Cloudflare

What they grade on

• Edge-first thinking — assume the answer runs in every PoP
• DDoS + abuse-mitigation reasoning from the first minute
• Zero-trust and TLS termination fluency
• Cost per request at global edge scale
• Latency-obsessed — every millisecond matters

Deal breakers

• Origin-centric design — treating the CDN as an afterthought
• No abuse / DDoS story on a public-facing service
• Ignoring cost per request at 100B+ RPS scale
• No thought for BGP or Anycast routing when relevant

Super helpful signals

• Discussing Workers or Durable Objects when relevant — 'compute at the edge' isn't a bolt-on
• Naming Anycast and its trade-offs (TCP session affinity, PoP failover)
• Referencing the Cloudflare Global Network (300+ POPs) with realistic numbers
• Talking about zero-trust access model (mTLS, JWT at edge)

Delivery advice

• Assume Anycast + every PoP has the code — don't design around a central origin.
• Consider abuse and DDoS from the first minute.
• Talk about cost per request at global scale — Cloudflare loves this.
• Reason about latency budgets under 100ms for user-facing operations.
• Discuss BGP path optimization + peering when relevant to network design.

Red flags to avoid

• Origin-centric design
• No abuse story
• Ignoring latency budgets or cost per request
• Assuming a central control plane can survive PoP-level partitions