Review 21 Personal QC

Purpose

Run the user's final "reviewer-before-reviewer" audit. This skill is stricter than ordinary polishing: it checks whether the manuscript can survive peer review on evidence, statistics, framing, reporting, and submission packaging.

Audit Order

1. Identify manuscript type: narrative review, systematic review, meta-analysis, translational cohort, prediction model, or mixed evidence article.
2. Read the manuscript, evidence matrix, citation audit, journal target report, and any numeric/statistical verification files.
3. Build a risk ledger with critical, major, moderate, and minor issues.
4. Separate four domains: evidence truth, numeric consistency, conceptual framing, and journal readiness.
5. Recommend concrete edits, not generic advice.

Personal Gates Distilled From Prior Work

• Meta-analysis gate: verify PMID/DOI, event counts, effect measure, heterogeneity, sensitivity analyses, absolute-risk translation, and PRISMA limitations.
• Translational manuscript gate: do not claim integration when molecular data and clinical cohort are parallel rather than patient-matched.
• Prediction-model gate: distinguish exploratory, internally validated, externally validated, and clinically deployable tools.
• Observational cohort gate: downgrade causal wording to association unless design supports causal inference.
• Outcome gate: rename study-defined outcomes precisely and justify thresholds.
• Figure gate: every panel needs source evidence, panel order, legend alignment, and permission/redraw status.
• Submission gate: verify declarations, ethics, funding, data availability, author contributions, target journal scope, and upload order.

Output

Create or update 18_final_qc/personal_qc_report.md with:

• Overall verdict: ready, minor revision, major revision, or not ready.
• Top 5 risks.
• Claim-to-evidence problems.
• Numeric/statistical problems.
• Framing and language changes.
• Missing reporting items.
• Journal-fit and submission-package gaps.
• Next revision checklist.

Stop Rules

Do not invent missing data, references, ethics statements, funding, author contributions, or statistical outputs. Mark gaps explicitly and request user confirmation.