When a WAF sits behind another reverse proxy or edge, the source address it logs is often the hop that connected to it, not the client. Trust only an explicit forwarded-client header after you strip or overwrite it at the trusted perimeter; if every hop appends without a trusted rewrite, spoofed client addresses sail into rate limits, allow-lists, and audit trails. Prefer a single rewrite at the outermost trusted proxy that sets the real client once, then ignore further inbound copies of that header from the open internet.
Ops prove-out after the rewrite: send a probe with a forged forwarded-client header and confirm the WAF decision and access log both show the address of the trusted hop (or the rewritten value you control), not the forged one. Compare allow-list and rate-limit counters before and after with the same client path — if blocks suddenly target your load balancer instead of abusers, the trust boundary is still reading the wrong hop. Keep health-check and internal probe paths on a separate allow path so they do not inflate the shared client-IP budget.