Data Center Power & Community Risk Assessor with Claude Agents SDK and Gentrace
Rising political and community backlash against data center power consumption threatens regional grid stability and development permits. Build an autonomous multi-agent evaluation pipeline using Claude Agents SDK and Gentrace to assess site viability, power load constraints, and public opposition risk. Achieve 90% accuracy in predicting municipal permit approval outcomes.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an agentic site risk assessment pipeline using Claude Agents SDK instrumented with Gentrace to evaluate power grid load and community sentiment.
How work is evaluated
Evaluates candidate agent accuracy in calculating site risk score and permit recommendation.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
recommendation_match
Checks if permit recommendation matches ground truth label.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
risk_score_mae
Mean absolute error of calculated risk score. • target: 5 • range: 0-100
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement multi-agent analysis using Claude Agents SDK
Trace and evaluate AI agent runs using Gentrace pipelines
Parse regional power utility capacity and municipal zoning regulations
Predict local legislative approval probabilities based on public backlash signals
Reference links and supporting material
Synthetic dataset of data center site applications including utility grid headroom, power request MW, and public hearing sentiment transcripts.
Challenge setup
Evaluates candidate agent accuracy in calculating site risk score and permit recommendation.
Challenge input
JSON object with site_id, grid_capacity_mw, requested_mw, transcript_text
Claude Agents SDK
Primary framework required for agentic decision making.
Gentrace
Required tracking framework for AI workflow monitoring.
Evaluated output
JSON object with risk_score (0-100), permit_recommendation ('APPROVED'|'REJECTED')
- Checks if permit recommendation matches ground truth label.
- Mean absolute error of calculated risk score. • target: 5 • range: 0-100
- Benchmark: DataCenterPermitBenchmark
- Risk Score Mae target: 5
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- RAI · Optional
- Ellipsis · Optional
- Gentrace · Required
- Gentrace · Required
Evaluation contract
- recommendation_match · Weight 1
- risk_score_mae · Weight 1
Metrics
- risk_score_mae · Target 5
Benchmarks
- DataCenterPermitBenchmark
- energy
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON object with site_id, grid_capacity_mw, requested_mw, transcript_text
Output: JSON object with risk_score (0-100), permit_recommendation ('APPROVED'|'REJECTED')
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Orchestration
- Gentrace · Evaluation and Tracing
Recipe freeze marker
No recipe freeze time is recorded.