Back to evaluations
Public evaluation

FraudDetectionAccuracy

The agent will be evaluated on its ability to accurately detect simulated fraud events in a stream of transaction data, minimize false positives, and deliver timely alerts with relevant context.

Evaluation type
task based
Challenge
Real-time Crypto Fraud Alert Agent with Claude Agents SDK
Difficulty
Advanced
Rigor
Unspecified

Evaluation overview

How the linked challenge is judged: tasks, benchmarks, and criteria count.

Tasks
2
Benchmarks
0
Criteria
0

Task templates

Inputs and expected outputs.

Task 1

FraudDetectionAccuracy

Given a stream of simulated crypto transactions, the agent must identify known fraud patterns (e.g., large conversions to Monero after a social engineering scam event) and trigger an alert.

Input format

{'transaction_stream': [{'tx_id': '...', 'from_addr': '...', 'to_addr': '...', 'amount': '...', 'currency': '...', 'timestamp': '...'}, ...], 'news_feed': [{'headline': '...', 'content': '...', 'timestamp': '...'}, ...]}

Output format

{'alerts': [{'type': 'fraud', 'detected_pattern': '...', 'suspicious_txs': ['tx_id_1', ...], 'context': '...', 'timestamp': '...'}]}

Task 2

FalsePositiveRate

Ensure the agent does not trigger alerts for normal, legitimate high-volume transactions.

Input format

{'transaction_stream': [...], 'news_feed': [...]}

Output format

{'alerts': []}