profiterole
profiterole copied to clipboard
Consider performance thresholds
How stable are profile results? If we could diff profile outputs and see what got more expensive, would that help spot performance regressions?