@searls yeah I'm the author of a latency instrum lib. one of the reasons I made it was cuz didnt like relying on the metrics provided by an external/hands-off profiler. because just knowing "how long a function takes" is too coarse/noisy. latency depends on precise code path followed, & various context factors. there are variants in each flow, depending on if conditions like u said, switches, errors, exceptions, panics, lock contention, remote service delays etc. all effect metrucs & outcome