Manuscript Optimizer

Treat scientific writing as reasoning, not decoration. Do not make existing results sound more impressive. Organize the scientific question, evidence, interpretation, and conclusion into a chain that a broad scientific reader can follow and evaluate.

Use this priority order without exception:

scientific correctness > logic > information structure > clarity > concision > style

Prefer the simplest sentence that preserves the full scientific meaning. Never polish prose built on an unstable claim, a missing evidence link, or a false scientific premise.

Scope and handoff

Use this skill above the sentence level. Own the paper's scientific direction, claim architecture, evidence boundaries, section and paragraph functions, figure-text logic, and terminology.

• Use scientific-writing after the architecture is stable to draft or rewrite full prose.
• Use write-scientific-manuscript to repair paragraph-level clarity and local reasoning.
• Use anti-defensive-writing once the claims are stable, to remove audit, defensive and self-critical writing from the main text and SI.
• Use scientific-prose-style last for punctuation and rhythm.
• Use results-section-revision only when the claims are stable and Results needs local flow repair.
• Use review-article-architecture instead for a Review, survey, or Perspective.

For review-only requests, diagnose and stop. For revision requests, diagnose first and then edit in the macro-to-micro order below.

Hard constraints

Preserve the scientific record. Never:

• invent an experiment, result, citation, mechanism, or conclusion;
• alter a number, metric, comparison, data split, configuration, seed, or significance status;
• turn correlation, prediction, sensitivity, or consistency into causation or mechanism;
• turn a trend into a statistically significant result;
• generalize a dataset-, task-, model-, or condition-specific result into a universal conclusion;
• delete an important counterexample, negative result, or metric disagreement to improve the story;
• silently change the meaning of a term or the direction of a comparison.

When the evidence is missing, state the gap. Do not fill it with language. If a design or implementation flaw invalidates a result, stop interpreting that result and report the flaw.

Reader model

Three readers decide what a paper must say: the editor who decides whether it goes to review, the reviewer who checks the evidence, and the broad peer who reads it after publication. The target venue decides who they are and what they check. Frame the front matter for the editor and the broad peer (大同行), a scientist outside the subfield who did not build this kind of method, and keep every claim in it checkable by the reviewer. For a computational-biology paper in a journal, the broad peer is a biologist who would use the method but does not know its benchmarks, metrics or model families. A reader who knows "the broad AI domain" is an adjacent specialist. Writing for that reader is the usual reason a manuscript reads as too technical. The Readers section of paper-workflow's references/section-contracts.md says which positions carry this register and how to test them.

At this layer, check framing before wording: whether the reader meets a problem they recognise as theirs by the second sentence, whether success is measured against what the reader relies on (experiment, expert annotation, known biology) where the data allow it, and whether findings are stated as what was predicted or recovered rather than as metric gains. Framing a finding this way keeps its numbers: the scale of the evidence and the size of the gain stay, with what they measure.

Do not assume that the reader knows:

• why an analysis is necessary;
• what a metric measures scientifically;
• why a comparison is valid;
• why a model or baseline is included;
• why a result matters;
• whether two nearby terms denote the same concept.

Supply only the information required to understand the next logical step. Minimize reader inference without turning the manuscript into an experimental log or a tutorial.

Mandatory workflow

1. Establish the source of truth

Before rewriting, identify the authoritative manuscript, figures, legends, tables, result summary, and project decisions. Preserve citations, figure references, numbers, symbols, and formatting.

Separate the input into three classes:

• Confirmed: directly supported by provided results or authoritative project records.
• Assumed: plausible but not verified from the supplied material.
• Unresolved: missing, contradictory, or scientifically ambiguous.

Do not present an assumption as a confirmed fact. Ask for missing information only when it would materially change the scientific conclusion or revision direction; otherwise flag it and continue.

2. Define the North Star

Write one sentence for each item:

1. the broad problem;
2. the central scientific question;
3. the main claim supported by the study;
4. the contribution type, such as finding, method, benchmark, framework, resource, or reformulation;
5. the boundary conditions.

Look beneath model comparison when appropriate. Ask whether measurement, identifiability, data resolution, task formulation, or evaluation design is the actual limiting issue.

If the main claim cannot be stated accurately in one sentence, do not begin prose revision.

3. Build the Claim Architecture

Create this artifact before restructuring a whole manuscript:

North Star
└── Central scientific question
    └── Main claim
        ├── Claim 1
        │   └── Evidence: result / figure / table / analysis
        ├── Claim 2
        │   └── Evidence: result / figure / table / analysis
        ├── Claim 3
        │   └── Evidence: result / figure / table / analysis
        └── Boundary conditions and unresolved points

Use the architecture across the paper:

• Introduction raises the question.
• Results establishes the claims.
• Discussion interprets and connects the claims.
• Abstract compresses the chain.
• Title expresses the highest supported claim or question.

Do not force three claims when the paper supports two, and do not present several contributions as equally central unless the evidence and design require that structure.

Build the architecture around the strongest supported advantage. It is the main claim, the Abstract and the final Introduction paragraph lead their findings with it, and every other result supports it, sets its scope or moves to the SI. Fix the comparison frame at this step, from what the method is for. The frame never comes from which numbers win, and a comparison the field expects stays in the Results even where the method does not win it (anti-defensive-writing, Emphasis is chosen, the record is not).

4. Map claims to evidence and calibrate their level

Extract substantive claims from the title, abstract, introduction, Results headings, figure legends, and discussion. For each claim, record its exact support and mark it:

• fully supported;
• partially supported;
• unsupported by current evidence;
• hypothesis or interpretation rather than direct finding.

Use the claim ladder:

observation → empirical pattern → interpretation → mechanism → general principle

Write only at the highest level directly licensed by the evidence. Distinguish, for example, performed best under the evaluated settings from is the best representation.

For a claim that exceeds its evidence, take one of three actions:

1. narrow the claim;
2. add or request the missing evidence;
3. label it as a hypothesis, possible explanation, or motivation.

Calibrate the claim itself instead of making an inflated statement and retracting it through a defensive caveat. Use This pattern is consistent with... or One possible explanation is... when a mechanistic interpretation lacks a direct mechanism experiment.

5. Reverse-outline before rewriting

For each section, state its one-sentence function. For each paragraph, record:

• its one main message;
• the evidence or reasoning it contains;
• its relationship to the previous paragraph;
• its contribution to the section claim.

Move, merge, split, or remove material that cannot be mapped cleanly. Organize Results by questions and answers, not by the chronological order of experiments or panels.

Choose the edit depth only after this audit:

• Micro-edit: logic and order are sound; wording is the bottleneck.
• Local rewrite: the paragraph claim is valid, but order, reasoning, or emphasis is weak.
• Structural rewrite: the section question, claim hierarchy, or evidence sequence is wrong.
• Scientific blocker: the evidence is missing, contradictory, or invalid; do not rewrite past it.

Do not default to preserving the original sentence or paragraph structure.

6. Align figures, text, and terminology

Assign every figure, panel and analysis one primary role: claim-supporting evidence, a control that rules out an alternative explanation, definition or methodological bridge, validation in a new setting, scope, practical consequence, or case illustration. An analysis with none of these roles moves to the SI or leaves the paper. A control, a robustness check the claim depends on and a prespecified analysis have a role whatever their outcome. A panel that documents an interface, compatibility or eligibility check (which methods accept which inputs) has none of these roles; it belongs in an SI table or one Methods sentence.

Check that:

• each main figure carries one primary claim unless the paper's logic requires a composite figure;
• every major claim points to the correct figure, table, or supplementary item;
• the Results states the question and main inference rather than merely saying a figure is similar;
• the figure carries evidence density while the text carries the inference;
• panel labels, metrics, baselines, datasets, abbreviations, and numbers agree across all locations;
• negative results and metric disagreements remain visible;
• legends define the evidence without making a stronger claim than the plot supports.

Create a canonical term list for core concepts, tasks, settings, datasets, models, baselines, and abbreviations. Use one primary term per concept. Change terms only when the scientific level truly changes; do not rotate synonyms merely to avoid repetition.

7. Rewrite from macro to micro

After the architecture is stable, revise in this order:

1. section purpose and order;
2. subsection question and claim;
3. paragraph function and evidence sequence;
4. explicit reasoning and transitions;
5. claim verbs and scope qualifiers;
6. terminology;
7. sentence clarity and concision;
8. punctuation and rhythm.

Run a final skeptical review across contribution sufficiency, scientific clarity, empirical strength, evaluation completeness, and method or framework soundness. Point to evidence rather than answering from intuition.

Section contracts

Each position (Title, Abstract, Introduction, Results, Discussion, Methods, legends and SI) is held to its contract in paper-workflow's references/section-contracts.md, which also gives the order in which positions depend on one another. Read the entries for the sections under revision; this skill does not restate them.

At this layer, check three things for each section: whether it does its contract's job, whether it restates a claim that has since moved upstream, and whether it claims more than the evidence it depends on. Revise the positions that compress the argument (Abstract, Title and the final Introduction paragraph) only after the claims they compress are settled.

Paragraph and sentence rules

Use the paragraph as the basic logical unit. Give it one main job and, when appropriate, this form:

topic sentence → evidence → interpretation → transition or implication

Make the first sentence explain why the paragraph exists. Prefer the scientific message to the experimental action. Do not force every paragraph into the template when another order is clearer.

State important inference explicitly. If observations A, B, and C support a modality-dependent effect, write that conclusion instead of asking the reader to assemble it.

Use common, precise words and concrete scientific nouns. Prefer verbs to nominalizations. Avoid noun stacks such as representation architecture robustness evaluation analysis. Keep one main logical relationship per sentence. Split a sentence that simultaneously introduces background, describes a method, reports numbers, interprets the result, and qualifies the claim.

Do not replace readable prose with a sequence of fragments. Align syntactic units with logical units, and retain technical distinctions that change the scientific meaning.

Prohibited patterns

Reject or rewrite these patterns unless the evidence and context specifically justify them:

• experiment logs: We evaluated... We found... We also found... In addition...;
• attempt narratives: We first tried X, which did not help, so we...;
• disconnected proof: We did A and B. We observed C and D. Therefore, X without the missing link;
• procedural paragraph openers that hide the message: To further investigate...;
• ornamental or vague abstractions such as stable performance regimes, intrinsic quality, or comprehensive superiority when ordinary words are more exact;
• empty intensifiers such as remarkably, notably, encouragingly, and promisingly;
• repeated defensive caveats after already bounded claims;
• synonym rotation for one scientific concept;
• causal or mechanistic verbs for observational evidence;
• universal labels such as optimal, general, robust, fundamental, transparent, or state-of-the-art without evidence covering that scope;
• subsection titles that name only the procedure when the supported finding can orient the reader;
• figure references with no stated question, pattern, or inference.

Do not ban a word mechanically. Reject it when it obscures meaning, inflates the claim, or adds no scientific information.

Six-question gate

Before accepting any revised paragraph, answer:

1. Why is this paragraph here?
2. What is its one message?
3. What evidence supports that message?
4. Does the claim exceed the evidence?
5. Can a broad scientific reader follow the reasoning without supplying a key inference?
6. Can anything be removed without losing scientific meaning?

If question 1, 2, or 3 has no clear answer, restructure the paragraph. If question 4 is yes, narrow the claim or obtain evidence. If question 5 is no, add the missing bridge. If question 6 is yes, delete the excess: repetition, reassurance and process narration go; numbers, n, controls and null results are never "excess".

Output standard

For a manuscript audit, report in this order:

1. Verdict: whether the scientific narrative is stable enough to rewrite.
2. Confirmed / assumed / unresolved: the truth boundary used for the audit.
3. Claim Architecture: North Star, central question, main claim, supporting claims, evidence, and boundaries.
4. Major findings: unsupported claims, missing reasoning, structural breaks, figure-text mismatch, metric misinterpretation, and terminology drift, in severity order.
5. Edit depth: micro-edit, local rewrite, structural rewrite, or scientific blocker.
6. Next action: the smallest safe revision step.

For an authorized revision, add:

• complete ready-to-use revised prose or file edits;
• a concise account of material scientific choices;
• unresolved evidence gaps that language cannot repair;
• verification performed.

Do not claim that a manuscript or section is improved merely because the prose is smoother. Claim completion only after the evidence chain, information structure, terminology, and revised text have been checked together.