Open
Description
The coercions benchmark has been consistently quite noisy (over 1% regressions/improvements) for the last several months, at least, and may have recently gotten worse. The debug profile seems noisiest. We should aim to benchmark under cachegrind (notably, using production artifacts) to attempt to understand where the noise originates.