1. **Urgency Gradient Classification**
   - Objective: Test an AI or workflow engine's ability to rank incoming messages by time-sensitivity and explicit priority markers.
   - Synthetic Input: Three fictional emails from "[Sender_A]", "[Sender_B]", and "[Sender_C]" regarding "[Project_Orion]". One contains a hard deadline of "2 hours until server maintenance window", one requests feedback by "end of business Friday", and one is a routine monthly industry newsletter.
   - Triage Task: Assign a priority tier (Critical/High/Medium/Low), extract all temporal constraints, and recommend response windows.
   - Evaluation Focus: Correct priority ordering, accurate deadline extraction, and appropriate latency recommendations.

2. **Cross-Functional Action Routing**
   - Objective: Evaluate parsing of multi-domain requests and correct internal delegation.
   - Synthetic Input: A single message from "[Client_D]" containing three distinct asks within one thread: a contract clause revision (Legal), a technical architecture change (Engineering), and a prorated billing adjustment (Finance).
   - Triage Task: Separate the composite request into discrete action items, map each to the correct department/team, and generate internal routing instructions.
   - Evaluation Focus: Request isolation accuracy, correct department mapping, and clear delegation logic.

3. **Ambiguity Resolution & Clarification Drafting**
   - Objective: Assess handling of incomplete, vague, or underspecified communication.
   - Synthetic Input: An email from "[Partner_E]" requesting to "schedule the sync for next week" and "send the deck", with no specific dates, time zones, calendar links, or file references provided.
   - Triage Task: Identify all missing parameters, draft a professional clarification email template, and propose a follow-up cadence.
   - Evaluation Focus: Completeness of missing-field identification, tone appropriateness, and logical follow-up structure.

4. **Multi-Attachment & Nested Request Processing**
   - Objective: Test systematic breakdown of complex messages containing multiple files and conditional actions.
   - Synthetic Input: A message from "[Vendor_F]" with three PDF attachments (invoice, compliance certificate, renewal form), plus instructions to approve the renewal by Friday, forward the invoice to accounting, and archive the compliance certificate.
   - Triage Task: Catalog each attachment with its purpose, extract conditional deadlines, and generate a step-by-step action checklist.
   - Evaluation Focus: Attachment-to-purpose mapping accuracy, deadline extraction, and comprehensive checklist generation.

5. **Security-Aware Filtering & Verification**
   - Objective: Evaluate risk-based triage in a business communication context.
   - Synthetic Input: Four fictional messages: (1) a standard vendor status update, (2) a generic-greeting email with an urgent payment link from a mismatched domain, (3) a verified sender requesting a routine quarterly report, and (4) a slightly altered sender address requesting a credential reset.
   - Triage Task: Classify each as Safe, Suspicious, or High-Risk, provide security reasoning, and recommend next steps (reply, verify via alternate channel, or quarantine).
   - Evaluation Focus: Classification accuracy, precise risk reasoning, and appropriate mitigation recommendations.