we use

minimize_latency

as our consistency guarantee, which means that we could theoretically have results that are as stale as our 5s configured quantization window, but it means we don't have to worry about that complexity