Public edition · 2026-09-25
Mendelian Randomization: validation
These results describe the checks actually performed on this edition. They do not certify scientific conclusions, grant outcomes or submission readiness.
Linux 6.8.0-138-generic x86_64 · Python 3.10.12 · R 4.3.2 · TwoSampleMR 0.6.11 · Codex CLI 0.154.0
Checks and scope
Frozen numeric comparison
passed19/19 identity and headline cells matched declared exact, source-display or frozen stochastic rules. Independent IVW/Egger/Wald/leave-one-out arithmetic passed.
Historical package branch
limitedTwoSampleMR 0.4.22 matched the current fixed-seed four-method output under R 4.3.2. The original R 3.5 system and unknown seed were not recovered.
Regression and transfer
passed83 tests passed. Nine case-1 pairs and one case-3 pair matched their scoped comparisons; these are same-archive transfer checks, not independent biological validation.
Historical proof figure generation
passedThe preserved public.2 evidence records six display types, 48 assets and 18 PDF pages passing deterministic replay and visual review; browser previews passed at 320, 390 and 768 CSS pixels. The 1.0.2 documentation update did not regenerate or requalify those scientific proof assets.
Clean package replay
passedExact 1.0.2 ZIP/source identity, safe extraction, 83 tests, invented R estimate and Python rendering passed. Qualified runtime bytes match 1.0.1 and public.2.
Native host
limitedCodex CLI 0.154.0 on Linux discovered the runtime-equivalent public.2 skill and ran the invented example. The 1.0.2 ZIP was clean-replayed but not separately host-registered; other hosts and operating systems were not tested.
Independent human scientific review
unavailableOwner-authorized, AI-assisted preparation only; no independent human scientific review was performed.
Package delivery and commerce
unavailablePreview only. The $5.00 one-off display price does not enable checkout, entitlement, customer package delivery or external marketplace submission.
Walker v2 Table 4/archive comparison
| Method | Published b | Reproduced b | Published SE | Reproduced SE | Published P | Reproduced P | Status |
|---|---|---|---|---|---|---|---|
| Inverse variance weighted | 0.5663 | 0.5662910 | 0.0905 | 0.0905047 | 3.924e-10 | 3.9238e-10 | Matched |
| MR Egger | 0.9711 | 0.9710530 | 0.2917 | 0.2917428 | 0.001091 | 0.00109103 | Matched |
| Weighted median | 0.571 | 0.5710485 | 0.07664 | 0.0755234 | 9.226e-14 | 3.9932e-14 | Matched* |
| Weighted mode | 0.571 | 0.5709661 | 0.1744 | 0.1811799 | 0.00131 | 0.00194873 | Matched* |
Verified figures
Iterations and unchanged source limits
Four reusable defects were fixed with preserved before/after fixtures: missing method lines and Egger intercept in generated figures, an incorrect all-sizes funnel warning, one-row T allele typing, and all-missing optional EAF typing. The installed/candidate delta is confined to run_mr.R, render_mr.py and their portable regression tests.
The weighted-median prose/table mismatch, missing embedded exposure unit, unresolved authentic-GWAS redistribution, unknown original seed and strong heterogeneity are source or interpretation limits; they were not edited away.
Rights and review roles
Original Nexitia-controlled package files and the invented example use Nexitia Skill License 1.1. Separately installed open-source dependencies retain their own licenses and are not bundled. Public Nexitia-generated comparison data, reports and figures remain CC BY 4.0 with Walker attribution and change disclosure.
The earlier GPL-labelled 1.0.0-public.2 edition and any rights already received under it remain preserved. Preparation and technical review were AI-assisted under the owner’s individual authorization; institutional ownership and legal enforceability were not reviewed by counsel.
Reproduction commands
Package regression suite
PYTHONDONTWRITEBYTECODE=1 python3 -m pytest -q testsInvented three-variant estimation
Rscript scripts/run_mr.R --harmonized examples/synthetic-harmonized.csv --effect-scale continuous --out synthetic-run --seed 153 --nboot 100Invented example rendering
python3 scripts/render_mr.py --run synthetic-run --out synthetic-figures --scale betaRemaining limits
- The flagship is a technical replay of Walker et al. 2019 v2 case 2, not a reproduction of the whole paper or an independent causal discovery.
- The exposure unit is catalogue-reported SD; the frozen archive does not embed that unit.
- A nonsignificant Egger-intercept test does not establish absence of directional pleiotropy, and high heterogeneity limits simple causal interpretation.
- Authentic benchmark associations are excluded from the customer/public package; users must acquire lawful inputs separately.
- Advanced correlated, multivariable, colocalization, one-sample, clinical and live-retrieval workflows are unavailable or uncertified.
Owner-authorized and AI-assisted. Edition 1.0.2 adds OpenGWAS access guidance without changing the qualified runtime files or the 1.0.1 license correction. Package, numerical and visual checks are recorded; no independent human scientific review, legal-counsel review or clinical certification is claimed.
Reproduction and package identity
Read the execution report and reproduction commandsmendelian-randomization-1.0.2.zip · 29 files · 207,191 bytes unpacked
b0401d5fae3c6263d8c8fad71d6b5fa5b45af7d9ecc43e45a970823a47ae24fd