Design a malicious-URL checking service using an isMalicious API

Design a backend service that determines whether a given URL is malicious.

You have access to an external dependency:

• isMalicious(url) -> bool (black-box API)

This external API is:

• Rate-limited (e.g., 1k QPS max) and sometimes slow (p95 500ms) or unavailable.
• Potentially costly per call.

Your service should expose an internal API such as:

• GET /reputation?url=... → returns { verdict: "malicious"|"benign"|"unknown", checkedAt, confidence }

Requirements

1. Low latency for repeated queries.
2. Must handle high traffic (e.g., 50k QPS reads) while respecting the external API limit.
3. Avoid repeatedly calling isMalicious for the same URL.
4. Provide reasonable behavior for timeouts/outages (no cascading failures).
5. Support background re-checking because reputations can change.

Explain:

• APIs and key workflows (read path, cache miss path, async refresh)
• Data model and storage choices
• Caching strategy, TTLs, and deduplication of concurrent requests
• Rate limiting / backpressure / circuit breaker strategy
• Scaling, partitioning, and reliability
• Metrics and operational concerns