What is measured
Latency, throughput, memory behavior, and operation-specific timing can be collected for key generation, encapsulation, decapsulation, signing, and verification workflows.
Measurement method
Benchmarks
Keystone's benchmark surface is designed for repeatable research measurement, not headline performance claims. Results depend on algorithm parameters, implementation libraries, hardware, and run context.
What is recorded
Runs need algorithm family, parameter set, library path, platform context, and repeat count so research comparisons can be interpreted instead of treated as absolute.
What is compared
Classical baselines and post-quantum candidates can be reviewed side by side when the execution environment and method are kept visible.
Adapter role
Native boundaries for liboqs and OpenSSL keep cryptographic work outside the visual layer while preserving a clear path back to result metadata.
Validation posture
The useful benchmark is the one another researcher can re-run, inspect, and question. Keystone favors recorded context over marketing shorthand.
Limitations
Different hardware, operating systems, load conditions, and library versions can change outcomes. The prototype does not turn a single run into a universal claim.