Known bugs in v1.0:

CPU benchmarking phase at start is relatively innacurate, requiring
significant compensation by the rolling-usage adjustment.  This calibration is
done while other processes are starting, probably causing capacity
underestimation.

The rolling-usage adjustment takes longer to settle than it ought.

CPU utilization code isn't portable, and may be vulnerable to changes in HZ,
at least as viewed from userspace.

The initial fill in the memory utilization code takes significant time.

