version: "1.0.0" name: radiology-writing description: "Draft/restructure manuscripts or theses from fixed evidence; not polishing, review or rebuttal. CN: 论文写作、英文稿、结构化摘要"

Venue-routed imaging and mechanism manuscript writing

Use this skill to construct imaging or mechanism-research prose — argument first, then sentences — in the selected venue's real shape. _Radiology_ is the fallback only for an imaging-centered manuscript whose venue is not yet selected; it is not the default shape for a standalone bulk RNA, single-cell, spatial or perturbation paper. For sentence-level polish use radiology-polishing.

Core stance

• Author evidence first. Never invent results, metrics, p-values, cohort numbers,

citations, mechanisms, or limitations. Missing input → explicit placeholder or a question.

• Write the argument before the sentences. One-sentence claim → section architecture →

paragraph jobs → prose.

• Match the selected venue family. Use _Radiology_ Summary/Key Results only on the RSNA route;

Nature/npj, Cell Press, JAMA Network, Lancet and Wiley routes are not interchangeable.

• Preserve the scientific packet. When a validated radiogenomics handoff exists, stable Claim

IDs, modality roles, independent unit, matched n, evidence state, claim branch/verdict/ceiling and protected placement are immutable. Return conflicts to the scientific ledger.

• Preserve the evidence-synthesis packet. For evidence-synthesis, consume the frozen protocol,

search, selection-flow, extraction, risk-of-bias/applicability, synthesis and certainty artifacts from radiology-systematic-review. Preserve the study-family/report/effect-row hierarchy and the synthesized claim ceiling; this route does not require a radiogenomics handoff.

• Bound the claim. Ambitious but evidence-bounded; calibrate verbs to evidence

(demonstrate → suggest → may reflect).

• Reporting-aware. Every Methods/Results element should satisfy the relevant checklist

item (CLAIM/TRIPOD+AI/CLEAR/STARD) — cross-check with radiology-reporting.

When to use

• Draft/rebuild any section: title, abstract, Introduction, Methods, Results or Discussion, plus

venue-specific front matter such as Summary statement/Key Results only when the selected route requires it.

• Turn Chinese lab notes / mixed drafts into submission-ready English.
• Chinese→English translation/reconstruction of manuscript text (language translation —

not to be confused with radiology-translation, which is clinical/bench-to-bedside translation). Translate the argument and intent, not clause order; open references/chinese-author-workflow.md first.

• Restructure a rejected draft to the _Radiology_ argument shape.
• Build or audit a monograph thesis, article-based/cumulative thesis, chapter package,

thesis-to-paper conversion or whole-thesis argument spine from fixed evidence. Institutional regulations and examiner decisions remain external human authority.

• 中文高频说法：“帮我把中文实验记录/组会思路写成英文投稿稿。”“按 Radiology 风格重构摘要和讨论。”

When to open extra files

Intake (identify before drafting)

• Target venue/shape: resolve RSNA, Nature/npj, Cell Press, JAMA Network, Advanced/Wiley,

European, NEJM, Science or Lancet before finalising front matter. For Cancer Cell, Cell Reports Medicine, JAMA Network Open or Advanced Science, open cell-jama-wiley-shape.md. If venue is undecided, use a reversible journal-neutral scientific spine; use the _Radiology_ fallback only for an imaging-centered paper.

• Scientific state: if a mechanism or radiogenomics packet is supplied, record its path,

canonical packet_sha256, distinct physical file SHA-256 when transported, source-manifest digest and writing_handoff_status; validate it before consuming W-HANDOFF-READY. If no packet exists, build a local map of scope, modality roles, independent unit, usable n, material Claim IDs, evidence locations, evidence states and wording ceilings from the supplied artifacts. Mark unresolved fields AUTHOR_INPUT_NEEDED; do not block traceable sections merely because another module was not invoked first. An untraceable Results or mechanism claim receives a visible scaffold or W-RETURN-TO-LEDGER, never invented science.

• Evidence-synthesis state: for study_scope=evidence-synthesis, record the review route,

lifecycle, protocol/search/selection-flow/extraction/risk-of-bias-applicability/synthesis/certainty artifact IDs, versions and SHA-256s, study-family hierarchy, pooling decision, certainty and supported/forbidden wording. If a decision-bearing artifact is missing or not fixed, write only a visible Methods/Results scaffold and return the affected item to radiology-systematic-review.

• Section(s) requested.
• Study type: diagnostic-accuracy, prediction model, radiomics, radiogenomics, reader

study, observational, trial, scoping review, systematic narrative synthesis, or meta-analysis.

• Core claim: what the study actually shows.
• Evidence: cohorts (n, source, dates), metrics with CIs, comparisons, validation.
• Method-evaluation state: Evaluation IDs, parameter/configuration IDs, benchmark/ablation

artifacts, selection/freeze points, evidence state and claim consequence when the paper makes a methodological or robustness claim.

• Boundary: where the claim stops (single-centre? retrospective? prevalence?).
• Limits: target word/figure counts (verify against current author instructions).
• Thesis state when applicable: degree/programme, monograph versus article-based form,

institutional rule source/date, included-paper status and reuse rights, chapter/version map, cross-chapter claim IDs, contribution statement, examination/defense milestone and embargo route.

If core claim, evidence, or boundary is missing, surface the gap and offer a scaffold with placeholders rather than inventing content.

Writing workflow

1. For full manuscripts or high-impact rebuilds, open

argument-spine-and-stage-gates.md and establish the project context, contribution-first gate, and results-as-validation map before drafting long prose.

1. For radiogenomics or mechanism work, validate an existing packet and establish a Claim-ID

drift ledger. Without a packet, create the bounded local scientific-state map described above and continue wherever evidence is traceable; escalate only affected scientific uncertainties to radiology-radiogenomics.

1. For evidence synthesis, select study_scope=evidence-synthesis, consume the fixed

radiology-systematic-review: writing-handoff, and verify all seven review-artifact roles plus the study-family/report/effect-row map before drafting Results or certainty language. Planned, author-reported or unresolved pooling output stays out of Results. Route a changed eligibility, extraction, bias, synthesis or certainty decision back to the review owner rather than creating a radiogenomics packet.

1. For target-venue writing taste, open journal-family-writing-style.md after confirming

the venue family; use it to adjust article shape, title/abstract rhythm, key points, and clinical relevance language.

1. For parameter or methodology evidence, consume a completed

radiology-method-evaluation matrix when available. If the evaluation is missing, separate Methods reporting facts from the unproven Results claim and return only the affected question to that module.

1. For thesis/dissertation work, open thesis-and-dissertation-architecture.md; freeze the

institution/programme rule passport and choose monograph, article-based, chapter-draft, thesis-to-paper or whole-thesis-audit. Keep a thesis-wide contribution spine, stable Claim IDs, included-paper/version/reuse-rights map and cross-chapter non-duplication ledger. Do not convert a defense outcome, examiner judgment or repository/embargo decision into a writing-model verdict.

1. One-sentence argument: *"In [population/modality], we show [advance] using [approach],

supported by [key result with CI], with [boundary]."*

1. Pick the architecture conditionally — RSNA/_Radiology_ route →

article-architecture.md and, when applicable, structured-abstract.md; Nature/npj route → nature-family-shape.md; Cancer Cell/Cell Reports Medicine/JAMA Network/Advanced-Wiley route → cell-jama-wiley-shape.md. For other verified families, use the applicable current guide/style reference. If the family is unknown or its exact shape is unverified, retain a reversible journal-neutral scientific spine and mark VENUE_SHAPE_UNVERIFIED.

1. Map each paragraph to one job: context / gap / objective / design / cohort / technique

/ analysis / result / comparison / validation / interpretation / limitation.

1. For full sections, define the section contract (purpose, inputs, allowed/forbidden claims,

evidence and validation) and, when author samples exist, the style profile from section-contract-and-style-profile.md.

1. For full sections, draft the topic-sentence chain first, carrying the stable Claim ID beside

each material sentence. If the claims do not flow, revise the chain before writing full paragraphs.

1. Draft from evidence outward — keep claims next to the numbers that support them.
2. Calibrate verbs to evidence; remove unsupported novelty/"first" claims.
3. Fit only the selected venue-family shape — require abstract headings, Summary statement, Key

Results and the _Radiology_ Discussion pattern only on the RSNA/_Radiology_ route. Do not import those elements into Nature/npj, Cell Press, JAMA Network, Lancet or Wiley routes unless the current target-journal instructions independently require them.

1. Self-review against the section contract and relevant reporting checklist; flag unmet

items. For full manuscripts, use a fresh-reader/adversarial pass before final prose ships.

RSNA / _Radiology_ route defaults

Apply this section only after the RSNA/_Radiology_ route is selected. Other routes use their own shape reference; an undecided venue keeps the journal-neutral scientific spine.

• Title — concrete: population/condition + modality/method + finding/role. State AI/

radiomics if central. Avoid slogans and "novel."

• Structured abstract — Background → Purpose → Materials and Methods → Results →

Conclusion; report design, cohort sizes/dates, primary metric(s) with CIs, a bounded conclusion. (structured-abstract.md)

• Summary statement — a single declarative sentence of the main finding.
• Key Results — up to 3 results/conclusions, ≤ 75 words, with summary data; don't

repeat the Summary statement; avoid vague language and abbreviations.

• Introduction — field/clinical stakes → specific gap → objective/hypothesis. Short; no

results dump.

• Materials and Methods — design + ethics/registration → participants/flow → imaging

technique → image analysis/reference standard/readers → model/feature pipeline → statistical analysis. (methods.md)

• Results — patient flow + characteristics → primary performance with CIs → comparisons →

validation/subgroups. Past tense, quantitative. (results.md)

• Discussion — first paragraph = concise summary of key findings, then relation to

prior work, then limitations, then a bounded conclusion. (discussion.md)

Output format

1. `Draft` — the requested prose in the selected venue-family shape, or the reversible

journal-neutral scientific spine when the venue is unresolved.

1. `Section outline` — 3–7 compact bullets (for a full section).
2. `Topic-sentence chain` — for full sections or major rewrites, the claim sequence before

paragraph expansion.

1. `Section contract` — purpose, inputs, allowed/forbidden claims, validation, and word/display

budget for full sections.

1. `Claim–evidence map` — `Claim ID | Claim | Evidence (with CI) | immutable ceiling |

Status: supported / needs input`.

1. `Stage gates` — for full manuscripts: contribution gate, results-as-validation gate,

citation/figure/reporting gates that are passed or still open.

1. `Venue style check` — if the target family is known: abstract shape, title/key-points

logic, clinical relevance language, and any guide-derived limits or VERIFY FROM GUIDE.

1. `Style calibration` — journal/discipline rules applied first; author-profile traits used

only where compatible and based on actual samples.

1. `Assumptions / missing inputs` — only material gaps.
2. `Reporting check` — checklist items this draft does/doesn't satisfy (→ radiology-reporting).
3. `Claim-ID drift ledger` — for a scientific handoff: immutable fields preserved, editorial

changes, any return-to-ledger request, and the packet digest consumed.

1. `Method-evaluation placement map` — when applicable: Evaluation/Parameter IDs and exact

Methods, Results, Figure/Table, Supplement/config/code and Discussion locations, preserving each evidence state and claim consequence.

1. `Evidence-synthesis placement map` — for review manuscripts: stable Claim ID, study-family

unit, protocol/search/flow/extraction/RoB/synthesis/certainty source anchor, evidence state, supported wording, forbidden stronger wording, and exact Methods/Results/display/Supplement/ Discussion placement.

1. `Thesis architecture packet` — when applicable: thesis mode and rule passport, contribution

spine, chapter jobs/dependencies, included-paper/version/reuse-rights map, cross-chapter claim/display/method overlap ledger, synthesis/conclusion ceiling and defense/repository handoff.

For Chinese notes: polished English first, then brief Chinese notes on structural choices.

Handoffs

• Sentence-level polish / house style → radiology-polishing.
• Statistics/CIs/tests behind the numbers → radiology-stats.
• Parameter/metric/method fit, fair benchmark, ablation and sensitivity evidence design or audit →

radiology-method-evaluation; this writing skill consumes fixed evidence but does not certify an unrun evaluation.

• Frozen Methods/Results handoffs from execution skills (radiology-acquisition-qc,

radiology-annotation, radiology-radiomics, radiology-deep-learning, radiology-transcriptomics-analysis writing-handoff outputs) are consumed as fixed evidence: integrate and reconcile them, never silently upgrade their claim ceilings.

• Checklist compliance → radiology-reporting.
• Figures/legends → radiology-figure.
• Publication tables and table footnotes → radiology-table.
• Finding/verifying citations to support a claim in Introduction/Discussion → radiology-citation.
• Draft scientific review is scope-aware: imaging-only → radiology-prereview;

mechanism-only → radiology-radiogenomics: manuscript-review; imaging-mechanism → mechanism/ bridge review in radiology-radiogenomics first, then imaging/editor synthesis in radiology-prereview; evidence-synthesis → radiology-systematic-review: audit/writing-handoff, then radiology-prereview: evidence-synthesis-review with scientific_handoff_digest=not-applicable. Do not treat any scientific review as all-files submission readiness.

• Full-project artifact state → radiology-pipeline; upload-ready package (cover letter

included — template at radiology-submission/assets/cover-letter-brief.template.md) → radiology-submission.

• Thesis defense deck and rehearsal artifact → radiology-paper2ppt; examiner or institutional

approval stays with the graduate programme; thesis repository/access/embargo operations route to radiology-data / radiology-research-ops as applicable.