06 / reproducibility

Benchmarks

Benchmark reachq as an algorithmic library: record the work, the query, the graph, the environment, and the invariant together.

Edit this pageUpdated with the source

Record a complete run

Commit, package version, Python version, operating system, and CPU.
Graph source and type, vertex count n, edge count m, density, and seed.
Algorithm, epsilon, omega, beta, hopbound, and refinement configuration.
Construction time, query time, memory, shortcut/hopset size, and approximation error.

Separate claim types

Theoretical work/depth bounds, microbenchmarks, and end-to-end measurements answer different questions. Keep them separate and never use beta alone as a proxy for practical performance.

Reproducible benchmark scripts live in the repository scripts.