Managing Revisions and Corrections in Policy Research
Keep research revisions traceable by recording what changed, why it changed and which conclusions or recipients may be affected.
A requested change to a policy research output is not one kind of event. It may be a copy edit, a correction of error, the arrival of evidence that post-dates the analysis, or a request for work that was never commissioned. Each carries a different obligation to the published record, a different approval route and a different cost. The practical discipline is to classify the request before editing anything, decide who owns the decision, and record the outcome against a version number. Researchers who skip the classification tend to make two opposite mistakes: silently amending substantive findings under the label of editing, or treating a genuine error as a matter of interpretation because reopening the text is inconvenient.
Why the category decides the process
A copy edit leaves meaning intact. A correction concedes that the published text was wrong on a point of fact, arithmetic, attribution or method. New evidence may leave the original finding standing while narrowing its conditions. A scope extension is new research wearing the vocabulary of revision. These differ in three ways that matter operationally:
- Effect on the record. Only a correction and, in some cases, new evidence require a visible note explaining what changed and why.
- Who approves. Copy edits usually sit with the author or editor. Corrections and evidence-driven revisions often require the person who held methodological oversight, and sometimes the commissioning body.
- Who pays and who is bound. A scope extension is a new commission with its own method note, review and budget; absorbing it into a revision obscures both the timeline and the basis of the original finding.
A four-part test to apply on receipt
Before opening the document, answer four questions in writing. The answers, not the requester's wording, determine the category.
| Test question | Copy edit | Correction | New evidence | Scope extension |
|---|---|---|---|---|
| Does it change the meaning of a sentence or the direction of a finding? | No | Possibly — the error may be load-bearing | Possibly — a finding may need qualification | Yes, by adding new findings |
| Does it concern material already within the commissioned scope? | Yes | Yes | Usually yes, but dated after the cut-off | No |
| Does it require new collection, analysis or review? | No | Verification only | Judgement about whether the finding survives | Yes |
| Does the public record need a note? | Rarely | Always | Usually, as a dated addendum | No — it is a separate output |
Two boundary cases recur. A change that looks editorial but reverses a hedge (replacing "is likely to" with "will") is not a copy edit, because it alters the strength of a claim. A request that looks like new evidence but actually identifies a source the original analysis misread is a correction, and should be handled as one.
Copy edits: keep the test explicit
The test is whether a careful reader would draw a different conclusion from the revised sentence. Hyphenation, house style, consistent capitalisation of institution names, reference formatting and typographical errors pass without a note, though the version number should still move. Changes to qualifying adverbs, to the order of a conditional clause, or to the precision of a number may fail the test even when they are presented as tidying. When in doubt, treat the change as a correction candidate and check the underlying source.
Corrections: concede the error precisely
A correction should name the location, quote or summarise the original text, state the revised text, give the date and the version, and identify what the error affected. Where the output informed a decision, that last point is the one readers need. A correction that says only "an earlier version contained an error" leaves the reader unable to judge whether their reliance on the document was misplaced.
Distinguish an error from a disagreement. A reader who disputes a judgement, a weighting or the political feasibility of a recommendation has not identified a correction unless the judgement rested on a factual or methodological mistake. Record the disagreement, respond to it, and leave the text standing if it survives scrutiny — but say so in the revision log rather than in the published note.
External frameworks treat the integrity of the published record as an institutional responsibility rather than a personal courtesy. The UK Research Integrity Office's Code of Practice for Research, a UK framework, addresses the correction and retraction of research outputs within institutional research integrity arrangements. That framing is useful for its implication: the correction route should exist before it is needed, and it should be documented rather than improvised. For a small research team or an individual specialist, the practical translation is to agree the route at commissioning — who may request a correction, who decides, where the note appears, and how long the corrected version and the original remain available.
New evidence: qualify, do not automatically withdraw
Evidence that arrives after the stated cut-off date is not evidence of error. The original output should have recorded its search window, the sources consulted and the date of the analysis; with that in place, a later judgment, statutory instrument or dataset can be handled as a dated addendum. The judgement to make is whether the new material undermines a load-bearing finding, narrows its conditions, or merely adds detail.
If the finding no longer holds, treat the matter as a correction and say which finding has been superseded. Should the finding hold under narrower conditions, an addendum with a cross-reference from the executive summary is usually proportionate. For material that is peripheral, note it in the log and include it in the next scheduled revision. What should be avoided is the slow drift of a document through unlogged updates, where the version a reader downloaded in March differs from the version available in September with no indication that anything changed.
Scope extensions: name them and price them
Requests to add jurisdictions, extend a time period, interview additional groups or model alternative scenarios are new work. They deserve a separate proposal, a method note and, where the original had independent review, a fresh review arrangement. Folding them into a revision creates three problems: the original finding acquires an authority it was not tested for, the additional work is not resourced, and the commissioning body loses the ability to see what it has actually bought. A short written response that separates the revision from the extension is usually enough, provided it is issued before any drafting begins.
A working revision log
The following fields are a suggested working format, not a validated instrument or an institutional requirement. They can be held in a spreadsheet or at the front of the project file.
| Field | What to record |
|---|---|
| Request reference and date received | A stable identifier and the date, so the sequence is reconstructable |
| Requester and their relationship to the work | Commissioning body, reviewer, reader, co-author |
| Provisional category | Copy edit, correction, new evidence, scope extension |
| Answers to the four tests | Does it change meaning; is it in scope; does it need new work; does the record need a note |
| Evidence checked | The source consulted to confirm or reject the point, with its date |
| Decision and decision owner | Accepted, rejected, deferred, or converted to a new commission |
| Output affected | Section, table, paragraph or finding |
| Version and publication date | The version that changes, and the date it takes effect |
| Note text and distribution | The wording of any published note, and who was told |
The log is not bureaucracy for its own sake. Its practical value appears at the point where a commissioning body asks, months later, whether a document it relied on has changed.
Fictional worked example
The following scenario is fictional and illustrative; the numbers and dates are invented and imply no standard.
A specialist prepares a fourteen-page regulatory comparison for a fictional regional development agency, published with a stated evidence cut-off of 30 April. Four requests arrive over the following year.
A communications officer asks for hyphenation and capitalisation to be brought into house style. Meaning is unaffected, so the document moves to version 1.1 with no published note.
A reviewer states that a table misreports a transposition deadline. The specialist checks the cited statute and confirms the error: the table gave a date that applied to a different provision. The document moves to version 1.2, and a correction note names the table, reproduces the original date, gives the revised date and states that the error affected one comparative ranking but not the report's overall conclusion.
A court judgment is handed down in the autumn, narrowing the interpretation of one provision discussed in a single section. The finding survives under narrower conditions. A dated addendum is added, with one sentence in the executive summary pointing to it; the original text stands.
The agency then asks for two neighbouring jurisdictions to be added. The specialist declines to treat this as a revision and issues a separate proposal with its own method note, review arrangement and delivery date.
Conditions and limits
No classification rule removes the need for judgement. The right route depends on whether the output is public, contractual or advisory; on what review rights the commissioning body retained; on whether the work informed a statutory or funding decision; and on confidentiality obligations that may require a correction to be delivered to the client rather than published. Where an output feeds a formal decision, the correction procedure of the publishing or deciding body governs, and the author's preference is secondary. Confidential engagements raise a further point: a corrected confidential report does not automatically correct a public summary derived from it, so the two need to be handled together. The considerations set out in this network's guidance on confidentiality for expert consulting apply directly to that pairing.
A next step that takes an hour
Choose one current or recent output and add three things before its next review round: a version number on the cover, a stated evidence cut-off date, and a short paragraph in the cover note setting out how corrections will be received and who decides. Then apply the four-part test to any request already in the file and record the result in the log. Teams commissioning work from specialists can make the same addition to their own templates; the guidance for organisations describes how project-based research is typically structured.