When a WAF detection path fails, the dangerous default is a silent fail-open that still serves traffic while you think you are protected. Prefer an explicit per-route response: fail-closed on login, checkout, and write APIs, and a monitored fail-open or challenge only on read-heavy public paths where a hard block would take the whole site dark. Keep health probes on a separate allow path so a detection outage cannot mark every upstream unhealthy.
Ops prove-out after pinning fail behaviour: force a detection error on a canary Host and confirm the protected write path returns the block or challenge you configured, while the public read path still serves under the softer policy. Watch blocked, challenged, and error counters by route for one deploy window — a spike in detection errors with flat blocks usually means you are still fail-open where you expected closed. Roll under peak and compare origin request rate; if origin climbs while the WAF reports healthy, requests are skipping the rule path entirely.