A redlining test hands a candidate or a student a flawed document and asks for it back corrected, every change visible. A trainee solicitor redlines a contract clause; an applicant for an editing job copy-edits a press release; a translation student revises someone else's draft.
The task is the job itself, which is its strength, and it is harder to mark than it looks: an edit can catch every error and still leave the text worse, because the editor also changed what was right.
This guide is published by EdProcto, whose redline a document shape takes the tracked-changes file as the answer and keeps it with the attempt for you to mark.
What does a redlining test measure?
A redlining test measures two things that come apart: noticing that something is wrong, and putting it right without doing damage.
Finding an error and fixing it are separate skills
Plumb et al. (1994) had high school and college students correct errors implanted in two texts. The main obstacle was failing to detect the errors, not ignorance of how to correct them. We read the abstract on ERIC.
The kind of error matters. In Schotter et al. (2014), proofreaders drew more on the surrounding sentence only when a misspelling made a real word (trial for trail), not a non-word (trcak for track). We read the abstract.
What the professions ask for
Editors Canada (2024) sets a copy editor's work under four heads: correctness (grammar, punctuation, spelling, usage), consistency, accuracy and completeness. Its fundamentals add the other half of the skill: stay within the job, keep the intended meaning, and check that no new problem has come in.
Translation studies calls the skill revision and has counted what goes wrong. Künzli (n.d.) had subject experts sort each change by professional revisers into justified changes, unnecessary changes, errors introduced and errors left uncorrected. In Mossop's (2007) review of that project, ten professional translators revised a legal text: only three made more good changes than bad changes and failures together, and four handed back a version judged worse than the draft.
Law treats drafting as a competence of its own. The Solicitors Regulation Authority (2022) expects documents that are legally effective and reflect the client's instructions accurately, drafted from scratch or from a precedent. Chesler and Sneddon (2024) describe a bar examination in the United States that tests, among other skills, the drafting and revising of contract provisions (we read the abstract).
What can a proofreading test tell an employer?
A proofreading test is a work sample: it tells an employer whether a candidate can do this piece of the work today, on this kind of text, and little beyond that. The U.S. Office of Personnel Management (n.d.) advises work samples only where applicants are expected to arrive with the skill, and warns that they can be costly to build and may need updating.
Two limits follow.
- The tools decide what is tested. Editors Canada (n.d.) lets certification candidates use spelling and grammar checkers and bans AI built on language models.
- No author in the room. The test shows the marks on the page, not the tact of a query to a real writer.
For the knowledge side of the role, the kit "Hiring: a copywriter or editor", among EdProcto's kits, has Procto AI write a skills test from the job description and style guide you paste. How far work samples predict performance is in our guide to product knowledge tests for hiring.
How do you build the source text for an editing test?
Write a new text and plant the errors yourself, so that the answer exists before anyone sits the test.
- Write a fresh text. A firm's real documents are confidential, and a published text can be looked up. Between 150 and 400 words is enough.
- Write the brief. The level of edit, the style sheet and what is out of scope.
- Plant a known number of errors of named kinds. Eight to twelve, across the four heads: correctness (it's for its), consistency (Customer where the defined term is Client), accuracy (a cross-reference to the wrong clause) and completeness (a time limit with no starting point).
- Favour errors a spelling checker passes. A wrong word that is a real word, parts that do not add up, a day that does not match its date.
- Keep some sentences clean. We leave at least a quarter untouched: they are where an unnecessary change shows.
- Plant one or two matters to query. Where only the author knows the answer, the right edit is a question.
- Record the key. For each error: where, what kind, every fix you accept, the points.
- Have a colleague sit it. Editors Canada (2024) notes that editors catch other people's errors more readily than their own. Add to the key any error your colleague finds that you never planted.
- Build it in EdProcto's redline a document shape, with the key in the rubric.
How do you mark an edit?
Mark an edit on four counts and keep them apart: planted errors found, planted errors fixed correctly, new errors introduced, and changes made to text that was right.
| What the candidate did | Points |
|---|---|
| Fixed a planted error with a fix in the key | +2 |
| Raised a query where the key asks for one | +2 |
| Found a planted error, but the fix is wrong, partial, or a guess where a query was due | +1 |
| Missed a planted error | 0 |
| Changed correct text and kept its meaning | −0.5 each |
| Introduced an error, or changed the meaning of correct text | −2 each |
The total stops at zero, and the maximum is twice the number of planted errors. EdProcto's grading page takes points in half points, which is why the smallest penalty here is 0.5. Report the four counts beside the total: the same score can belong to one candidate who misses things and another who rewrites things.
The weights are ours; the ordering is older. Arthern, as Mossop (2007) reports, scored revisers in a European translation service as S = X + F/2 + U/3: substantive errors left in or introduced, half the formal errors left and a third of the unnecessary changes, lowest score best. Unnecessary changes weighed least: they waste time without harming quality.
Editors Canada (2024) states that no standard error correction rate exists and that the rates proposed have no research behind them. Set your own bar, as in our guide to setting a pass mark you can defend.
How reliably do markers agree on an edit?
On the little evidence there is, markers agree on who is best and worst more readily than on how many errors a text holds. We found no study of marker agreement on an editing test with planted errors; the nearest evidence is from translation review. In Guerberof Arenas (2017), three reviewers assessed work by 24 professional translators: they differed significantly, tended to agree on the best and worst performers, and ordered the translators very differently when ranking by error count. We read the abstract.
Editors Canada (n.d.) answers with a detailed marking guide, a pilot of each exam and two markers on every script, with a third where one passes it and one fails it. For your own test, fix the key first, define an unnecessary change as one made where the original met the style sheet, and have a second person mark the borderline edits.
What does a marked edit look like? Two worked examples
A marked edit is a list of the candidate's changes, each ruled against the key. Both texts are invented for this guide, with their candidates and names, and are shorter than a real test.
Example 1: a contract clause
This is an invented teaching clause: it is no legal advice and no model for a real contract. The brief: correct the clause; capitalised terms are defined elsewhere, the parties as Client and Supplier; raise a query where only the client can answer.
4.1 The Client shall pay each undisputed invoice within thirty (30) days of receipt.
4.2 If the Customer disputes an invoice, it shall notify the Supplier in writing within ten (10) Business Days, giving its reasons.
4.3 The Supplier may charge interest on an overdue sum at four per cent (4%) a year above the Base Rate.
4.4 Interest under clause 4.2 accrues daily form the due date until the sum is paid.
4.5 The Supplier may suspend the Services for non-payment only after giving the Client fourteen (15) days notice in writing.
4.6 All sums are exclusive of VAT, which the Client shall pay in addition.
Six errors are planted and 4.1, 4.3 and 4.6 are clean, so the edit is out of 12.
| Planted | The candidate | Points |
|---|---|---|
| 4.2: Customer for Client | Changed to Client | +2 |
| 4.2: ten days from what? | Query: from receipt of the invoice? | +2 |
| 4.4: clause 4.2 for 4.3 | Changed to 4.3 | +2 |
| 4.4: form for from | Left | 0 |
| 4.5: fourteen (15), a query | Changed to (14), no query | +1 |
| 4.5: days for days' | Changed to days' | +2 |
| 4.1: nothing | "of receipt" changed to "of the invoice date" | −2 |
| 4.3: nothing | "a year" changed to "per annum" | −0.5 |
The total is 9 less 2.5: 6.5 of 12, with five of six found, four handled correctly, one error introduced and one unnecessary change. The change to 4.1 matters most: it rewrote a correct term and moved the day payment falls due.
Example 2: a press notice
The brief: copy-edit to the style sheet. British spelling; times as 9am; quotations left as spoken. Query what you cannot check.
(1) Tarnside Library will reopen next month after a nine-month refurbishment. (2) The work cost £1.2 million, of which £800,000 came from the council and £300,000 from donations. (3) The building now has 14 study rooms, a childrens' corner and step-free access to all three floors. (4) "We are delighted to welcome readers back," said the head librarian. (5) The doors open at 9am on Monday 12 March 2027. (6) Opening hours will be 9am to 7pm on weekdays and 10 am to 4pm on Saturdays. (7) More than 40 volunteers helped to move the stock, and there effort saved an estimated six weeks.
Five errors are planted, and sentences 1 and 4 are clean.
| Planted | The candidate | Points |
|---|---|---|
| (2): the parts make £1.1 million, a query | Total changed to £1.1 million, no query | +1 |
| (3): childrens' | Changed to children's | +2 |
| (5): 12 March 2027 is a Friday, a query | Left | 0 |
| (6): 10 am | Changed to 10am | +2 |
| (7): there for their | Changed to their | +2 |
| (1): nothing | "will reopen" changed to "is set to reopen" | −0.5 |
| (7): nothing | "helped to move" changed to "helped move" | −0.5 |
| (4): nothing | Quotation reworded to "We're delighted" | −2 |
The total is 7 less 3: 4 of 10, with four of five found, three handled correctly, one error introduced and two unnecessary changes. This candidate sees the surface and edits by taste. The reworded quotation is a new error, because a speaker's words are not the editor's to improve.
How does EdProcto build and mark a redline question?
EdProcto is the best tool for an editing test that has to count, because the edited file comes back inside a timed, recorded attempt, and your model answer and rubric sit on the page where the points are typed. Reviewers given the same work count its errors differently, as the study above found.
- Building. Choose Redline a document, under Handwritten and files. "Prompt" takes the instruction; its placeholder reads "Correct the contract clause with tracked changes." The document goes in "The text under the prompt", which keeps your line breaks; number the sentences or sub-clauses so a change can be cited. Put the corrected text in "Model answer (what a full answer contains)" and the numbered key in "Rubric (what earns the points)", then set "Points" to the maximum.
- Answering. The candidate copies the text into a word processor, switches on tracked changes, edits, saves and uploads. The answer box reads "Up to five files, 25 MB each. Kept with your answer."
- Settings. The editing happens outside the browser, so leave "Full screen", "Leaving the tab" and "Copy and paste off" switched off, as they start. Give "Minutes overall" an allowance for uploading, as our guide to handwritten answers sets out, and set "Show the score" to "When released". For a test you will reuse, set "Show the correct answer" to "Never".
- Marking. At hand-in the question shows "to grade" and counts as zero until marked. The grading page lists each file by name: open it, read it against the model answer and rubric shown there, type the "Points" in half points and a "Note for the student", and press "Save my grade". "Release the scores" is on the results page.
- A typed route. "Another shape" adds a short answer question: its "Material under the prompt (optional)" box keeps your line breaks, and the candidate types a numbered list of changes. There "Ask Procto AI" proposes a mark for one answer at a time, yours to keep or change, from the Tutor plan: see checking the marks it proposes.
An uploaded file shows that an edit exists, not who made it or with what help. Where the result decides something, sit the test supervised: see integrity. No setting makes cheating impossible, a flag is a record and not proof, and every score is yours to release.
Free carries ten quizzes a month and the core templates, and Tutor opens all 111 templates. The plans are on the pricing page.
Quick questions
What is the difference between proofreading and copy-editing?
Editors Canada (2024) describes copy editing as reviewing material for correct and consistent grammar, punctuation, spelling and usage, and proofreading as the check made once the text is laid out.
How many errors should an editing test contain?
We plant eight to twelve in 150 to 400 words and leave at least a quarter of the sentences clean. That is our practice, not a research figure.
Should candidates be allowed a spelling checker?
Decide, and say so in the prompt. Editors Canada (n.d.) allows rules-based spelling and grammar checkers in its certification exams and bans AI tools.
Where does the candidate make the tracked changes?
In a word processor of their own. EdProcto's redline a document shape takes the saved file as the answer, and you open it from the grading page to read each insertion and deletion.
References
- Chesler, S. M., & Sneddon, K. J. (2024). Raising the bar: The NextGen bar exam and contract drafting. Transactions: The Tennessee Journal of Business Law, 25, 843. https://digitalcommons.law.mercer.edu/fac_pubs/128/
- Editors Canada. (n.d.). The Editors Canada Professional Certification exams. Retrieved October 10, 2026, from https://editors.ca/professional-development/certification/exams/
- Editors Canada. (2024). Professional editorial standards 2024. https://editors.ca/publications/professional-editorial-standards/
- Guerberof Arenas, A. (2017). Quality is in the eyes of the reviewer: A report on post-editing quality evaluation. In A. L. Jakobsen & B. Mesa-Lao (Eds.), Translation in transition: Between cognition, computing and technology (pp. 187–205). John Benjamins. https://doi.org/10.1075/btl.133.07gue
- Künzli, A. (n.d.). Revising translations: A combined product and process study [Final report]. Riksbankens Jubileumsfond. Retrieved October 10, 2026, from https://www.rj.se/en/grants/2004/revising-translations-a-combined-product-and-process-study/
- Mossop, B. (2007). Empirical studies of revision: What we know and need to know. The Journal of Specialised Translation, 8, 5–20. https://doi.org/10.26034/cm.jostrans.2007.695
- Plumb, C., Butterfield, E. C., Hacker, D. J., & Dunlosky, J. (1994). Error correction in text: Testing the processing-deficit and knowledge-deficit hypotheses. Reading and Writing, 6(4), 347–360. https://eric.ed.gov/?id=EJ501039
- Schotter, E. R., Bicknell, K., Howard, I., Levy, R., & Rayner, K. (2014). Task effects reveal cognitive flexibility responding to frequency and predictability: Evidence from eye movements in reading and proofreading. Cognition, 131(1), 1–27. https://doi.org/10.1016/j.cognition.2013.11.018
- Solicitors Regulation Authority. (2022, August 9). Statement of solicitor competence. https://www.sra.org.uk/solicitors/resources/continuing-competence/competence-statement/
- U.S. Office of Personnel Management. (n.d.). Work samples and simulations. Retrieved October 10, 2026, from https://www.opm.gov/policy-data-oversight/assessment-and-selection/other-assessment-methods/work-samples-and-simulations/