1. **Urgency Prioritization Benchmark**
   *   **Scenario:** A batch of 10 incoming emails is provided, ranging from critical server outage alerts and deadline warnings to routine newsletter digests and social event invitations.
   *   **Task:** The model must rank the emails on a scale of 1 to 5 based on immediate business criticality and time sensitivity, ensuring that system-critical alerts are flagged above non-urgent informational updates.

2. **Phishing and Malware Detection Benchmark**
   *   **Scenario:** A dataset containing 5 emails: two appear to be legitimate internal policy updates, one is a generic "you've won a prize" message, one contains a suspicious link to a look-alike login domain, and one is a direct message from a known vendor with a PDF attachment.
   *   **Task:** Classify each email as "Safe," "Suspicious," or "Malicious," and identify specific red flags (such as mismatched sender domains or urgent calls to action) for the flagged items.

3. **Departmental Routing Benchmark**
   *   **Scenario:** A stream of 8 incoming inquiries from generic users asking about password resets, invoice discrepancies, vacation policy changes, and product demos.
   *   **Task:** Automatically assign each email to the correct department bucket (IT, Finance, HR, or Sales) based on keyword analysis and context, without generating any actual routing commands.

4. **Actionability Classification Benchmark**
   *   **Scenario:** A set of 6 emails from simulated colleagues. Some emails explicitly request a review or approval, some are purely informational "FYI" updates, and some are meeting cancellations that require no response.
   *   **Task:** Tag each email with one of three action labels: "Requires Response," "Requires Review," or "No Action Needed," helping to simulate an intelligent inbox that highlights only items demanding user intervention.

5. **Customer Sentiment Escalation Benchmark**
   *   **Scenario:** Four support ticket emails from generic customers regarding a delayed shipment. One customer is neutral, one is politely inquiring, one is expressing mild frustration, and one is using aggressive language demanding a refund.
   *   **Task:** Analyze the sentiment of each message and flag the high-intensity negative email for immediate human escalation, while marking the others for standard automated responses.