Performance

These are absolute baseline measurements of common consumer journeys, not an optimization comparison. Lower scores are better.

The accepted run measured ReṼoman 0.90.0 at revision 7b1603db on an Intel Core i7-9800X with Linux 6.8.0-138-generic and Amazon OpenJDK 25.0.4.1. JMH used average time in milliseconds per operation, one thread pinned to logical CPUs 0 through 7, five forks, ten one-second warmup iterations, and twenty one-second measurement iterations per fork. Each row therefore contains 100 samples and reports a 99.9-percent error interval.

Each execution journey uses deterministic in-process responses supplied through Kick.httpClient(HttpHandler). The first six rows time one complete public ReVoman.revUp call. The rendering row times only Rundown.toJson(Verbosity.VERBOSE) on a prepared successful rundown. Fixture construction and other setup are outside the measured bodies.

The figures include only in-process ReṼoman engine work. They exclude DNS resolution, sockets, TLS, remote-service latency, cold JVM startup, and caller work outside the measured public call.

The complete manifest, environment snapshot, raw JMH CSV, and CPU, allocation, and lock profiles are available in the accepted evidence directory.