BENCHMARKS · EVIDENCE

Reproducibility

Provenance

Record the full baseline and candidate revisions, dirty state, OS, architecture, CPU, memory, and Rust, Node, Python, and backend versions.

Resources and fixtures include SHA-256 and logical case counts, not just paths.

Commands

Preserve the official wrappers and every environment override. Morphology Python tests run from tools/morph-compare with discovery that includes both import roots.

Run morphology-query and regex comparisons with scripts/benchmark-search-baseline.sh so resource checksums, fixtures, and the process model remain fixed.

Results from direct Docker or cargo commands that bypass lock and fixture contracts are not approved reports.

Artifact verification

Schema validation checks that JSON reports, Markdown tables, and site snapshots contain the same values. Charts consume only snapshot fields.

A changed source-report revision or SHA-256 requires regenerating the snapshot in the same change.