# Writing and editing English articles Use this guide to draft, revise, and translate English articles that readers can understand on the first reading. Choose what readers need, explain how the claims follow from the evidence, and then polish the sentences. Follow the publication's conventions and stay within the requested scope. Use the examples to judge similar passages. ## Readers and purpose Identify the intended readers, what they already know, and what they should be able to explain, do, or decide after reading. Go beyond naming the topic: identify the relationship readers should understand or the judgment they should be able to make. Use that goal to choose the opening, depth, and material to include. An overview can present the main points and their relationships across the requested scope; a tutorial should teach readers to perform the task. Decide which prerequisites to assume and which to explain: programming experience does not imply knowledge of statistics or finance. For substantial work, record the reader's main question, the supported answer, the knowledge assumed, and the evidence needed in a short working brief. When an article serves several purposes, choose which one determines the reading order. Place secondary comparisons, surveys, and reference material where readers can use them without interrupting an explanation or procedure. Keep the brief and other editorial notes out of the published article and its HTML comments. Announce the audience only when it helps readers judge the scope. Organize the article around its central question and answer. The body should explain the meaning and scope of the title and summary, the main reasons for the answer, and assumptions that affect the conclusion. Account for reading time and the number of new concepts as well as clarity within each sentence. ## Requirements, defaults, and article choices | Level | How to apply it | Scope | | --- | --- | --- | | Requirements | Meet these in every article. | Factual and logical accuracy, evidence, necessary conditions, faithful attribution, and consistent technical meaning across languages. | | Defaults | Follow these unless an established convention or a reader's need justifies a different choice. | Spelling, voice, terminology, notation, and presentation. | | Article choices | Choose for the reader's goal; treat sections with the same role consistently. | Structure, headings, depth, examples, and tone. | Give alternatives comparable headings, terminology, units, assumptions, and comparison criteria. Sections with different roles, such as a derivation and a procedure, may use different forms. Let the explanation determine sentence length, examples, and equation placement. Preserve coherent existing conventions during a small edit. During a substantial revision, reassess structure, tone, and headings together. Choose departures from defaults to help readers, and retain the same standards of accuracy and evidence. If a passage seems "AI-like," identify the specific defect: unnecessary repetition, irrelevant detail, generic interpretation, mechanical rhythm, or claims that exceed their evidence. Revise that defect while preserving accuracy, useful explanation, accessibility, and honest attribution. Detector scores and word counts cannot establish whether an article explains its subject well. ## Content selection Review each proposed topic with the definitions, examples, comparisons, qualifications, and cross-references it requires. Temporarily remove the group and ask what the intended reader loses: an explanation of the answer or its reasons, a condition needed to judge its scope, or information needed to apply or verify it. Omit the group if removing it leaves no gap at the intended depth. A topic's relevance to the field, factual correctness, or later mentions do not by themselves justify including it. Apply the same test to existing and new material. Do not add examples or later uses merely to justify introducing a concept. For retained material, use "Main text, notes, and captions" to decide where it belongs. Before adding an explanation, check what the surrounding passage already establishes. Fill the remaining gap and account for any further terms, background, or calculations the addition would require. ## Article patterns Choose a starting sequence for the article's purpose and adjust it to the reader's knowledge and intended depth: | Purpose | Starting sequence | Evidence and emphasis | | --- | --- | --- | | Understand a concept or mechanism | Problem and answer → concrete example → mechanism → implications and conditions | Use representative examples and explain why the result follows. Detailed derivations can remain in notes when the goal is an overview. | | Compare alternatives or estimate a result | Question → main assumptions → method and results → effect of changing conditions → conclusion | Align comparison criteria, units, and conditions. Distinguish sourced values, assumptions, calculations, and judgments; show which assumptions affect the conclusion. | | Complete a procedure | Intended outcome → prerequisites → steps → verification → common failures | Give executable actions and observable checks of success. Include failure cases that help readers complete the task. | | Consult a specification or method | Scope and use → definitions → conditions and behavior → examples → limitations | Make notation, guarantees, and implementation rules easy to locate. Use explicit formal definitions. | | Understand an experience or choice | Situation and purpose → experience or observations → reasons for the judgment → conditions where it applies | Use concrete first-person observations. Keep firsthand experience, research, and recommendations distinguishable. | A section can use a different sequence when its purpose changes. For example, a conceptual article may include a procedure for implementing the method. ## Translating and synchronizing articles Edit each language version for its intended readers while keeping shared facts, technical definitions, the author's position, and the strength and conditions of claims consistent. Each version must support its own conclusions. Its outline, examples, and length may differ from the other version's; rebuild the article when that would explain the subject better. Apply the same standards for reading effort and relevant detail in both languages. Choose background for the audience's knowledge and purpose; language alone does not establish subject expertise or a need for more detail. Apply "Content selection" to material from either version and to proposed additions. Verify new examples and derivations as carefully as retained ones. ### Read both versions before drafting Read the source article and any existing English version before drafting, even when replacing the English from scratch. Record two kinds of information in the working notes: - **Shared content:** definitions, assumptions, claims, mechanisms, examples, equations, results, evidence, attribution, and limitations. Identify which requested corrections affect claims in the English article and why. - **Candidate explanations:** openings, comparisons, examples, interpretations, and terminology that may help the English reader. Separate factual corrections from changes made for a particular audience. Apply corrections wherever they affect the English article, including fluent passages with obsolete definitions or invalidated examples. If the versions disagree materially and the requested change does not resolve it, check the underlying derivation, implementation, or source. ### Rebuild the explanation in English Outline the explanation for English readers around the problem, the choices and their reasons, and the results and their meaning. Organize by these units of meaning instead of mapping Japanese sentences to English paragraphs. - Combine short source sentences when a subordinate clause makes their relationship easier to follow. Split a dense sentence when it contains separate steps or claims. - Supply subjects and logical connections that Japanese leaves implicit when their meaning is clear from context. Check ambiguities before assigning an actor, cause, or scope. - Reorder or regroup sections to preserve the dependencies of the argument. One version may develop a method from a use case while the other presents a result and then derives it. - Replace examples when another makes the mechanism clearer. Explain prerequisites the audience needs and compress background it already knows. Keep the support and qualifications needed for each conclusion. - Distinguish the author's choices, mathematical consequences, observations, and predictions. Fluent wording must preserve their different kinds of support. Update cross-references and figure numbers when the organization changes. ### Review meaning and English Check that shared claims incorporate the relevant corrections with their conditions and strength intact. Read the English on its own and outline its opening, comparisons, and conclusion. Check what each part helps readers understand, whether connections and references are clear, and whether the article supports its answers. Repair interruptions or missing connections; keep a shared outline when it serves both audiences. Reassess additions that fill no gap and omissions that leave one. Combine these checks with "Editing and review". ## Voice and point of view Use familiar verbs and concrete descriptions. Choose a tone for the article's purpose and keep it stable across passages with the same role. Definitions and specifications usually suit neutral prose; personal observations can be conversational and use contractions. Use a supplied style sample to understand word choice, formality, explanatory depth, and paragraph development. Follow explicit instructions for the current article before habits inferred from the sample. Copying sentence lengths or punctuation frequencies will not capture those choices. Keep the sample's facts, experiences, and opinions separate from the draft author's. Without a sample, preserve clear phrasing and deliberate stylistic choices in the draft. Use "I" for the author's decisions, experience, or argument, and "my" for the author's work. Use "we" and "our" for an identified group. Inclusive "we" can guide the reader through a derivation, as in "We can now substitute this value into the equation." Distinguish that use from claims about a research team's work. When translating Japanese self-reference, use these pronouns instead of repeatedly naming the author or writing "the author" or "this author." Use "our family" for family experience. Preserve the author's role when discussing publications. "My earlier paper" can refer to work with coauthors; use "our earlier paper" when the collaborators are established. Mention lead authorship or coauthorship when the distinction matters and is supported. Replacing a lead author's reference to their work with "a paper I contributed to" can obscure that role. Use "this article" when identifying its purpose or scope helps the reader. Otherwise, state the point directly. Remove announcements that only repeat what the surrounding text already establishes. Before removing a preview, check whether it explains the author's reason for writing, addresses the reader, or marks a transition. For example, "I use this script every week, so here's how it works" connects experience to a reason for sharing it. Preserve that role when revising; a statement of what the author does can lose the invitation to the reader. Write idiomatic English that an international technical audience can follow. Prefer familiar word combinations to slang or cultural references that need extra explanation. Retain concrete experiences and candid judgments in personal reviews. Keep "I used," "I researched," and "I would choose" distinct; a recommendation based on features is not firsthand experience. Do not invent experiences, emotions, discovery stories, or verification work to make the prose sound personal. Connect judgments to their reasons and consequences. Explain which comparison changes a choice or which boundary case breaks an implementation. "This is important" leaves that connection unstated. Preserve truthful disclosures of tools and assistance, including whether the author independently verified an analysis. Separate checks performed during editing from explanations readers need in the article. Describe formatting, conversion, or verification work when omitting it would change the interpretation of the material or when the publication requires it. If "The model's output is reproduced unchanged" establishes the comparison, do not add that Markdown was converted to HTML for display and the wording was not edited. Disclose omissions, data selection, or other processing that affects how readers interpret the material. ## Sentences, paragraphs, and connections ### Organize around the reader's next question Open with the problem, observation, or task that motivates the article. State the main result early enough to orient readers, then develop the reasoning in the order they need to understand it. Introduce terms as they become useful. A concrete example may reduce the background needed before explaining a method. Include an anecdote or rhetorical question when it clarifies the subject; avoid delaying the answer for suspense. Replace an opening that promises to "explain the structure, examine the details, and discuss the implications" with the supported finding and its relevance. Keep the conditions needed to interpret it: a result on one dataset must not become a general guarantee. When announcing a list, put scope or methods notes before the announcement so the items follow it directly. When introducing a method or tool, explain what problem it solves. If existing approaches motivate it, show how the proposal differs and what that changes in practice. For example, "Caching the result lets later requests reuse it without repeating the calculation" connects a mechanism to its benefit. Once the motivation is clear, develop the explanation without restating the benefit at each step. Give each paragraph a clear purpose. Start from something the reader understands and develop the new point. A claim followed by its mechanism and consequence is one useful structure. Keep a comparison or causal explanation together when splitting it would make readers reconstruct the connection. A worked example can clarify an earlier claim; a synthesis can explain how earlier results fit together. Choose order and depth for the reader's goal. Explain relationships readers may struggle to follow in more detail, and be brief about what they already understand. Passages with different roles or importance need not use the same number of sentences or the same progression. Keep corresponding items consistent in comparisons and procedures so readers can find the information they need. Use headings, lists, and tables to make definitions, steps, or comparisons easy to locate. Keep connected reasoning in paragraphs. Give list items labels when they help readers find or compare information; remove labels that merely repeat the following sentence. Removing a label and removing list formatting are separate decisions. Keep lists when readers need to scan independent requests, responsibilities, deadlines, or steps. Turn a list into prose when doing so makes the connections easier to follow without making needed information harder to find. ### Make relationships explicit Use a clear subject and a verb that states what happens. Prefer "The cache stores the result" to "Storage of the result is performed by the cache." Passive voice is useful when the object or procedure is the topic and the actor does not matter, as in "The samples are sorted by timestamp." Check that subjects, verbs, and complements express the intended relationship. Name what acts or has a property. "The load consumes energy during the eclipse" identifies the consumer more clearly than "The eclipse requires energy." Compare the same kinds of things: "a computer doing useful work" and "a resistor" are devices, whereas "useful computation" and "a resistor" mix an activity and a device. Check headings and summaries for these errors too. When one subject governs several verbs, check that it performs each action in the process being described. Distinguish an event that triggers processing from the component that carries it out. Name the component if omitting it obscures who performs an operation or the sequence of steps. Keep established shorthand when the context makes the actor and sequence clear. Explain relationships readers would otherwise have to infer to follow the next claim. After distinguishing projected area from radiating surface area, for example, say which the comparison uses for equipment size and which the calculation uses for heat rejection. Build on what the context establishes and check unresolved claims before explaining them. Use "because," "so," "although," or "whereas" to express the actual relationship. Adjacency in a source does not establish causation. Remove connectives such as "Furthermore" or "Importantly" when they add no connection. Keep alternatives parallel. For example: "The first method sorts each block and then merges the results. The second merges the blocks and then sorts the result." The repeated structure makes the difference in operation order easy to see. Explain its consequence when that matters to the current question and is not already clear from the context. Use a noun after "this" when its referent would otherwise be unclear: "this approximation," "this ordering," or "this bound." Repeat a technical term when precision requires it; replacing it with a near-synonym for variety can suggest a different concept. Keep lists parallel in grammatical form and level of detail. Check that a shared verb or modifier applies naturally and accurately to every item. Let the material determine the number of items. Keep comparison criteria and conditions consistent, but give each alternative the space its evidence requires. Remove items added only to reach a preferred count or make the sides look balanced. ### Revise rhythm through meaning Read neighboring paragraphs together. Repeated openings such as "The X ...", actions used as subjects ("Doing X causes Y"), semicolon pairs, and conclusions introduced by "so" or "therefore" may reveal repeated explanations. Consolidate those before changing sentence openings or punctuation. Choose sentence length for the explanation, independently of any Japanese character-count target. A short sentence can state a result; a longer one can keep its reason or condition attached. When the actor matters, name it: "The parser rejects rows without a timestamp" is more direct than "Omitting the timestamp causes row rejection." Use an action as the subject when the action is the topic, or a semicolon when it makes parallel alternatives easy to compare. Move from familiar information to the new point, even when that means keeping the same opening word in consecutive sentences. Split a sentence when readers must hold too many relationships at once; combine sentences when a condition, cause, or contrast becomes easier to follow. Avoid both overloaded sentences and disconnected short ones. Use questions, fragments, and unusual punctuation only when they fit the voice and purpose. When several conditions modify a noun, or conditions interact with negation, compare the sentence with a version that explains the subject, conditions, and action separately. Choose the version whose modifiers and negation readers can understand without rereading, even if it needs more words or sentences. ### Remove padding while preserving meaning Decide which explanations to keep before shortening individual sentences. Preserve the relationships needed to understand the claims: a shorter phrase can make reading harder if it introduces an unfamiliar term or hides an inference. When shortening a passage, check whether the removed words identified the actor, limited the scope of a term, or made clear which statement a condition applies to. Restore words whose omission leaves readers to infer a relationship that was explicit. Brevity alone is not an improvement. Respect the requested length. Use direct constructions when the longer form adds no useful distinction: | Before | After | | --- | --- | | Due to the fact that the buffer is full, the writer waits. | Because the buffer is full, the writer waits. | | The implementation performs a comparison of the keys. | The implementation compares the keys. | | The index is stored in order to avoid a second scan. | The index is stored to avoid a second scan. | | It is important to note that the bound applies only to sorted inputs. | The bound applies only to sorted inputs. | | This document serves as the user guide. | This document is the user guide. | | The package boasts a command-line tool. | The package has a command-line tool. | Use "is," "are," or "has" to identify something or state what it contains. Keep "serves as" or "functions as" when it distinguishes a role, behavior, or temporary use from identity. Review "not X, but Y," "not just X, but Y," and contrasts split across sentences for the information each half supplies. For example, "This is not just a cache. It is a way to reuse results" can become "The cache lets later requests reuse results." Keep a negative half that corrects a relevant misunderstanding or states a distinction the reader needs. Remove a contrast that only makes the positive claim sound more impressive. Review repetition by purpose as well as wording. The opening gives the main answer, the body develops its reasons, and the conclusion explains what readers can understand or decide from the results. Merge or remove passages that repeat an explanation without serving another purpose. A conclusion can restate a key finding; it need not summarize every section or have a separate heading. After several examples or a long explanation, check whether readers can bring the main findings together and follow the next topic. If they must reconstruct how the individual explanations relate to the article's main question, briefly summarize the relevant findings. When the conclusion moves into reflection or advice, make its connection to the preceding explanation clear. Let the length and complexity of the explanation, and the size of the shift in topic, determine how much recap is useful. Check whether different words refer to the same thing within a sentence or neighboring sentences. For example, if "feedback" and "input" refer to the same comments, "We will use your feedback to assess the draft and revise it based on your input" can become "We will use your feedback to assess and revise the draft." Keep the distinction when the words refer to different material or serve different roles. Before editing a summary sentence, identify what it helps readers do at that point. Consider omitting or merging it if it only restates the preceding explanation for the same purpose. Keep it when it brings examples together, recalls the main points of a long explanation, or connects to the next topic, even if the information has appeared before. Read versions with and without it, and compare how easily readers can follow the explanation while retaining the point and its conditions. - **Remove:** "You can apply online without visiting the office. This lets you submit your application without going to the office." The second sentence repeats the first. - **Keep:** "The office accepts applications only during weekday office hours, but online applications are accepted at any time. People who cannot visit during office hours can still apply." The second sentence explains who benefits from the difference in opening hours. - **Recap earlier material:** After several examples of edited prose, write "In these examples, repeated explanations were reduced, and missing reasons or steps were added," then move on to advice about requesting edits. Even if each change has already been discussed, bringing them together helps readers connect the comparisons as a whole to the next topic. Explain a recurring concept where readers first need it. Later passages should develop the explanation or help readers recall earlier material they need to follow it. A reference article may also need a short reminder for readers entering midway. After an example, retain an interpretation that answers the reader's question. Remove a generic ending such as "Understanding this requires considering the relevant factors" when the preceding explanation already identifies those factors and their effect. Keep comparisons where they affect the explanation or decision. Remove recurring endings such as "The alternative also does this (see §X)" when the comparison section already covers them. Use cross-references to locate evidence, definitions, or procedures readers need, and make the reason for following them clear. Use a comparison section to explain which differences affect the choice. Collect checks where readers will use them. A checklist can repeat facts to prevent mistakes during a procedure; its introduction and conclusion need not recite them again. In a diagnostic table, distinguish observed failures from consequences inferred from the specification. Verify an inference before presenting it as expected behavior. Describe what readers should notice in a table rather than announcing its row headers. "The table compares cost, capacity, and latency" adds little when those labels are visible. Keep introductions that identify comparison cases, units, assumptions, or a non-obvious reading convention. Prefer an informative label such as "Estimated payload bytes" to "Memory use" followed by a paragraph explaining that the estimate omits most kinds of memory. End a section when its explanation is complete. Add a comparison, caveat, summary, or preview only when readers need it there. Replace announcements of importance and vague praise with a supported description of what changes, under which conditions, and why it matters. Name the operation with verbs such as "stores," "compares," or "allocates." Remove intensifiers such as "dramatically" when they substitute for evidence. Prefer "use" to "utilize" when they mean the same thing. Keep technical verbs that express distinct claims: "prove," "estimate," "approximate," and "suggest" are not interchangeable versions of "show." Check who acted, what caused the result, and what was completed after shortening. "The team began trying to deploy the system" does not mean "The team deployed the system." Preserve the difference between an intention, an attempt, an ongoing process, and an achieved result, along with the original conditions and attribution. ### Use headings that explain the section's purpose For headings at the same level with the same role, align grammatical form, specificity, and the aspect being compared. Choose a pattern suited to the sections: | Pattern | Useful when | Example | | --- | --- | --- | | Topic noun phrases | Organizing concepts, specifications, or estimates | "Equal keys in a stable sort" | | Questions | Answering the reader's questions in sequence | "What happens to equal keys in a stable sort?" | | Declarative clauses | Building an argument through supported findings | "A stable sort preserves the input order of equal keys" | | Instructions | Guiding actions the reader will perform | "Preserve the input order of equal keys" | Default to short topic noun phrases for a technical explanation when another pattern offers no clear benefit. A section must answer its heading's question, support its claim under the stated conditions, or provide the steps it promises. Keep headings specific when aligning their form. A procedure inside a conceptual article may use instructions. Conventional headings such as "Introduction," "Conclusion," and "References" identify a role in the article and can retain their usual form. When renaming an existing heading, preserve the section's role and tone. An abstract phrase that could fit many articles does not make a heading more specific. Check whether the new heading makes a personal closing observation sound like a general lesson or claims more than the body supports. Keep capitalization consistent, preserving proper names and acronyms. If numbering sections, use the same scheme for comparable sections; a reference list may remain unnumbered. Update cross-references after renaming or renumbering. Read the headings alone to check the argument's order and the treatment of parallel sections. A new article title should identify the subject and a supported contribution or practical question. Keep title and summary edits within the requested scope. ## Terminology and definitions Introduce terms where readers need them to understand an object, operation, or result. A term mentioned in the introduction may be better explained later, where readers use it. When the operation is the topic, explain it while introducing its name. Keep phrases such as "called X below" only when readers need the name at that point. For example, "Dividing each count by the positive total normalizes the counts to proportions that sum to 1" explains normalization through its action and result. Adding "This process is called normalization" would repeat the name without advancing the explanation. Joining that addition to the preceding sentence or putting it in parentheses leaves the repetition intact. When readers only need to understand an operation, its description may be enough. "Storage returns 90% of the energy used to charge it" explains "90% round-trip storage efficiency" and prepares readers for a calculation using 0.90. Retain the technical name when readers need it later. When replacing a technical term with a plain description, check that the description preserves the objects, conditions, and properties the term distinguishes. If the description broadens the meaning, keep the term and explain it. For example, replacing "lossless compression" with "making a file smaller" drops the property that the original data can be recovered exactly. Use standalone definitions when the meaning, scope, or distinction is the subject, especially in specifications and notation sections. "Let ..." or "Here, ... denotes ..." can introduce a formal definition. Use a term consistently after introducing it. Explain an extension or new distinction where it arises; repeat a definition when a passage read on its own needs it. State unit conventions and representation details when ambiguity could change a calculation or implementation, even for expert readers. When a specialized statistic matters, explain what its value tells readers in the body. State the scope of definitions and conventions. Identify a choice made for a particular method or model so readers can distinguish it from a property of the underlying concept. Use established English terms. Define an unfamiliar acronym when it becomes useful; omit an acronym that will not be used again. Distinguish ordinary technical nouns from official names. Japanese glossary capitalization does not determine capitalization in English prose. Keep official product names, API names, symbols, and code identifiers exact. Judge English word choice by meaning, usual combinations with neighboring words, and fit with the audience. Check whole phrases: verbs with their objects and prepositions, and nouns with their modifiers. A phrase that translates poorly into Japanese may be ordinary English, so vocabulary review lists from other languages need to be reassessed for English. Choose articles, number, and quantifiers deliberately. "A method" can introduce one possible method; "the method" points to an identified one. "Each input," "at least one input," and "the inputs in this experiment" make different claims. Check these choices when translating Japanese expressions that leave number or definiteness implicit. ## Technical reasoning and strength of claims Explain what a result means at the depth the article requires. An overview can explain the mechanism and consequence; an article teaching a method should walk through the input, key steps, and result. Choose representative examples for that purpose and state what they illustrate. Continue with the same example when it helps explain why something happens, which conditions matter, or how to respond. Show which information or action changes. When using a different example, check what new distinction or relationship it clarifies. Before combining or omitting examples, check the conditions, outcomes, and effects on readers each illustrates. Even when examples illustrate the same mechanism, retain a brief explanation of any difference readers need to understand the subject or make a decision. Keep numbers and comparisons that support a conclusion, explain a mechanism, inform a choice, or help readers reproduce or verify a result. Omit unused calculations and their methods. If a value is followed immediately by a warning that it cannot answer the question, reconsider its purpose. A counterexample demonstrating a metric's limitation can be useful when that is the question. ### Explain quantities and calculations Choose the order of prose, notation, and equations for the reader. A specification may start with a definition followed by symbol explanations; an introduction may need intuition or an example first. For quantities that affect the conclusion, explain what is measured or calculated and what the value is relative to: - **Averages:** State the period, population, or other averaging basis when it affects interpretation: "43 kW averaged over the orbit" rather than just "43 kW on average." - **Fractions and per-unit values:** Identify the denominator and its scope. "Heat input per square meter of each face" distinguishes a value per area from a total per face; say whether the area includes inactive parts when that matters. - **Several values:** Distinguish separate examples, bounds, observed ranges, and uncertainty intervals. State what changes and what stays fixed in a comparison. Put the information needed to interpret a result in the body or table heading. For example, if two scenarios require 2,450 m² and 4,180 m², label each result with its scenario. A bare range, "2,450–4,180 m²," can make readers mistake two examples for bounds. Explain the roles of terms and operations when readers would otherwise have to infer the mechanism. For example, to calculate net heat rejection from two faces, subtract absorbed power from emitted power for each face, then add the results. Choose the explanation's position relative to the equation according to what readers already understand. Preserve comparison direction, operation order, units, ranges, and quantifiers when editing calculations. Distinguish exact results from approximations, bounds from typical values, and per-case results from averages. Retain edge cases that affect the conclusion or its use. A smoother sentence must not turn a sufficient condition into a necessary one or a possibility into a guarantee. ### State interpretations at the strength the evidence supports State definitions, verified facts, and their consequences directly within their scope. Use "may," "suggests," or "I expect" for possibilities, inferences, and predictions where appropriate. Use "can" for capability or behavior that occurs for some inputs. For example, retain "The ratio can exceed 1" if that is true for some inputs but not all; "exceeds 1" would change the claim. An interpretation should explain what an observation supports and why. "Consistent with" means the observation fits an explanation without identifying it as the cause. Replacing it with "suggests that the system was designed to ..." can invent both a cause and an intention. Discuss uncertain mechanisms when they help answer the article's question, and distinguish them from observations. During a prose edit, preserve the original reasons and evidence for a claim. Do not substitute a different reason merely because it supports the same conclusion. Changing "I chose this route because it has fewer transfers" to "I chose this route because the fare is lower" replaces the reason for the choice. When filling a gap in an explanation, retain its reasoning and add the connection readers need to follow it. | Evidence available | Appropriate claim | Scope to preserve | | --- | --- | --- | | A formula under explicit assumptions | State the derived result directly under those assumptions. | A hypothetical calculation does not describe what a deployed system actually did. | | A configuration or specification | "The configured limit is ..." | A declared limit is distinct from the largest case tested. | | Measurements on selected inputs | "For these inputs, implementation A used less memory than B." | Identify the scope of the comparison. Keep exact inputs and procedures in a note unless they affect interpretation. | | An observation compatible with several mechanisms | "One possible explanation is X; the observations do not distinguish X from Y." | Name only relevant alternatives supported by the evidence. | | Results from a simplified test | State what the simplified test establishes. | Do not merge it with a full-system test into a claim that the full system passed both. | Keep comparisons within their evidence. "Faster in this benchmark" is narrower than "faster," and a complexity bound differs from measured runtime. Fewer operations do not by themselves establish a proportional runtime improvement: operations can take different amounts of time. Preserve the workload and quantity being compared. The same rule applies to other subjects; for example, a purchase-price comparison alone does not establish total cost of ownership. Explain why a design choice supports the claimed benefit. Caching results requires opportunities for reuse before it can shorten response times. Distinguish benefits shared with existing methods from the proposal's new contribution. Give terms such as "robust," "stable," "optimal," and "significant" a precise meaning and appropriate support. For judgments such as "excessive," "fair," or "desirable," state the purpose or criterion. Keep summaries within the claims and conditions the body substantiates. Verify that an attributed source exists, supports the particular claim, and establishes the stated conditions. One study does not establish that "researchers agree" or that "many studies show" a result. Correct unsupported conclusions instead of protecting them with "may" or a disclaimer. Distinguish a disproved claim from one whose source you cannot verify. Report unresolved evidence to the author instead of inventing an attribution or silently deleting the claim as false. Explain the evidence for a factual correction. Failure to find information does not establish that it is private or unpublished. ### Support recommendations For recommendations and time-sensitive facts such as prices, schedules, and facilities, state the research date and identify details readers should confirm before acting. Present prices as ranges or estimates when the evidence calls for it. Explain selection criteria before a list or comparison. Include differences and drawbacks that affect the intended use, and distinguish sourced facts from ratings and recommendations. Disclose affiliate links plainly near the beginning, before readers encounter them. ### Avoid defensive qualifications Use "is designed to," "aims to," or "is intended to" for a goal supported by evidence, including when its outcome remains unverified. If a definition or verified result establishes that a method rewards matches, write "The method rewards matches." Calling this a design goal would weaken an established claim. Introduce shared assumptions at the start of the article, section, or example, then state their consequences directly. Repeat a condition when the topic changes or omitting it would change the interpretation. Remove recurring phrases such as "under these conditions" while the scope remains clear. Test a caveat or negative clarification by removing it temporarily. Keep it when its absence changes the claim's scope or certainty, or leaves a specific misunderstanding that affects the reader's understanding, decision, or action. Review repeated phrases such as "X alone does not determine Y" or "remains to be checked" together to see whether they serve different purposes. | Purpose of the statement | Editorial action | | --- | --- | | It changes how the result can be used. | Keep it beside the claim, with the reason when readers need it. | | It repeats an assumption that still clearly governs the passage. | Keep the shared assumption and remove the repetition. | | It repeats a distinction already clear from a positive explanation. | Remove the negative restatement unless it resolves a specific remaining ambiguity. | | It repairs an overly broad claim or label. | Narrow the claim or label first; retain any remaining necessary condition. | | It denies an irrelevant stronger claim. | Remove the denial and any discussion introduced only to support it. | | It identifies an open question central to the article. | Say what remains unknown and, if useful, what observation would resolve it. | | It merely says more research could be done. | Remove it. | For example, "This is computed, not measured" can become "The calculated storage requirement is ..." when the calculation's basis remains clear. "If condition C holds" can establish a hypothetical calculation without a second disclaimer about whether C held in a deployment. When changing a negative statement into a positive one, preserve its scope: "A alone does not determine the result" does not establish that A and B together fully determine it. ## Main text, notes, and captions Keep the answer, its main reasons, and assumptions that affect the conclusion in the body. Include representative examples that explain the mechanism. Use notes for supporting details readers can skip without misunderstanding the main point. Choose footnotes, appendices, or expandable notes to suit the publication, and omit unrelated material. In an overview, place details needed only to reproduce a calculation or measurement in notes. These may include coefficient sources, derivations, exact inputs, execution environments, and checks. A statistic can appear in a table or figure without being repeated in prose. In a tutorial, keep the steps and conditions readers need to perform the task in the body. If a precise value requires a lengthy setup that interrupts the main explanation, move the value and setup into a note together and explain the result's meaning in the body. The visible claim must remain accurate with the note closed. Give each note an entry point that tells readers what it adds, using its link text or the passage to which it is attached. Start with the additional evidence or explanation. A long note can help readers verify or extend a result; give unrelated topics separate entry points. Reassess note placement when translating. English readers may need a definition kept in a Japanese note, or an English clause may replace a full Japanese sentence. Read the body without following notes, closing expandable ones. Readers should understand the answer and main reasoning without notes or figure captions. State what figures and tables establish without reciting every value. Use labels such as "Figure 1" and "Table 1," with a space before the number, and give figures concise, informative titles. Captions should identify what is shown and how to read the relevant relationship, adding an interpretation when useful. Check each figure or table with its caption on its own: retain the subject, units, and conditions needed to interpret it, even when the body also states them. Keep reproduction details in notes unless they affect the comparison's meaning. A detailed table and its setup can go into a note together. Alternative text provides access to information in the image. Preserve the relationships and values needed for that role even when they also appear in the body or caption. Judge repetition within each reading path: the body alone, the body with a note, a figure or table with its caption, and the image's accessible description. Each may need information also present in another path. Translate chart labels, legends, alternative text, table headings, and interactive controls as well as the body. Retain terms or identifiers already appropriate in English. Check that labels fit without changing the data or meaning. ## English notation and local context - Use English punctuation and spacing: spaces between words and after commas and sentence punctuation, ordinary parentheses, and a space after a colon. Replace Japanese full-width punctuation in English prose. Preserve mathematical and code syntax. - Use customary compound modifiers when they clarify the grammar, such as "fixed-size buffer." Avoid inventing compressed noun strings when a clause or prepositional phrase would be easier to understand. - Use conventional number and unit forms, such as "32 MB," "5 ms," and "20%." Preserve the claim's units and precision. Write dates unambiguously in prose, for example "July 8, 2026." - Distinguish mathematical notation, code, and ordinary prose. Use code formatting for identifiers, and emphasis for distinctions that need attention. Ordinary technical terms can remain in plain text. - Use natural English word boundaries. Remove manual breaks inside ordinary words; reassess phrase groupings carried over from another language. - Identify the country, currency, institution, or service when readers need that context. Preserve the original setting and the author's experience; avoid inventing a local equivalent for English readers. Use established English names while preserving official titles and bibliographic attribution. ## Editing and review Scale this sequence to the change. Use it for a new article or substantial revision; a small edit can retain the article's existing choices and focus on the affected passages. Combine stages when convenient, while resolving accuracy and structure before polishing sentences. For revision, use the text supplied for the current edit as the reference. Do not restore removed explanations merely because they appeared in an older version. Perform these checks before returning the revised text, keeping comparison notes out of the article. 1. **Select content.** Use the working brief to settle the reader's goal, the supported answer, and the article pattern. Check sources, calculations, and assumptions. Apply "Content selection" to new and inherited material. Correct unsupported conclusions. 2. **Organize the explanation.** Use "Main text, notes, and captions" to place the selected material. Briefly outline the roles of the headings and paragraphs, then check the order and depth against the reader's goal. For a substantial revision, compare this with a short outline built around the central question or a concrete example, keeping the topic and scope intact. Check that the author's perspective, emphasis, and natural voice survive. Choose the structure that makes the explanation easier to understand or needed information easier to find. Keep alternatives parallel where that aids comparison. Choose the English structure for its readers, independently of the Japanese where useful. Read the end of each section together with the opening of the next. Check whether readers can follow the connection when the topic or level of detail changes. If the transition feels abrupt, consider changing the order, briefly recalling earlier material, or explaining the relationship. 3. **Check gaps and repetition.** Read the body without notes, then follow the notes, then inspect figures and tables with their captions independently. Check alternative text as an accessible description of the image. Fill gaps in reasoning and necessary conditions in each reading. Remove detours and explanations that repeat the same purpose. Use "Avoid defensive qualifications" to review caveats and negative restatements. 4. **Edit sentences.** Read the revised English without looking at the source. Check subjects, actions, word combinations, references, modifiers, and connections. Read neighboring paragraphs aloud to hear repeated sentence shapes. Simplify cumbersome or needlessly formal wording while keeping repetition that makes comparisons precise. Check terminology, pronouns, tone, headings, spelling, punctuation, units, and dates in prose, metadata, notes, captions, labels, alternative text, and controls. 5. **Compare and select changes.** Compare the revision with the supplied text; check factual corrections against verified evidence. Check subjects, conditions, quantities, the scope of technical terms, comparison direction, operation order, causality, attribution, and claim strength. Check averaging bases, denominators, ranges, and the distinction between observed, calculated, and inferred results. Verify shared claims across languages. Also compare the role of each sentence or heading and the way the author addresses the reader. For each substantive edit, identify what it improves in accuracy, fluency, or ease of reading. Revert changes that make the passage worse, and wording changes with no identifiable benefit. Better organization does not establish that every wording change is better. When unsure about an edit, compare a version that reverts only that change, keeping it consistent with the surrounding passage. Keep the original wording when neither version is more natural or clearer and the request does not require a change. 6. **Recheck changes made during review.** When a check leads to another edit, repeat steps 4 and 5 for that passage and the surrounding material it affects. If the structure changes, repeat step 2 as well. Check for newly introduced repetition, ambiguity, or shifts in meaning. Repair links, citations, and cross-references, and confirm that the edits stay within the requested scope. Use word and phrase counts to locate passages for review. Verify counts and review claims against the source, and state what was counted: prose, code, tables, captions, alternative text, or notes. Assess each match in context before changing it. Sentence length, opening words, punctuation, and pronoun frequency have no prescribed targets; they do not establish authorship or a universal standard for human writing. Judge a length change by what readers gain or lose, within the requested length. Before delivery, remove conversational wrappers accidentally copied into the article, such as "Certainly!", "Here is the revised version," "I hope this helps," or offers to expand the draft. Remove generic statements about a model's knowledge limits that add nothing about the article's sources or research scope. Preserve truthful disclosures of tool use, research dates, and uncertainty needed to interpret the conclusions. Read the finished article as its intended reader. For each central claim, check whether readers can explain why it holds and what changes as a result. If the text and the assumed background knowledge do not let readers follow the connection, supply the missing explanation while preserving the original reasons and evidence. Check that the article answers the central question at the intended depth, each passage serves a purpose, readers can follow the reasoning, and the claims match their evidence. Finish when the review finds no remaining concrete problem and no further change has an identifiable benefit.