Measure narrowly. Publish the limitations with the number.

RuleOak benchmarks are useful for implementation regression and local characterization. They are not independent audits or universal production promises.

npm run benchmark
npm run benchmark:policy
npm run benchmark:mcp
npm run benchmark:evidence

Measured areas

• policy decision latency and scaling;
• receipt issuance/verification and lease replay rejection;
• evidence append latency;
• stdio JSON-RPC and loopback HTTP throughput;
• malformed-fragment handling;
• memory change during sustained evidence append.

Record the environment

Node.js version, OS, architecture, CPU, memory, storage/filesystem, sample count, benchmark options and durability settings such as fsync materially affect interpretation.

Use percentiles carefully

p50 describes the middle observed result; p95/p99 describe the slower tail in that run. They do not automatically predict a production deployment with different policies, networks, providers or disks.