Certificate verification bugs in the runtime bite harder than cipher misconfig: a check that returns success while skipping hostname or chain validation leaves every outbound dial looking healthy until a name mismatch or intercepted peer shows up in production. Pin the verify mode your clients actually need, fail closed when the leaf name does not match the dialed SNI, and keep the trust store path under the same config review as the renew job so a library upgrade cannot silently weaken checks.
Ops prove-out after a runtime or OpenSSL bump: dial a host with a mismatched leaf and confirm the client fails before any application payload; then dial a correct name and confirm session reuse still works under load. Alert on verify-error counters separately from connect timeouts so a broken check cannot look like network blips. Most sudden open-trust incidents are a default that stopped verifying, not a cipher change.