Google Cloud API Credentials (for Translate, TTS).
+-   Google AI Studio API Key (for Gemini).
+-   Microsoft Entra ID (Azure AD) App Registration (for MSAL).
+
+### Step 1: Environment Variables
+Create a `.env` file in the root directory based on `.env.example`.
+
+```bash
+# Database
+MONGODB_DB=accessible_video
+
+# Redis
+REDIS_URL=redis://redis:6379/0
+
+# AI Services
+GEMINI_API_KEY=your_gemini_key
+GOOGLE_CLOUD_PROJECT_ID=your_project_id
+GOOGLE_APPLICATION_CREDENTIALS=./path/to/key.json
+
+# Auth
+MSAL_CLIENT_ID=your_azure_client_id
+MSAL_TENANT_ID=your_azure_tenant_id
+
+# Sentry
+SENTRY_DSN=your_sentry_dsn
+```
+
+### Step 2: Start Infrastructure
+```bash
+docker-compose up -d mongodb redis
+```
+
+### Step 3: Backend Setup
+```bash
+cd backend
+pip install -r requirements.txt
+# Copy config if necessary
+cp config/mongod.conf.example config/mongod.conf
+```
+
+### Step 4: Frontend Setup
+```bash
+cd frontend
+npm install
+# Ensure .env.development or .env.local is configured with VITE_API_URL
+npm run dev
+```
+
+### Step 5: Start Workers
+In separate terminal windows:
+```bash
+# Default Worker
+celery -A celery_app worker -Q default,ingest,notify,render -c 4
+
+# Dedicated Workers
+celery -A celery_app worker -Q tts -c 8
+celery -A celery_app worker -Q ffmpeg -c 1
+celery -A celery_app worker -Q whisper -c 1
+```
+
+## Environment Variables
+
+| Variable | Description | Required |
+| :--- | :--- | :--- |
+| `MONGODB_DB` | MongoDB database name | Yes |
+| `REDIS_URL` | Redis connection string | Yes |
+| `GEMINI_API_KEY` | API key for Gemini 2.5 Pro | Yes |
+| `MSAL_CLIENT_ID` | Azure AD Client ID | Yes |
+| `MSAL_TENANT_ID` | Azure AD Tenant ID | Yes |
+| `SENTRY_DSN` | Sentry Data Source Name | Optional |
+| `VITE_APP_ENV` | App environment (dev/prod) | Yes |
+
+## Key Services & Entry Points
+
+### 1. Frontend (`frontend/src/main.tsx`)
+-   **Auth**: Initializes MSAL `PublicClientApplication` for Microsoft SSO.
+-   **Monitoring**: Initializes Sentry with PII filtering.
+-   **State**: Provides global contexts for Toasts, Notifications, and WebSocket connections.
+
+### 2. WebSocket (`frontend/src/contexts/GlobalWebSocketContext.tsx`)
+-   Manages real-time job status updates.
+-   Uses `useJobStatusWebSocket` hook.
+-   Dispatches toast notifications based on `getStatusMessageConfig`.
+
+### 3. Backend API (`backend/app/main.py` inferred)
+-   **FastAPI** handles REST requests.
+-   **Celery** offloads heavy processing:
+    -   `ingest`: Video metadata extraction.
+    -   `whisper`: Speech-to-text transcription.
+    -   `render`: TTS synthesis and FFmpeg video compositing.
+    -   `notify`: WebSocket dispatching.
+
+### 4. Database Models (MongoDB)
+-   **Job**: Core entity tracking processing status, assets, and metadata.
+-   **User/Client/Reviewer**: RBAC models.
+-   **AuditLog**: Immutable log of reviewer actions.
+
+## API Reference
+
+*Note: Full OpenAPI spec available at `/docs` in running API.*
+
+### Authentication
+-   **POST /auth/login**: Triggers MSAL redirect flow.
+-   **GET /auth/callback**: Handles MSAL token exchange.
+
+### Jobs
+-   **POST /jobs**: Create a new accessibility job.
+    -   Body: `{ video_id, target_languages, services: ["cc", "ad"] }`
+-   **GET /jobs**: List jobs with filters (status, owner).
+-   **GET /jobs/{job_id}**: Get job details and progress.
+
+### Quality Control
+-   **GET /qc/list**: Get jobs awaiting review.
+-   **PUT /qc/{job_id}/approve**: Approve job.
+-   **PUT /qc/{job_id}/reject**: Reject job with comments.
+-   **PUT /qc/{job_id}/edit-vtt**: Update VTT content.
+
+## Deployment
+
+1.  **Build Images**: Ensure Docker images are up to date.
+2.  **Production Env**: Set `VITE_APP_ENV=production` and `SENTRY_DSN`.
+3.  **Security**: Ensure `HTTPS` is enforced at the reverse proxy (Nginx/AWS ALB).
+4.  **Scale Workers**: Increase concurrency in `docker-compose.yml` or use Kubernetes for horizontal scaling.
+
+## Known Gotchas
+
+1.  **FFmpeg Concurrency**: The `ffmpeg-worker` is configured with `concurrency=1`. Do not scale this worker beyond 1 instance per CPU core to avoid file locking issues.
+2.  **TTS Quotas**: Google Cloud TTS and ElevenLabs have rate limits. Monitor usage and implement backoff strategies in the TTS worker.
+3.  **WebSocket Reconnects**: The frontend handles auto-reconnects. Ensure the backend WebSocket server is stable; unexpected disconnections may require a manual page refresh if reconnect fails.
+4.  **Signed URLs**: Generated download links expire after 24 hours. If a user tries to download later, they must re-request the link from the API.
+5.  **MSAL Config**: Ensure `msalConfig.ts` matches the Azure AD redirect URIs exactly, including the `basename` (`/video-accessibility`) if deployed to a sub-path.
\ No newline at end of file
diff --git a/01 Projects/video-accessibility-old/USER_MANUAL.md b/01 Projects/video-accessibility-old/USER_MANUAL.md
new file mode 100644
index 0000000..7644008
--- /dev/null
+++ b/01 Projects/video-accessibility-old/USER_MANUAL.md	
@@ -0,0 +1,107 @@
+---
+auto_generated: true
+manual_updated_at: 2026-05-18
+modified: 2026-05-18
+---
+
+# User Manual: Accessible Video Processing Platform
+
+## What This Tool Does
+
+The **Accessible Video Processing Platform** is an AI-powered solution designed to make video content inclusive and accessible to diverse audiences. It automates the complex workflow of generating closed captions, audio descriptions (with voiceovers), and multi-language translations for video content.
+
+Instead of manually creating accessibility assets, users upload a video, and the platform uses advanced AI models (Gemini 2.5 Pro, Google Cloud TTS, Whisper) to generate high-quality, synchronized accessibility layers. The platform includes a robust quality control (QC) workflow where human reviewers verify and edit the generated content before final delivery.
+
+## Who Uses It
+
+This platform serves three primary user roles:
+
+1.  **Clients/Content Creators**: Users who upload videos and manage their accessibility projects. They initiate jobs, track progress, and download final accessible assets.
+2.  **Reviewers**: Linguists or accessibility experts who review generated captions and audio descriptions, make necessary edits, and approve the content for delivery.
+3.  **Administrators**: System managers who oversee user accounts, organizational settings, audit logs, and glossary management.
+
+## How to Access
+
+### Prerequisites
+- A web browser (Chrome, Firefox, Edge, or Safari).
+- Valid organizational credentials (Microsoft Account/Enterprise SSO).
+
+### Login
+1. Navigate to the application URL.
+2. Click **Login**. You will be redirected to the Microsoft Authentication Library (MSAL) login screen.
+3. Enter your organizational credentials to sign in.
+4. Upon successful authentication, you will be redirected to the **Dashboard**.
+
+## Main Workflows
+
+### Workflow 1: Creating a New Accessibility Job (For Clients)
+
+1.  **Navigate to Upload**: From the Dashboard, click **New Job**.
+2.  **Upload Video**: 
+    - Drag and drop your video file into the upload area or click to browse.
+    - The system validates the file format and size.
+3.  **Configure Settings**:
+    - **Source Language**: Select the original language of the video.
+    - **Target Languages**: Select one or more languages for translation (supports 50+ languages).
+    - **Services**: Check boxes for required services:
+        - *Closed Captions (CC)*
+        - *Audio Descriptions (AD)*
+        - *Multi-language Translation*
+4.  **Submit Job**: Click **Start Processing**. 
+    - You will be taken to the **Jobs List** where the new job appears.
+    - A toast notification confirms the submission.
+
+### Workflow 2: Tracking Job Progress
+
+1.  **Job List View**: View the status of all your jobs in real-time.
+2.  **Real-Time Updates**: 
+    - Use the **WebSocket** status bar to see live progress.
+    - Toast notifications will pop up when status changes occur (e.g., "Caption Generation Complete").
+3.  **Job Details**: Click on a specific job ID to view:
+    - Detailed progress bars for each AI service (Whisper, Gemini, TTS).
+    - Logs of the processing stages.
+    - Estimated time remaining.
+
+### Workflow 3: Quality Control Review (For Reviewers)
+
+1.  **Access QC Queue**: Navigate to **QC List** from the admin menu.
+2.  **Review Item**: Click on a job marked as **Ready for Review**.
+3.  **Edit Captions**:
+    - Open the **VTT Editor**.
+    - Play the video and read the synchronized captions.
+    - Click on any caption line to edit the text directly.
+    - Save changes to update the VTT file.
+4.  **Review Audio Descriptions**:
+    - Listen to the generated TTS audio.
+    - If the tone is inappropriate or inaccurate, flag the section for re-generation or manual note.
+5.  **Approve/Reject**:
+    - Click **Approve** to send the assets for final delivery.
+    - Click **Reject** with a comment to send it back to AI processing for regeneration.
+6.  **Audit Trail**: All review actions are logged automatically in the **Audit Log** for compliance.
+
+### Workflow 4: Downloading Final Assets
+
+1.  **Navigate to Downloads**: Go to the **Downloads** page.
+2.  **Select Completed Jobs**: Filter by status "Approved".
+3.  **Download Assets**:
+    - Click the download icon next to each asset:
+        - *VTT File*: For captions.
+        - *Audio File*: For audio descriptions.
+    - Assets are delivered via secure, time-limited signed URLs.
+
+## FAQ
+
+**Q: How long does processing take?**
+A: Processing time depends on video length and the number of target languages. Typically, captions take 2-5 minutes per hour of video, while audio descriptions may take longer due to TTS synthesis.
+
+**Q: Can I edit captions before they are processed?**
+A: You can provide a **Glossary** in the settings to influence the AI's terminology. However, the primary editing window is during the Quality Control phase.
+
+**Q: Why is my job stuck on "Processing"?**
+A: Check the **Job Detail** logs. If it remains stuck for more than 10 minutes, try refreshing the page. If the issue persists, contact an Administrator to check the worker queues.
+
+**Q: How are my files secured?**
+A: Files are stored in Google Cloud Storage with signed URLs that expire after 24 hours. The platform also enforces strict Role-Based Access Control (RBAC).
+
+**Q: Who can approve my work?**
+A: Only users with the **Reviewer** or **Admin** role can approve jobs. Clients cannot self-approve their own submissions.
\ No newline at end of file