Revision Integrity Audit

A revision succeeds when it repairs the intended problem and preserves what readers need to understand the research. Fewer words, smoother sentences, and a persuasive response letter are not sufficient evidence of success.

Establish the comparison

Identify the before and after versions, the requested change, authoritative study evidence, and whether the user wants findings or edits. Use supplied files or explicit excerpts, not a remembered earlier draft. If a version is missing, audit the available text and identify what cannot be compared. Do not invent a change history. Preserve original files during an audit.

Choose the relevant discipline, subfield, and study design from the materials. Review scientific validity with the methodological knowledge the claim requires, then review reader comprehension and expression. A proof needs its hypotheses and quantifiers checked; an interpretive study needs its analytic commitments and evidentiary relationships checked. Do not impose experiments, statistical tests, or reporting conventions from another design. Roles can be separate passes by one agent. Additional agents require authorized scope; a role title does not establish credentials or independent verification.

Define success before judging the rewrite

Reconstruct the substantive question and the passage's job. Record briefly what a reader should be able to understand or conclude, the source of that expectation, what the earlier version supplied, and what the current version supplies. Derive expectations from the research question, retained claim, actual review comment, or supplied requirement. Label an inferred reader need as an inference. Do not invent a reviewer's intentions or a venue rule.

Judge two questions separately. Did the revision address its intended concern? Did it preserve or correctly repair the scientific content and reasoning? A version can improve readability yet leave the concern unresolved, or answer a reviewer while introducing a different error. If no change objective was supplied, assess preservation without inventing a target defect.

Compare meaning and reader function

Read the surrounding passages in both versions. Map consequential propositions and their relationships, not just matching words. For substantial edits, read comparison decisions and examples.

Check the research question, motivation-to-method bridge, distinct component roles, operational definitions, comparator conditions, populations or support sets, endpoints and denominators, assumptions and quantifiers, evidence links, uncertainty, interpretation, and claim boundaries when they matter to the edited passage. Do not turn this into a universal reporting checklist. Preserve valid professional terms; classify a misleading name separately from an incorrect operation.

Check additions as carefully as deletions. New causal language, implied mechanisms, broader populations, or stronger guarantees may appear even when all numbers remain unchanged. A replacement such as “overall success” can silently erase a composite endpoint. Splitting a sentence can detach a condition from the claim it restricts. Fluent connectors cannot replace missing premises.

Check whether a smoother or shorter version retains the relationships that made the facts meaningful. It may preserve all nouns and values yet lose why a method follows from the problem, why a comparison matters, or how a finding qualifies an interpretation. Accept changes to rhythm and structure that improve these relationships; do not restore a mechanical template or require novelty in every sentence.

For an introduction revision, compare the before/after narrowing line and the target venue's needs. Verify that removing explanation or compressing conference prose has not erased the unresolved issue, its consequence, or the reason the contribution follows. For a journal, check that gap and motivation remain explicit without restoring repetitions or detours. Accept deletion of explanation that served no distinct reasoning function. If a premise is lost, restore only the missing bridge, not the entire earlier paragraph or a stock gap/motivation template.

Distinguish preserved meaning, useful compression, supported correction, justified relocation, harmful loss, unsupported addition, and unresolved change. The earlier version is a baseline, not ground truth. Do not restore an error, unsupported claim, or redundant defense merely to maximize similarity. Judge the replacement against the evidence and intended scientific meaning.

For moved material, inspect the destination, reference, and point of use. An accessible appendix may carry detailed settings, but readers still need the operation and consequential scope where the claim is made. If the destination was not supplied, mark the relocation unverified rather than declaring it lost or safe. Retain restatements that serve different readers or section functions, such as a standalone abstract and the full method.

Repair only the actual loss

Recommend the smallest restoration, bridge, correction, or local rearrangement that keeps the valid gains of the new version. Restore the missing meaning rather than copying the entire earlier paragraph. If the retained claim requires unavailable evidence, specify the dependency or narrow the claim within authorized scope; wording cannot substitute for that evidence.

For paired replies, verify both the disposition of the concern and the manuscript passage itself. A statement that a definition was added is not the definition. Keep reviewer-directed acknowledgments and change-status language in the response, while the paper explains the research directly. Recognize already resolved concerns without adding text just to show activity.

Draft concise connected prose with concrete subjects and stable terminology. Default to no explanatory colons or sentence-interrupting dashes; retain compound hyphens, notation, ranges, ratios, exact quotations, URLs, and required templates. Author instructions take precedence. Avoid both repeated assurances and deletions that remove necessary conditions or interpretation.

Deliver and stop

Give the overall outcome, consequential differences with locations in both versions, evidence, reader consequence, and the smallest action. Include explicit useful-compression and keep decisions. A compact comparison can use expected function, before, after, judgment, and action, but do not produce a sentence inventory when a few findings suffice.

Distinguish an acceptable revision, a partial repair, a regression needing repair, and a change that cannot be verified from the supplied material. Several can apply to different passages. Do not call the whole manuscript verified after inspecting only excerpts. Report source comparison, rendering, and submission separately.

Stop when the in-scope changes have supported dispositions and requested repairs are checked. Do not equate a word-count target with success, force a second rewrite, or restore the old version wholesale.