1. **Multi-Department Routing:** A high-volume corporate inbox receives messages that must be categorized into specific departments (e.g., Billing, Technical Support, Sales, or Legal). The benchmark tests the model's ability to distinguish between overlapping themes, such as a customer complaining about a technical bug that led to a billing error.

2. **Executive Priority Scoring:** A simulated CEO's inbox containing a mix of internal updates, newsletters, urgent requests from board members, and low-priority networking invites. The model must assign a priority score (1-5) and provide a one-sentence justification for the ranking based on the sender's role and the urgency of the language used.

3. **E-commerce Action Item Extraction:** A set of customer emails regarding orders (e.g., requests for refunds, address changes, or shipping status updates). The benchmark evaluates the model's ability to extract structured data—such as Order IDs, Product SKUs, and requested actions—from unstructured, conversational text.

4. **HR Candidate Intent Classification:** An HR portal receiving emails from job applicants. The model must classify the intent of each email into categories such as "Application Submission," "Interview Rescheduling," "Salary Negotiation," or "Withdrawal of Application," while ignoring irrelevant "Thank You" notes.

5. **IT Help Desk Sentiment & Severity Mapping:** A series of technical incident reports. The model must analyze the sentiment of the user (e.g., frustrated, neutral, panicked) and map the technical description to a severity level (P1 through P4) based on whether the issue affects a single user or the entire organization.