GitHub

V2 CRAN Benchmark Manifest

Date: 2026-07-03

This manifest is the runnable benchmark inventory for the V2 CRAN release candidate. It is not a release signoff by itself. A row is release-grade only when it was run from a clean installed package, saved its result artifacts, and passed the strict gate for the promoted solver family. V2 CRAN is the internal release-boundary name used by the tracker and audit files; the package version for this release candidate is eigencore 1.0.0.

The companion documentation scope audit is docs/v1-doc-scope-audit.md. That audit maps user-facing docs, migration docs, design notes, and release evidence to V1 requirements; it does not relax the benchmark stop rule below.

The companion completion audit is docs/v1-completion-audit.md. It is the single checklist to consult before any V1 completion claim.

The contribution-facing interpretation is docs/contribution-methods-artifact.md. It summarizes the current benchmark wins, losses, and non-promoted solver boundaries without changing the runnable commands below.

V3 promotion gates live separately in docs/post-v1-benchmark-gates.md and inst/benchmarks/post-v1-gate-manifest.R. Those gates are for broadening solver claims after the V2 CRAN release; they do not change the CRAN release evidence below.

The source-tree all-surface quick smoke is:

Rscript inst/benchmarks/run-post-v1-gates.R --tier=smoke --gates=post_v1_svd_hard_surface,post_v1_operator_sidecars,post_v1_randomized_svd_hard_surface,post_v1_generalized_preconditioned_surface,post_v1_shift_invert_boundaries,post_v1_nonsymmetric_matrix_free_surface --load-all

It is a contract/provenance regression check. Speed and memory ratios printed by quick fixtures are diagnostic unless the installed-package strict gate below also supports the promoted release claim.

Use an installed package path for release evidence, for example:

"R CMD INSTALL --library=/tmp/eigencore-bench-lib . R_LIBS=/tmp/eigencore-bench-lib Rscript inst/benchmarks/

Read the original on github.com ↗