# Publication Surface Scrub Internal IDs, user-interface choices and audit artifacts describe how the paper was produced. They do not belong in the research article. Apply this boundary when drafting, assembling another layout and reviewing a revision, including the title area, abstract, running headers, captions, tables, declarations and scientific supplements. Do not copy task status or a continuation note into a template's author note, draft-status block or abstract. Do not generate standalone "Draft status", "Review status", "Delivery status" or equivalent explanatory paragraphs in the article, even when the underlying status is true. Keep that communication in the delivery notes. Removing the heading while leaving its production narration in ordinary prose does not fix it. Use saved choices to select the actual figures, not to narrate that the user confirmed them. Keep figure adoption, pending independent review, visual-check completion, package readiness and download status in task notes or the local work-package README. A new layout must not inherit an old production-status paragraph merely because it preserves the preceding manuscript's text. Remove bracketed claim/evidence/source/numeric/method/outcome/result tags, workflow filenames and gate narration from reader-facing prose. Repair the source and regenerate only the outputs authorized by the current request; an audit-only request does not authorize replacing the user's paper. Preserve the scientific boundary in natural prose. Do not delete a real limitation, author-input requirement, or disclosed uncertainty merely because it originated in an audit. Separate a mixed paragraph by its meaning. For example, move "the user confirmed these figures" to work notes, while retaining an unresolved measurement unit or study-setting limitation where it affects interpretation. Do not invent author, ethics or recruitment facts. Use concise academic wording or an explicit author placeholder when needed; a local draft is not automatically submission-ready. Reproducibility methods, participant choices in an HCI study and a systematic review's screening process remain scientific content. Do not remove them merely because they mention a user, workflow or review. PaperSpine itself must never add an AI-use disclosure, model/tool name, AI-generated label or workflow provenance marker to any reader-facing paper output. If an external submission system separately requires a disclosure, keep that requirement in submission metadata and author instructions; do not silently add it to the manuscript body. Also remove defensive audit voice. The final paper should not repeatedly say “within the supplied materials,” “user-authoritative,” “the evidence ledger shows,” or stack `may/might/potentially` around an observed result. Follow `references/assertive-scientific-writing.md`: state supported findings directly and keep one clear qualification at the point where the inference changes. ```bash python scripts/publication_surface_check.py --source "" --source "" --json ``` Pass the actual manuscript paths, including requested DOCX/Markdown outputs; do not rely on the historical output-directory default. The checker reports located candidates and extraction failures; it does not rewrite the paper or prove that every paraphrase is absent. Read the full current article and inspect all rendered pages independently, including text inside figure images that a text extractor cannot see. Resolve each finding by meaning, preserve legitimate scientific content, and recheck the affected source and output. Keep work notes and review reports out of this manuscript-only scan.