Time handling is easy to underestimate in a distributed limiter. If application servers calculate refills using their own clocks, clock skew can produce inconsistent decisions. Keeping time calculation close to the authoritative Redis state makes the behavior easier to reason about.