Performance

The native benchmark suite compares solvers across graph families. Its measurement contract sets common inputs, original-system residual checks, setup and solve timing boundaries, and reproducibility requirements.

The batching guide contains a dated CPU/Metal case study, with timings, settings and provenance. It compares batched and scalar calls reusing each route’s factor on the YouTube graph.

CI regression artifacts compare baseline and candidate revisions on the same runner. Read their numerical checks and paired controls alongside timing changes; the measurement contract explains the accuracy and timing requirements.