Municipal 311 Citizen Triage Multi-Agent Routing Benchmark Suite
Create an end-to-end benchmark script to stress-test and evaluate multi-agent civic response workflows across edge-case emergencies.
Use this template when auditing multi-agent conversational pipelines deployed across public 311 citizen helplines. It outputs an automated simulation script measuring routing fidelity, escalation safety compliance, and latency.
Role: Chief Civic Technology Evaluator specializing in multi-agent routing systems and emergency dispatch workflows.
Context
- Municipal Jurisdiction: {{municipality_jurisdiction}}
- Evaluation Corpus: {{citizen_inquiry_dataset}}
- Department Directory: {{department_routing_targets}}
- Safety & Escalation Protocols: {{escalation_safety_rules}}
- Performance SLA: {{latency_sla_ms}}
- Framework Engine: {{benchmark_framework}}
Task
Construct an automated multi-agent triage evaluation script that stress-tests civic routing accuracy, emergency escalation safety, and latency performance across {{municipality_jurisdiction}} citizen support pipelines.
Method
- Load and parse the {{citizen_inquiry_dataset}} into categorized synthetic citizen interaction scenarios.
- Define expected routing topology based on {{department_routing_targets}}.
- Construct asynchronous concurrent evaluation workers utilizing {{benchmark_framework}}.
- Implement critical path safety assertion checks derived from {{escalation_safety_rules}} for emergency triggers.
- Measure and record per-request round-trip latency, evaluating against {{latency_sla_ms}}.
- Compute confusion matrices and routing precision/recall scores across municipal service boundaries.
- Generate an automated markdown evaluation scorecard detailing routing bottlenecks and safety violations.
Constraints
- MUST abort execution or flag a critical failure if any emergency trigger in {{escalation_safety_rules}} is misrouted.
- MUST NOT allow average latency to pass if it exceeds {{latency_sla_ms}}.
- Script must be fully asynchronous to simulate real-world municipal load conditions.
- Use clean error handling for resilient execution during edge-case ingestion.
Output format
Provide a comprehensive {{benchmark_framework}} benchmark script structured as:
- Initialization and Async Client Configuration.
- Dataset Ingestion and Emergency Scenario Fixtures.
- Routing Precision and Safety Protocol Evaluator Engine.
- Concurrent Execution Runner and Markdown Scorecard Generator.
Self-review
- Ensure emergency escalation rules from {{escalation_safety_rules}} are prioritized over general routing.
- Validate that latency tracking accurately captures p95 and p99 against {{latency_sla_ms}}.
- Verify that the async worker pool scales cleanly across {{citizen_inquiry_dataset}}.
Explicit role, a named task, and discrete steps the model can follow.
Background, inputs and variables the model needs before it starts.
Hard boundaries — what the model must and must not do.
A named, field-level shape for the response.
Ordered work items that force analysis before an answer.
Length and structure that travel across frontier models.
Signal density — instruction weight without padding.
Documented variables so the scaffold adapts to new inputs.
Quality bar, assumptions and behaviour when inputs are thin.
How much real usage the template has behind it.