Description

This pull request:

• Corrects seven cross-package drift findings in @stdlib/blas/base, surfaced by majority-vote analysis across the 107 non-namespace-aggregator leaf packages in the namespace.

Namespace summary

• Members analyzed: 107 leaf packages (the assert, ndarray, and wasm sub-namespace aggregator packages were excluded since they aggregate other packages rather than implement a routine).
• Features analyzed: file tree, package.json top-level / scripts / directories / keywords shape, README ##/### section list, manifest.json shape, test//benchmark//examples/ filename sets.
• Features with a clear majority (≥75%): every package.json top-level key and every directories sub-key (100%); four universal keywords (stdlib, blas, array, ndarray); the canonical file tree (README.md, package.json, lib/{index,main}.js, test/test.js, examples/index.js, docs/types/{index.d.ts,test.ts}, docs/repl.txt, benchmark/benchmark.js); the canonical ## Usage / ## Notes / ## Examples README sections.
• Features excluded for no clear majority: the optional __stdlib__ config block (60% prevalence, split evenly between {} and {"wasm": false}); gypfile (47%); browser (40%); the ## Notes heading on enum-utility packages (intentionally empty); the ## See Also content (auto-populated).

Per outlier package

blas/base/drotg

README.md opens the <section class="examples"> block but jumps straight to the code fence without the ## Examples heading. The heading is present in 105/107 (98%) of siblings — including the analogous dasum, ccopy, dswap READMEs — and is a structural template element, not a content judgment. Heading restored.

blas/base/srotg

Same drift as drotg: the <section class="examples"> block is missing the ## Examples heading present in 105/107 (98%) of siblings. Same restoration applied.

blas/base/dnrm2

package.json keywords lacked ndarray despite the package shipping a working .ndarray method (lib/ndarray.js, c_dnrm2_ndarray). The keyword is present in 99/107 (93%) of siblings and in every level-1 strided-vector peer (dasum, daxpy, dswap, ...). Keyword added.

blas/base/snrm2

Same omission as dnrm2: .ndarray method ships but ndarray was missing from the keywords. Present in single-precision peers (sasum, saxpy, sswap). Keyword added.

blas/base/gnrm2

Same omission for the generic variant: .ndarray method ships, peer gasum carries ndarray. Keyword added.

blas/base/gswap

Same omission for the generic swap: .ndarray method ships, typed peers dswap and sswap carry ndarray. Keyword added.

blas/base/zcopy

README.md jumped directly from <!-- /.c --> to the links section, with no <section class="related"></section> scaffold. The empty scaffold is the hook the related-links auto-populator keys off, and it is present in 106/107 (99%) of siblings — ccopy, dcopy, scopy all carry it. Only the empty scaffold (plus the surrounding HTML comments) was added; the inner content remains generator-owned.

Related Issues

Does this pull request have any related issues?

None.

Questions

Any questions for reviewers of this pull request?

No.

Other

Validation

• Structural extraction: per-package walk of file tree, package.json shape, README section list, manifest.json shape, and test//benchmark//examples/ filenames; majority computed at the 75% threshold (⌈107 × 0.75⌉ = 81).
• Three-agent drift validation (run in parallel against the 7 candidate corrections):
  • Opus semantic-review — confirm the deviation is not justified by a difference in what the package does.
  • Opus cross-reference — confirm tests/examples do not rely on the deviation and applying the fix does not break a documented contract.
  • Sonnet structural-review — confirm the majority pattern is what should be applied (and not an inapplicable element).
• All seven candidates returned confirmed-drift / safe-to-fix from every reviewer and advanced.

Deliberately excluded from the run:

• The __stdlib__ config block split (60% prevalence, no clear majority shape).
• Missing benchmark/benchmark.js in dgemm/sgemm/ggemm: these packages benchmark across 40 parameter-combination files instead, by design.
• Missing benchmark/benchmark.js and docs/repl.txt in shared/xerbla: these are utility packages (shared include dir, BLAS error handler) with no public computation to benchmark or REPL-document.
• The ## Notes heading on enum-utility packages (diagonal-types, layout-resolve-str, ...): the <section class="notes"> element is present but intentionally empty, which is the convention for the enum family.
• The <!-- <related-links> --> placeholder in the links section (only 25% prevalence — no majority).
• ## See Also content in <section class="related">: auto-populated by the related-links generator, exempt from manual edits.

A local report with the full audit trail (feature distributions, dropped candidates and rationale, namespace random seed) lives at ~/drift-reports/drift-blas-base-2026-06-22.md.

Checklist

• Read, understood, and followed the contributing guidelines.

AI Assistance

• Yes
• No

If you answered "yes" above, how did you use AI assistance?

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

Disclosure

This PR was authored by Claude Code running as a scheduled cross-package drift-detection routine: structural features were extracted across all 107 non-aggregator leaf packages of @stdlib/blas/base, deviations from the ≥75% majority pattern were surfaced, three reviewer agents (semantic, cross-reference, structural) validated each candidate fix in parallel, and the resulting one-line README and package.json corrections were applied per-package. A human will audit and promote the PR out of draft.

@stdlib-js/reviewers

Generated by Claude Code