Validation and review

Implementation baseline: d166be532720abed66f1d9816b148207ccaa6064. Review scope: new CHD Quarto document, repo-local offline modules/tests and its own rendered outputs. Existing website changes and other posts are excluded.

Behavioral and integration checks

  • All five test files pass via Rscript offline/run-tests.R, including fresh local input reconstruction, blocked scientific remote calls and full exports.
  • All 14 R source/test/runner files parse successfully. R has no static typechecking configuration in this repository.
  • The compact real GBC regression fixture matches pinned-reference estimates. An additional development check compared the full original local GBC bundle against the exact reference commit: all 33 pair scientific fields match.
  • CHD retains 41 leads, 18 target SNPs, 23 pair rows, five pooled estimates and 22 usable comparisons/decompositions. All 23 total associations remain usable. SBP/rs10139550 remains unavailable for indirect/residual inference.
  • Outcome extraction coverage is 40/41 leads, 984/992 distinct instrument SNPs and 18/18 target SNPs. Expanded VCF duplicate records are retained for checks.
  • CSV round trips preserve row counts. Geographic/covariance sensitivity CSVs have 46/69 rows, respectively, and 12 separate inference families are exported.
  • A fresh report-only Quarto render succeeds with scientific remote access blocked. All 10 PNGs are embedded; native MathML needs no external scripts. All 35 local download links resolve.
  • All exported pages were visually inspected: comparison PDF 5 pages, decomposition PDF 4 pages and pooled PDF 1 page. Dense LDL/SBP labels and unavailable components remain readable.

Standards

Independent standards review found no actionable documented-standard violations or material maintainability findings. Input reconstruction, statistical core, inference, plotting and orchestration follow the documented offline workflow.

Spec

Independent specification review identified one provenance defect: queried counts were derived from successful outcome matches. It was corrected to use independent inventory denominators, with regression assertions for 40/41 and 984/992 coverage. The report now distinguishes returned association records from requested instrument identities. No other missing requirements, scientific implementation errors or scope creep were identified. The user’s new-QMD instruction supersedes the original issue’s notebook-edit delivery requirement.

Final findings: Standards 0; Spec 1 corrected, 0 unresolved.