bayes for days

perf(candidates): sample the volley timings instead of taking every one master

Ten clock reads an exchange, against a measured 2.4 million exchanges in one unit's decision: the instrumentation was 8% of what it was measuring. One call in 64 is timed now and the reading scaled back up, on a fixed stride so nothing about it can reach a result.


+35 -9
3 changed files