Skip to content

Structured Peer Review

Pattern · All patterns
Evidence · 48 claims (30 for, 18 mixed) · 56 studies (17 causal, 11 quant-synthesis, 9 associational, 8 review, 4 qualitative, 4 theoretical, 3 design), q1–q4 · 12 of 56 report an effect size · 34 claims rest on one study

Description and scope

A reusable policy for improving a draft through peers' criterion-referenced comments, and for building reviewers' sense of quality by having them write those comments: a reviewable draft (an essay, a proposal, a prototype, a recorded performance) and a short set of criteria → reviewers practise on an anchor draft of known quality → each reviewer reads one to three peers' drafts and writes comments that name a criterion, point to a place in the draft and suggest a change → the author reads the comments, decides which to act on and writes a revision plan → the author revises before the work is judged → both roles reflect on what they will do differently in their own next draft. The intended change is in the quality of the revised work (skill-performance, scored on the task's own rubric), in each reviewer's own later work (near-transfer to their next draft), and, as a by-product, in how accurately and helpfully learners comment on drafts (assessment-accuracy).

It sits beside two converted pages and does not repeat them. The collaborative evaluation pattern owns learning to judge quality: judging alone, comparing judgments in a group and checking them against a reference. This page starts where that one ends, with a single author's draft and comments meant to change it; where reviewers cannot yet tell strong from weak work, run collaborative evaluation on anchor drafts first. The Assessment for Learning principle owns the general relationship (evidence read against shared criteria, low stakes, a revision step before the outcome is judged); this pattern is one configuration of it in which the evidence is read by peers. The Peer Feedback/Peer Review principle and Process-Based Writing are the principles it applies; the writing workshop is a larger routine that can contain it.

No claim in this wiki tests structured peer review as a whole: none compares the full cycle here (calibration, criterion-referenced comments, required revision, reflection) with a version lacking one of its parts, or with teacher comments alone, on the revised work. The claims that bear on it are of five kinds: peer feedback on writing against no feedback, self-assessment or teacher feedback (one meta-analysis, postsecondary); peer assessment against no assessment or teacher assessment on academic performance (one meta-analysis); whether giving feedback goes with gains in one's own work (one meta-analysis pooling both roles and one correlational study); what rubric guidance changes in reviewers' comments and ratings (two quasi-experiments); and how accurate and helpful secondary-school reviewers' comments are (two correlational studies). Smaller, weaker sets shape single steps and rows: which comments given and received go with later project scores (one correlational study of 181 secondary students online), what reviewers attend to while reviewing (one think-aloud study of twelve), and how peer marks compare with instructors' and what biases them (tool evaluations read second-hand in one systematic review, and observational analyses of MOOC peer grading). One claim limits the model: in the peer-assessment meta-analysis, a script or rubric did not raise the effect over freeform review. Further claims bear on single steps rather than the cycle: rubric-referenced self-assessment of one's own draft, contingent fading of support in tutoring, the specificity of AI-generated feedback and the revisions that follow it, a theoretical case for training reviewers, and students' doubts about novice peers in a technical subject. Each is carried to this pattern by extrapolation, and the page says how far. The sequence below is the earlier page's five steps, with a calibration step and a revision plan added, each labelled with its evidence status; the response-dependent policy is an untested design proposal.

Inputs: establish the design brief

Field Record or elicit
Learner and starting observation Before the first review, each learner's written comments on one anchor draft whose quality the teacher has already judged: how many comments name a criterion, point to a place and suggest a change, and whether the problems they flag are real ones. Record prior knowledge of the domain and genre, age band, language of instruction, confidence in writing or in critiquing, and status or friendship differences in the class. Preserve unknowns rather than labelling learners novice.
The draft and the task What is reviewed (essay, report, design, code, presentation), how long it takes to read, whether its quality turns on correctness (a calculation, a factual or technical claim) or on qualities informed readers weigh (argument, structure, audience fit), and how far it is from finished. A draft that cannot still change is not worth reviewing for the author.
Criteria and review prompts Who writes them (given, co-constructed from contrasting exemplars, or chosen by the author), how many, whether each has a described level and an anchor example, whether they are simple and objective (format, presence of a citation) or subjective and complex (quality of reasoning), and what the comment prompt asks for.
Context and activities Setting (classroom, online-instructor-led, online-self-paced, workplace-clinical), single-session or days-weeks, how many reviewers per draft, named or anonymous, synchronous or asynchronous, where comments are written, how much time authors have to revise, and whether a teacher sees the comments.
Learner-valued goal Ask each author what they want from the review (a better grade, a clearer argument for a real reader, to know whether the draft works at all, reassurance) and each reviewer what they expect to get from reviewing. Record disagreement with the designer objective.
Designer objective Which capability is the target: a better final version of this piece, better own writing on the next piece, accurate and useful commenting, or the habit of using comments to revise. Name the one the assessment will weigh.
Outcome The revised draft scored on the task rubric against the first draft (skill-performance), at the final submission; the reviewer's own next draft (near-transfer); or comment quality scored against a teacher's reading of the same draft (assessment-accuracy). The amount of revision is not this outcome: authors often revise in response to comments of mixed quality.

A request to "add peer review to the course" does not specify these. Ask what draft, against which criteria, how many reviewers, how much time to revise and what will be scored before choosing anonymity, prompt format or the number of cycles.

Sequence and conditional policy

One review cycle (about 60–75 minutes of contact time over two meetings, or the same steps spread over four to seven days online; proposed):

  1. Collect a draft that can still change (before the session). Set the draft deadline far enough before the final deadline that authors have two to five days to revise. Evidence status: untested; the earlier page's step 1. The writing meta-analysis and the peer-assessment meta-analysis both compared feedback on work that was then revised or retested, not on finished work.
  2. Calibrate on an anchor draft (15 minutes; early cycles only). Give three or four criteria, each with a described level. The teacher thinks aloud through one anchor draft: criterion, place in the draft, level, and one change that would raise it. Reviewers then write comments on a second anchor and compare them with the teacher's. Evidence status: rater training in peer assessment [+M] was the strongest moderator of peer assessment's effect in a 58-study meta-analysis; guided-rubric training [+M] helped MOOC writing raters place samples at the right level, mostly on subjective, complex criteria. That short, task-aligned training units make peer comments more specific, criterion-referenced and evidence-based than a rubric with generic guidance is an expectation stated for EFL speaking classes [+W], derived from earlier writing studies and not tested in that article. Why anchors may help: in a think-aloud study of twelve students giving feedback on argumentative texts, eight compared the draft with their own picture of what such a text should look like [+W], and that comparison was what let them identify and evaluate errors; calibration aims to make that picture shared and accurate. Qualitative, no comparison. The earlier page's step 2.
  3. Review two or three peers' drafts in writing (20–30 minutes per draft), so that each draft receives two or three reviews. Each comment names a criterion, quotes or points to a place, and suggests a change; each review opens with at least one thing that works and should stay, because reviewers left to themselves attend to errors. No marks. Evidence status: peer feedback on writing [+M]: across 24 higher-education studies, writers who gave and/or received structured peer feedback improved their writing more than no-feedback controls (g = 0.91) and than self-assessment only (g = 0.33); the nature of the feedback moderated the effect and the number of peers did not. Peer assessment [+M] improved academic performance over no assessment across 54 studies (g = 0.31). Several reviews per draft rest on one small SWoRD experiment [+W] (28 students, reported in a systematic review with no effect size): drafts improved most when the author received several peer reviews and least with a single expert reviewer. Asking for problems and suggestions follows Lu & Law's online peer assessment study [+W] (181 secondary students), where the number of suggestions and of problem-identifying comments [+W] a student gave predicted their own project score; the "works and should stay" opening follows the same study's finding that positive affective comments received [+W] predicted the author's project score, and a think-aloud study in which twelve students focused more on errors than on what worked [~W], both when giving and when processing feedback. All three are correlational or descriptive. The two-to-three-draft load is a proposal. The earlier page's step 3.
  4. Author reads, chooses and plans (10 minutes). The author sorts each comment into "will act on", "will not, because…", and "need to ask", writes a two- or three-line revision plan, and rates each review's usefulness (helpful, partly, not) with one reason the reviewer will see; where comments conflict, the author decides by the criteria and may ask the teacher. Evidence status: untested proposal. The usefulness rating is carried from peerScholar [+W], where letting students grade the reviews they received raised a low correlation between expert and peer marks (one tool evaluation, reported second-hand, no coefficient printed); it is an accountability move for reviewers, not shown to change the revision. It is added because comments are often not acted on: in one classroom study of feedback in a digital game [~M], only 4.5% of feedback instances to fifth-graders went from being noticed to measurable progress, with the largest loss between reading and acting. That was an on-screen agent's feedback, not a peer's. The rationale for writing a plan is self-monitoring [+M]: a review and a theoretical synthesis argue that feedback becomes useful when learners compare their current performance with a target and revise; neither studies peer review.
  5. Revise before the work is judged (two to five days). Before submitting, the author scores the revision against the same criteria. The final version is submitted with the revision plan, and the teacher scores the final version, not the comments. Evidence status: carried from the comparisons in steps 3 and 7; no claim isolates the revision requirement. The self-check is carried from rubric-referenced self-assessment [+M]: in two non-randomised studies (162 middle-school writers; 218 pre-service teachers), self-scoring one's own draft against a rubric came with higher-quality work, with no standardised effect size; the second also found more stress and a performance-avoidance orientation. The earlier page's step 4.
  6. Reflect in both roles (5–10 minutes). Each learner writes one thing they changed because of a comment and one thing they noticed in a peer's draft that they will use or avoid in their own next draft. Evidence status: untested; the earlier page's step 5.
  7. Check the reviewer's own next draft. Score each learner's next piece on the same criteria they reviewed against. Evidence status: the assessor benefit [+M]: in 43 undergraduate teacher-education students, the quality of feedback a student gave predicted their own final project, controlling for its initial quality, while the feedback received did not; correlational, so it supports including the reviewer role, not a claim that reviewing causes the gain. In 181 secondary students' online peer assessment [~W], the same pattern held: giving suggestions and problem-identifying comments predicted the giver's project score, while the cognitive feedback a student received did not predict theirs, and the authors conclude weak students did not benefit from receiving more of it. That is the reason this step checks the reviewer's own work and not only the author's revision.

Across cycles (four to six cycles over a term, proposed; untested):

  • Cycles 1–2: teacher-written criteria with anchors; calibration (step 2) every cycle; anonymous drafts; comment prompts with sentence frames ("Criterion…; at…; consider…"); the teacher reads a sample of comments and returns two or three to each reviewer with a note.
  • Cycles 3–4: learners co-construct or refine one or two criteria from a strong and a weak exemplar; calibration shrinks to one anchor; frames optional; authors attach one question they want reviewers to answer.
  • Cycles 5–6: named review where trust allows; authors choose the criteria they want feedback on; reviewers comment without frames; the teacher reads comments only where an author flags a disagreement.
  • Move on to the next stage when, on two consecutive cycles, most of a reviewer's comments (for example, four of five) name a criterion, point to a place and suggest a change, and the teacher agrees the flagged problems are real. Keep a reviewer on anchors and frames while fewer than half do. Both thresholds are proposals to set locally. Withdrawing support as each reviewer's comments show they can do without it is carried from contingent scaffolding [~M]: mostly small one-to-one tutoring studies (long division, the circulatory system), where support adjusted to each response beat fixed or absent support; in one meta-analysis explicit strategy training did better than scaffolding, and no study here fades support in peer review.
Observation Discriminating probe Proposed action, with its uncertainty
Comments are praise or generic ("good flow", "fix grammar") Ask the reviewer to point to one place in the draft for each comment, and to say which criterion it concerns. If they can, the prompt is the problem: require criterion, place and suggestion in each comment. If they cannot find a place, the criterion may be beyond their domain knowledge: narrow the criteria or calibrate again on an anchor pair that differs only on it. Neither branch is tested for peers; revising an AI-feedback prompt toward essay-grounded guidance [+W] lowered the odds of low-specificity feedback in one semester's writing course, which is a prompt to a model, not to reviewers.
Comments flag problems that are not problems, or miss serious ones Compare the reviewer's comments on an anchor with the teacher's. If the reviewer misjudges the anchor too, criterion knowledge is missing: more calibration, and a teacher pass on drafts where correctness matters. If they judge anchors well but not peers', friendship or status may be colouring the review: make review anonymous. In two secondary-school writing studies, reviewers' own writing did not predict whether they found real problems, so do not assume stronger writers catch more.
Authors ignore the comments, or revise only surface features Read the revision plan against the comments; ask the author why a substantive comment was not used. If the author did not understand it, have reviewer and author talk for five minutes. If the author disagreed with reasons, that is a legitimate decision. If the author had no time, move the draft deadline earlier. If the comments themselves were all surface-level, ask each reviewer for at least one comment on meaning, organisation or argument. Proposed; in one two-class quasi-experiment [+W], revisions after AI-supported diagnostic review held larger shares of meaning-change, organisation and argument work than control revisions, which stayed more surface-level; AI feedback, not peers', and no effect size.
Authors change everything a reviewer said, including wrong advice Ask the author which criterion each change served. If they cannot say, they may be deferring to any reader: keep the "will not, because…" column and ask for at least one reasoned rejection per cycle. Proposed.
Reviewers write useful comments but their own next draft does not improve Ask what they saw in a peer's draft that they could use, and score their next draft on the same criteria. If they cannot name anything, add the reflection step's second prompt and have them review their own draft against the criteria after reviewing a peer's. If they can, the gap may be in execution, not judgment: give practice on that criterion. The assessor benefit is correlational, so a null here is not surprising.
Peer scores (where reviewers also score) run high or low against the teacher's for some drafts Score a sample of drafts yourself and compare; check whether the gap follows the reviewer's own attainment (strong reviewers marking weak drafts down, weak reviewers marking strong drafts up) or a shared group between reviewer and author. If it follows attainment, assign each draft reviewers of mixed attainment and average their scores; if it follows a shared group, hide authors' names and origins and assign reviewers across groups; either way, spot-check a sample (reactive control) and tell reviewers their comments' quality is what counts (proactive control). Basis: in MOOC peer grading, bias was least between similar-scoring peers [~W], the best students marked the worst work down and the weakest marked the best work up, and graders rated work from their own country 3.6% higher [~W]; both observational. The two kinds of quality control [+W] come from an inventory of peer-assessment practice, untested as a comparison.
Learners hold back criticism, or one learner's review hurts another Compare named and anonymous comments where both are available; ask privately. Make review anonymous to authors (the teacher still sees who wrote what), add a "what works and should stay" field, and keep marks out of review. Anonymous sharing [~W] is one course's authors' account, untested.

Fitting the design to a situation

The three facts that most change the decision:

  • Whether reviewers can see quality in this kind of draft. Rater training was the strongest moderator of peer assessment's effect, and rubric guidance helped most on subjective, complex criteria; but stronger secondary-school writers found real problems no more often than weaker ones, only commented more helpfully. If unstated, ask: given one strong and one weak draft, can learners say which is stronger and point to why, without help?
  • Whether the author can still revise before the work is judged. The tested comparisons are of feedback on work that was then improved or retested; comments on finished work have nowhere to go. If unstated, ask: how many days lie between the draft deadline and the final deadline, and will the final version be scored?
  • Whether quality turns on correctness or on judgment. Where it turns on technical or factual correctness, peers can be confidently wrong; where it turns on structure, clarity and audience fit, peer comments are where the writing evidence lies. If unstated, ask: what does a serious error in this draft look like, and could a classmate spot it?
If the brief says… Then change… Basis
Reviewers are novice in the genre or domain Run collaborative evaluation on two anchors first; give three criteria with described levels and sentence frames; one draft per reviewer; the teacher reads and returns a sample of comments each cycle; keep the teacher's own comments on content accuracy. Rater training [+M] (moderator across 58 studies, roles pooled); rubric guidance [+M] (MOOC writers; graduate students)
Learners are advanced or practising professionals Let the author name the criteria and the questions to answer; drop frames and anchors; ask for one substantial alternative rather than a list of fixes; use the rubric as shared vocabulary, not a score sheet. Rubric guidance [~M]: in a graduate course a rubric changed comment content and restrained reflective and divergent commentary
Learners are child (under 12) Two criteria phrased as questions, with a picture or worked anchor; review in structured pairs, spoken first and then one written suggestion; the teacher models with a class draft each time; revise in the same or the next lesson. Collaborative writing [+S]: in one small randomised study, 10–11-year-olds writing in structured pairs gained more in their individual writing; that was co-writing, not peer review. Peer assessment [+M] found no difference by education level
Learners are adolescent secondary-school writers Anonymous drafts, four or five reviewers per essay with a well-specified rubric, and the average of their ratings shown to the author beside the comments; teacher reads flagged disagreements. Peer accuracy [~M]: AP English students' averaged rubric ratings correlated with expert scores about as well as teachers' did, in lower- as in higher-performing schools; correlational. Web-SPA [+W]: 76 high-school computing students' peer marks were broadly consistent with the instructor's and work improved after review (a before-and-after report with no comparison group, read second-hand)
Learners have low confidence or low literacy, or are struggling writers Short drafts or one section; a "what works and should stay" field first; spoken comments recorded or scribed; one criterion per cycle; the teacher adds one direct comment so the author is not relying on peers alone. Untested proposal. Process writing [~M] did not improve struggling writers' compositions (grades 1–12), which is a reason not to rely on the cycle alone for them; in one online peer-assessment study, the cognitive feedback students received did not predict their project scores [-W] and the authors conclude weak students did not gain from receiving more of it (correlational), so let these learners review as well as be reviewed
Quality turns on correctness (a calculation, code, a clinical or technical claim) Peers comment on clarity, structure and whether claims are supported; the teacher or an expert checks correctness unless there is an objective check reviewers can run (does the code pass the tests, does the answer match a worked one), in which case give them that check and let them judge against it. Peer accuracy [~M]: secondary-school writing only; peer ratings of programming assignments [+W] correlated with instructors' and were more accurate on objective than subjective judgements (one tool evaluation, read second-hand); students new to statistics [~M] valued peers' alternative explanations but did not trust peer feedback for final application, because their peers were beginners too (one qualitative case study); the split of roles is untested
Criteria are subjective or complex (argument quality, audience fit), not simple and objective Spend calibration on those criteria; do not train or prompt for simple ones, which a checklist can cover. Rubric guidance [+M]: MOOC writing raters, benefit concentrated on subjective, complex criteria and negligible on simple ones; in programming, peers judged objective qualities more accurately than subjective ones [+W] without training, which is where calibration has least to add
online-instructor-led or online-self-paced, asynchronous Use a tool that assigns drafts at random, hides authors' names, locks a review once submitted, shows the author all reviews together and lets the author rate each review's usefulness; a calibration anchor before the first review; a deadline for the revision plan. Assessor meta-analysis [+M]: computer-mediated peer assessment did better than paper; peer assessment [+M] found no online/offline difference; rating received reviews [+W] raised peer–expert agreement in one tool's evaluation. The other tool rules are untested
Mixed modes: some learners in the room, some remote, or phone-only Run review asynchronously for everyone in one shared document or form, so no one reviews by ear; keep drafts short enough to read on a phone (one to two pages, or one section); in-room learners write comments in the same place rather than talking them across the table. Untested proposal
One teacher for a large class, or no teacher reaches each pair Two to three reviewers per draft at random, mixing reviewers' attainment; the teacher reads comments only where an author flags a disagreement, spot-checks a small sample of drafts against the peer scores, and gives one class-wide note on the commonest problem; use averaged ratings rather than single reviews. Peer feedback on writing [+M]: number of peers did not moderate the effect; several peer reviews beat one expert's [+W] on draft-to-final improvement in one 28-student experiment; peer accuracy [~M]: averaged ratings, one subject, correlational; MOOC grader bias [~W] varied with the gap between grader's and author's attainment, observational
single-session only One short draft brought to the session; 10 minutes on one anchor; one review of one draft; 10 minutes to plan and start revising in the room; the author leaves with the plan. Untested proposal
High stakes: peer ratings would count towards marks Keep review formative; mark the final version (teacher) and, if anything, the quality of comments given against the teacher's reading; never the peer rating of a peer's work. Grading the quality of reviews rather than peer-assessed marks [+W] reduced tensions and raised student–instructor mark agreement over three semesters of one undergraduate course (reported second-hand, no effect size); tying part of the reviewer's mark to review quality is a proactive quality control [+W] in one inventory of practice; rubric use [~M] came with more stress and a performance-avoidance orientation in pre-service teachers' self-assessment, carried to peer review
Adults in a work setting (design teams, clinical or teaching placements) Review the real work product against the workplace's own standards; the author states what decision the draft must support; a senior colleague reviews alongside peers and requires a revise-and-resubmit where evidence is missing. Probing feedback with revise-and-resubmit [+W]: a case study of teacher candidates in clinical placements, instructor feedback not peer; progress was not linear
The product will go to a real audience (a client, a publication, a public showing, a hiring panel) Reviewers read as that audience: one criterion is "what would this reader do or misunderstand"; where possible include one reviewer from outside the class; the final version goes to the audience after revision, not the draft. Authentic audiences [+S]: seventh-graders' compositions to distant peers were rated higher than to the teacher; a small study, not of review
Status or friendship differences, reviewers and authors from distinct sites or countries (a MOOC, a cross-site course), or reviews have hurt before Anonymous to authors and with authors' names and origins hidden from reviewers (named to the teacher); "what works and should stay" first; random assignment of reviewers, across sites where there are several; no marks on reviews. Anonymous sharing [~W], authors' account of one undergraduate physics course, no test; in one MOOC, graders rated work from their own country 3.6% higher [~W] than work from elsewhere, a bias of graders and not a difference in the work, observational
Peer comments are combined with AI feedback on the same draft Keep the peer step and its revision plan; let the AI comment on surface and technical points and ask peers for audience, argument and context; have authors say in the plan which source each change came from. See the AI-feedback evidence on the Peer Learning principle, which this page does not repeat: there, adding AI feedback to peer feedback [+M] raised rated writing over peer feedback alone in one university L2 quasi-experiment. Here, AI-supported diagnostic review [+W] went with more meaning-level revision in one two-class study. The division of labour is untested

Fourteen of the sixteen rows cite a claim, and each Basis cell says where the claim's learners, task or design differ from the row's; the low-confidence, correctness, status and AI rows lean on a qualifying, weak or second-hand claim, and the low-confidence and AI rows' moves are themselves untested proposals. Two rows (mixed modes, a single session) are untested proposals with no claim. The large-class, high-stakes, online and status rows were sharpened on 2026-10-19 by reports of peer-assessment tools and MOOC peer grading; those are tool evaluations read second-hand in one systematic review, or observational analyses of grading bias, and none compares the full cycle.

Choosing configurations from evidence

Do not rank structured peer review against teacher comments from these claims: the meta-analytic comparison with teacher feedback is inconclusive, the one small experiment favouring several peers over one expert is read second-hand, none tests the full cycle, and none measures the time it costs. Do not read a long list of revisions as learning, or a reviewer's fluent comments as their own improved writing. A precise abstention names the missing comparison (the full cycle against peer comments without calibration, or without a required revision plan, on the revised draft and on the reviewer's next draft, at the end of a term) and the local observation that would stand in for it.

The evidence by question

Further evidence, not yet read against this model

Claims the earlier page cited, and others that bear on it, not used as core evidence above. Each was re-read against this model on 2026-10-15 or 2026-10-19 and found not to bear on a step or a row; each label says what it is and how far it bears.

Illustrative design instance and observation record

Invented illustration, not a tested course. A part-time master's module in public health (adult, postsecondary, most students working) runs online-instructor-led with one tutor for 60 students, some joining weekly seminars by phone; the assessed piece is a 2,500-word policy brief written for a named local decision-maker. Students have written essays before but not policy briefs (novice for the genre). The table changed the default in four ways: review runs asynchronously in a form, so phone users read a one-page section rather than the whole brief; the first cycle opens with collaborative evaluation of two anchor briefs the tutor has scored; one criterion is "what would the decision-maker do or misunderstand after reading this"; and peer ratings do not count towards the mark. In cycles 1–2 each student reviews two anonymous briefs with sentence frames; the tutor returns three comments per reviewer with a note; authors submit a revision plan within three days and revise over the following week. By cycle 4 authors name the two criteria they want feedback on, and the frames are optional. One student says she wants "something my director would actually read"; the designer's objective is a brief that meets the criteria. They agree to score the final brief on the rubric and to ask the student whether she would send it.

Record: draft and the criteria as given → each reviewer's comments on the anchor, against the tutor's → reviewers assigned, named or anonymous → each comment: criterion named, place, suggested change, real problem or not → author's revision plan (acted on, rejected with reason, asked) → final version scored on the rubric against the draft → reviewer's next draft scored on the same criteria → candidate interpretations (criterion not understood, domain knowledge missing, comments ignored for lack of time, deference to any reader, status or friendship) and their basis → reobservation at the next cycle. Preserve each author's stated purpose and any change in it. An agent may summarize this record but must not fabricate missing inputs or treat the number of revisions as improvement.

Elements and limits

Peer review, peer assessment, feedback, rubrics, exemplars, revision, assessment, reflection and peer discussion. Principles the pattern applies: Peer Feedback/Peer Review, Formative Assessment, Assessment for Learning, Multiple Methods of Assessment, Process-Based Writing and Authentic Audiences and Purposes. Calibrated Peer Review is a strategy that implements steps 2 and 3.

This pattern is scoped to peers commenting on an individual's draft that the author then revises before it is judged. Learning to judge quality without a draft to improve is collaborative evaluation; co-writing a shared text is collaborative learning; the teacher-side loop of reading evidence and choosing a next action is formative assessment. The policy supports observation and design reasoning; whether the full cycle improves the revised work and the reviewer's own next draft more than peer comments without calibration or a required revision plan, for which learners, at what time cost and at what horizon, remains to be tested.

Claims

Examples

  • Graduate seminar proposal reviews using a shared rubric.
  • Design teams critiquing prototypes before final presentation.
  • Writing classes using peer comments to guide draft revision.
  • Calibrated Peer Review — learners rate calibration samples of known quality before reviewing peers' work and receiving reviews of their own.

Key Sources

Page metadata
Field Value
type pattern
id structured-peer-review
title Structured Peer Review
description A reusable policy in which learners comment on one another's drafts against explicit criteria and each author then revises before the work is judged; expected to improve the revised work, and possibly the reviewers' own later work, where reviewers are calibrated on the criteria, comments point to the draft and suggest changes, revision is required and stakes are formative; no claim tests the full cycle, though peer feedback on writing and peer assessment are each tested against no feedback.
status review
grain_size lesson
author McAndrew, Goodyear, and Dalziel
canonical True
generated by claude/unspecified on 2026-10-19
trust unverified — no one has confirmed this page's content
sources 4 — topping-1998, nicol-2014, lu-2012, luxton-reilly-2009

Frontmatter is the page's machine-readable record — see the Schema & Guide.