1. **Time-Sensitive Priority Sorting**
   - **Description:** Evaluate the system's ability to rank incoming messages by urgency based on explicit deadlines, priority markers, and temporal language.
   - **Input Specification:** A batch of 500 synthetic emails with randomized timestamps, subject lines containing phrases like "URGENT: System Alert", "FYI: Monthly Digest", and "ACTION REQUIRED by 17:00 UTC", alongside body text with varying deadline references and escalation cues.
   - **Expected Output:** A ranked list from highest to lowest priority, with each email assigned a priority tier (Critical, High, Medium, Low) and a recommended response SLA.
   - **Evaluation Metrics:** Precision@K, mean reciprocal rank, priority tier accuracy, and ranking latency per 100 messages.

2. **Phishing & Social Engineering Detection**
   - **Description:** Assess classification performance on synthetic emails designed to mimic common attack vectors versus legitimate internal/external communications.
   - **Input Specification:** 1,000 emails with synthetic sender domains (e.g., @secure-auth-portal.net, @internal-hr-sys.com), varying degrees of urgency language, fabricated login URLs, mismatched display names, and randomized trust indicators.
   - **Expected Output:** Multi-class labels (Legitimate, Phishing, Spam, Newsletter) with confidence scores and extracted suspicious indicators.
   - **Evaluation Metrics:** Macro-averaged F1-score, false positive rate (weighted heavily), detection latency, and indicator extraction precision.

3. **Cross-Departmental Routing & Categorization**
   - **Description:** Test the system's capacity to route emails to appropriate functional queues based on semantic content, metadata, and organizational keywords.
   - **Input Specification:** 800 synthetic emails covering topics like invoice discrepancies, software license renewals, employee onboarding, vendor contract updates, and campaign approvals. Each contains department-specific terminology and synthetic routing tags.
   - **Expected Output:** Category assignments (Finance, IT, HR, Legal, Marketing, General) with target queue routing and confidence thresholds.
   - **Evaluation Metrics:** Multi-label classification accuracy, per-class F1-scores, routing latency, and threshold calibration stability.

4. **Conversation Threading & Deduplication**
   - **Description:** Evaluate the system's ability to group related messages into coherent threads, identify exact/near duplicates, and maintain context across replies and forwards.
   - **Input Specification:** 600 emails simulating 150 distinct conversation threads, featuring randomized reply chains, forwarded messages, slight subject variations (Re:, Fw:, Updated:), and synthetic Message-ID/In-Reply-To/References headers.
   - **Expected Output:** Thread cluster IDs, deduplication flags, canonical subject assignment per thread, and reply order reconstruction.
   - **Evaluation Metrics:** Thread recall, precision, duplicate detection accuracy, clustering purity, and processing throughput.

5. **Attachment Compliance & Data Loss Prevention (DLP)**
   - **Description:** Measure the system's ability to inspect email attachments for policy violations, restricted file types, size limits, and synthetic sensitive data patterns.
   - **Input Specification:** 500 emails with synthetic attachments following naming patterns (e.g., "Q3_financials.xlsx", "employee_ssn_list.csv", "marketing_assets.zip", "test_payload.exe"), varying file sizes, and embedded synthetic PII/PCI patterns in metadata and file contents.
   - **Expected Output:** Compliance status (Allowed, Flagged, Blocked), risk category, policy rule matches, and recommended action (quarantine, strip, deliver).
   - **Evaluation Metrics:** DLP detection rate, false positive rate, policy rule coverage, and maximum processing latency per attachment.