For simplicity, each WindowedAnalyzer instance tracks it's own history of values.

At the cost of extra CPU cycles to re-calculate latency speed (and maybe latency acceleration), we could cut the amount of history stored by 1/2 or 1/3.