Skills 2026-10-05 14:57 UTC
medsci-skillsv6.1.0
Version [v6.1.0] committed by Yoojin-nam. Update Category: Feature Addition & Enhancement. Commit details: docs(write-paper, revise, design-study): guidance drawn from co-author edits (#685). This update introduces new capabilities and module extensions.
CommitRaw commit message
docs(write-paper, revise, design-study): guidance drawn from co-author edits (#685) * docs(write-paper, revise, design-study): guidance drawn from co-author edits Four patterns that senior co-authors repeatedly corrected by hand, written as guidance with the case each rule must leave alone: - revise: unify terminology with a one-page term table before editing sentences (terms that name different things stay distinct). - revise: when a response letter is compressed or converted after drafting, diff its evidence item by item, not its word count; cut material moves to the Supplement and the response says so. - revise, write-paper: words left under a journal cap are not a target, and a word count or cap does not belong in the manuscript text unless the journal asks for it there. - write-paper: a general-purpose method validated on one task leads with the method and puts the disease in a validation clause; the four title components still apply, and a disease-first title stays when the disease is the contribution. - design-study: fix the reader-study analysis model at design, after the outcome, link, sampling, estimand and weights are defined; a crossed reader/case mixed model for a conditional estimand, a GEE when the population-average quantity is the estimand. Documentation only: no detector, verdict or script changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEWfQeHwpb5Dap2bBuXC6 * docs(write-paper, revise, design-study): tighten three of the new rules - design-study: a population-average primary analysis may be a GEE or a marginalized mixed model, and either way the variance must handle reader and case dependence; clustering on reader alone ignores shared cases. - write-paper: the abstract states the aim actually performed, so a paper that only validates an existing method does not claim development. - revise: after compressing a response letter, evidence that no longer fits goes into the manuscript or Supplement with an exact pointer; redundant explanation can simply go. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEWfQeHwpb5Dap2bBuXC6 * docs(revise): keep reviewer-only evidence in a compressed letter Only evidence the paper itself needs moves into the manuscript or Supplement when a response letter is compressed; evidence that only answers the reviewer stays in the letter in condensed form. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEWfQeHwpb5Dap2bBuXC6 * docs(design-study): conditional and marginal effects differ only under a non-linear link A GEE used as a robustness check reports a different quantity from the conditional mixed model only when the link is non-linear (e.g. logit); with an identity link the two coincide. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEWfQeHwpb5Dap2bBuXC6 * docs(design-study): state which quantity a GEE estimates instead of assuming agreement Whether a GEE matches the conditional effect depends on both the link and the random-effects structure, so the guidance no longer states when they coincide; a GEE robustness check states which quantity it estimates. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AXEWfQeHwpb5Dap2bBuXC6 --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>