A Secondary 2 student in Clementi hands in a balanced-looking argumentative paragraph. The teacher writes, “Develop the explanation.” The student looks at the paragraph again and adds two sentences. They repeat the claim in different words. The page is longer. The explanation is not better. Neither person is being unreasonable: the teacher can recognise the missing mechanism, while the learner cannot yet see what the instruction is asking for. The interesting question is not whether the learner has been told the criteria. It is whether the learner can recognise those criteria inside an actual piece of work.
Co-constructing success criteria offers one route across that gap. The teacher does not surrender the learning goal or ask the class to vote on academic standards. Instead, learners examine examples, argue about what makes an answer successful, test tentative criteria against contrasting performances and refine language until the criteria can guide a next attempt. The classroom moves from “Here is the rubric” to “Let us find out what this rubric must be able to distinguish.” If the conversation changes the next piece of work, a standard has become usable rather than merely visible.
The 50-second answer. Co-constructed success criteria are a bounded collaboration in which a teacher and learners make the quality features of a task explicit through evidence. The teacher defines the curriculum destination and guards disciplinary accuracy. Students compare examples, propose what counts, inspect borderline cases and turn abstract words such as clear or well supported into observable decisions. The mechanism is shared interpretation: students become better able to identify the gap between current and desired work, explain that gap and act on it. It can improve assessment literacy, self-checking and feedback uptake when paired with modelling and revision time. It is not the same as writing marking schemes by popularity, and it cannot compensate for weak teaching, unavailable examples or vague subject knowledge.
This article owns the bounded process of jointly developing and testing task-specific success criteria. The existing How Feedback Works guide owns the wider feedback loop, and the How Metacognition Works guide owns the learner’s wider monitoring system. Here the narrower problem is how pupils learn to see the target itself with enough precision to make feedback useful. We will follow the mechanism through primary and secondary English, mathematics, science, parent conversations and tutorial design.
The gap between having a rubric and being able to use one
Imagine a student is given four rows: relevance, evidence, reasoning and language. Each row has a description. The student dutifully reads it, then writes a response. Asked to assess their answer, they choose “good” for each row. The words have been seen, but the student has not learnt to tell a strong explanation from a weak one. The difference is equivalent to knowing the labels on a map while being unable to navigate the streets. A rubric is a representation of quality. Assessment literacy is the ability to use that representation to make justified judgements.
Teachers often know the hidden examples behind their adjectives. Relevant evidence may mean a quotation whose exact wording can support the claim, rather than any quotation about the same topic. Accurate method in algebra may mean preserving equality at every transformation, not merely arriving at the answer. Scientific explanation may require linking a mechanism to observed evidence while ruling out a plausible alternative. Students may interpret these phrases as politeness, fluency or impressive vocabulary instead of disciplinary quality.
Co-construction works when a class puts the invisible distinctions on the table. It is not enough to ask, “What does good writing look like?” Children can produce a list of harmless abstractions: neat, creative, detailed, interesting. The decisive move is to put two pieces of work beside each other and ask, “Both look plausible. Which answers the question better? What precise feature makes that judgement defensible?” Now the criterion is being tested against performance. The learner must articulate its observable consequence.
This is especially useful when students can reproduce the wording of marking criteria but cannot diagnose their own answer. It is less urgent when criteria are already transparent and the obstacle is missing prerequisite content. A pupil who does not understand conservation of mass needs scientific teaching, not an extended democratic conversation about what counts as a good science answer. The method must be matched to the problem.
Stage one: the teacher fixes the destination without fixing every word
Begin with the learning intention. In an English lesson, perhaps the goal is to support an inference using relevant details and reasoning. In mathematics, it might be to justify that a method will work for a class of problems. In science, it could be to explain how an independent variable affects a measured outcome under controlled conditions. These goals are not negotiable simply because students have preferences. A good process protects curriculum expectations while inviting learners to interrogate what successful evidence looks like.
The teacher first selects two or three anonymous, invented samples that differ in ways worth discussing. Crucially, the samples must be safe to criticise. If a student’s real work is displayed without permission, quality dialogue can become a public ranking of people. Fictional or carefully anonymised specimens allow students to analyse the work rather than protect a friend. Include one sample that is fluent but shallow and another that is less polished but well reasoned. That contrast prevents a superficial criterion from dominating.
Before discussion, let learners make an individual judgement. “Which response is more convincing?” “Where did the writer actually support the claim?” “Would your answer change if the examiner asked explain rather than describe?” Collect differences in judgement. These disagreements are data: they show which features pupils are using, and which they are not yet able to see. The teacher can then model the disciplinary reasoning that resolves the disagreement.
A subtle design choice matters here: avoid giving a perfectly worded rubric before the exercise if the purpose is to discover its meaning. Otherwise learners may simply hunt for terms already supplied. However, do not withhold essential safety rules, scientific definitions or assessment requirements. Co-construction is a way of making criteria intellectually owned, not a guessing game in which the teacher possesses a secret password.
Stage two: compare examples that really differ, not cartoons of good and bad
Take two Primary 6 comprehension answers to a question about why a character hesitates. The first says, “He was nervous because he did not know what to do.” The second says, “He hesitated because the empty corridor reminded him that nobody would answer if he called, so entering alone felt unsafe.” The second makes a more specific inference from the situation, but what if the passage never mentions an empty corridor? Then the extra detail becomes invented evidence. Longer and more vivid is not automatically better.
Try a stronger contrast. Answer A quotes an accurate detail but does not explain its relevance. Answer B gives a plausible explanation but cites evidence from the wrong paragraph. Answer C links an accurate textual detail to an interpretation and respects the question’s scope. Ask students to rank them, justify the decision and name a criterion capable of separating C from A and B. “Uses evidence” is too blunt: both A and B can appear to use evidence. “Selects accurate evidence and explains how it supports the precise inference” is closer to a usable criterion.
Now invite a challenge case. Could a concise one-sentence answer satisfy the criterion? Yes, if it contains the needed relationship. Could a long response fail? Yes, if its elaboration is unsupported. This challenge prevents students from turning a meaningful criterion into a mechanical length rule. The criterion should survive reasonable variation in style, not prescribe a single superficial shape.
Co-construction becomes educational when every candidate criterion earns its place by distinguishing examples. Students may prefer the answer that sounds nicer; the teacher can ask for the exact evidence behind that preference. The standard is not a negotiated mood. It is a shared, evidence-tested description of quality in a particular task.
Stage three: turn adjectives into decisions a learner can make
Consider clear explanation. Clear to whom, about what, and according to which knowledge? A useful criterion should help a learner decide what to write or revise without having to ask the teacher to translate it. In science, students might refine “explains well” into “names the relevant process, links that process to the observed change, and uses the measurements without claiming more than they show.” In English, “a good topic sentence” becomes “states a focused claim that the rest of the paragraph can develop.” In mathematics, “shows working” becomes “records the transformations that preserve the relationship and makes a checking route visible.”
The distinction is between a trait and an actionable test. “Interesting” is a reaction; “introduces a contrast that the following evidence resolves” is a test. “Detailed” describes volume; “identifies the dependent variable and explains why it changes” describes a cognitive job. “Neat” describes presentation; “labels units so another person can verify quantities” links presentation to mathematical meaning. A criterion should not become so elaborate that the pupil cannot remember it, but it must be precise enough to change a decision.
Ask pupils to apply each candidate criterion to a borderline answer. If the class cannot agree whether it passes, the wording may need refinement or a supplementary example. The teacher may discover that the criterion itself combines two different constructs. For instance, “uses accurate scientific language” can hide both terminology knowledge and causal reasoning. Separating them allows a more diagnostic response: the learner knows the vocabulary but has not connected the mechanism, or the reasoning is sound but terms are inaccurate.
Write the final version in language learners can use, then connect it to formal assessment terms. Do not permanently replace subject vocabulary with vague everyday substitutes. A Primary 4 class may say “show which detail makes you think that” before learning textual evidence. The phrase is a bridge into disciplinary language, not a rival to it.
Mathematics example: a correct numerical answer can still be weak evidence
Three Secondary 1 students solve a ratio question. Given that red and blue counters are in the ratio 3:5, with 64 counters altogether, all three write 24 red and 40 blue. One student guesses from a remembered example, another writes 64 ÷ 8 = 8 and multiplies by 3 and 5, and the third explains why 3 + 5 represents eight equal parts of the same whole before calculating. Their answers coincide. Their displayed reasoning does not.
If the task is simply to report the numbers, the first answer may satisfy the immediate marking target. If the task is to explain and justify a ratio method, the evidence of understanding differs sharply. Ask students to co-develop criteria for that latter task: identifies what the ratio units represent; shows why the total is eight equal parts; maps answers back to the requested quantities; checks both sum and ratio. Now present a changed task: the difference, not the total, is 64. The standard should tell learners why blindly dividing by eight will fail.
Students can test a draft criterion such as “uses the correct formula.” Is there a single formula? A ratio bar model, unit method and algebraic approach may all be valid. A rigid criterion could reward imitation and penalise an equally sound method. Revise it to “chooses a representation that preserves the given relationship, shows how the unknown is determined, and verifies the result.” That statement accommodates more than one strategy while demanding the invariant.
This is a useful boundary: co-constructed criteria should support mathematical reasoning, not turn the classroom into a competition over which working style the teacher prefers. The teacher remains responsible for accuracy and for recognising equivalent, valid methods. Students contribute the language that helps them recognise that validity in action.
Science example: describing results is not explaining them
In a lower-secondary experiment, pupils place equal masses of sugar into water at different temperatures and time how long it takes the sugar to dissolve under otherwise comparable conditions. One report says, “The sugar dissolved faster in the warmer water.” Another says, “Warm water makes it fast.” A third describes the pattern and offers a particle-level account, while carefully noting whether stirring and particle size were controlled. These responses do not all answer the same intellectual question.
Ask groups to sort the samples into observation, comparison, explanation and overclaim. Someone may insist that a table of measurements is already an explanation. The teacher can ask, “Where does this sentence tell us why the rate might change?” Another learner may offer a plausible mechanistic account that the experiment itself did not directly measure. The teacher can distinguish background scientific theory from what the dataset alone establishes. That is not pedantry; it is evidence literacy.
Criteria might become: report the trend accurately; identify controlled conditions that matter; connect the result to the relevant scientific model; mark speculation as interpretation rather than measurement; acknowledge anomalous results where they affect the claim. Not every Primary Science question requires all five. The intended task must set the weight of each feature. A two-mark answer should not be punished for failing to resemble a full laboratory report.
Co-construction also protects against false consensus. If classmates agree that “hot water contains more sugar particles” is a good explanation, the teacher must correct the science. Collective participation is valuable, but the empirical world does not vote. The students’ contribution is to help make the quality of scientific reasoning visible within accurate knowledge.
The rubric is a reference instrument, not the finished learning
A household measuring cup carries markings, but its usefulness depends on how it is used. Someone must know what is being measured, the unit, the level, and the consequence of error. Success criteria are similar: they make qualities inspectable but do not manufacture the underlying capability. A writing checklist cannot create a persuasive idea; a mathematics rubric cannot supply a missing fraction concept; a science criterion cannot replace a valid model of particle behaviour.
Treat the jointly constructed criteria as a temporary reference instrument. Students should use them on the next attempt, not merely celebrate the poster. Provide a short revision window in which learners select one criterion, identify a precise gap, change their answer and explain why the change is an improvement. Then introduce a new task with different surface features. If the criteria matter only when the same teacher reminds them, they have not yet become independent tools.
There is also a risk of checklist worship. A learner can tick all the boxes by inserting formulaic phrases without improving thought. A criterion such as “includes a quotation” may produce an irrelevant quotation. A criterion such as “uses three connectives” may produce a paragraph with clumsy or misleading logic. The remedy is to attach criteria to meaningful performance and judge quality holistically as well as analytically. A good criterion helps the learner notice a necessary feature; it should not become a recipe for looking sophisticated.
When the task is creative or open-ended, co-construction must allow genuine alternatives. A compelling narrative opening might create suspense, establish voice or present a vivid conflict. It need not do all three in a fixed order. Criteria can state purposes and constraints while leaving room for different artistic solutions. Quality has standards without always having one shape.
A Clementi tutorial scene: three answers, one better conversation
Picture a three-student tutorial at a Clementi study table. The students have written short explanations about why a metal spoon becomes warm in a hot drink. One answer says heat rises; one says particles touch; the third says energy transfers through the metal from the warmer region towards the cooler handle. Rather than marking the first two red and praising the third immediately, the tutor asks each child to predict which explanation a careful reader could use to account for the warming handle.
The first student notices that “heat rises” cannot explain a spoon laid horizontally. The second realises that particles touching is not enough to specify what transfers or in which direction. The third discovers that the expression “warmer to cooler” needs a reference to energy transfer, not simply temperature labels. Together they propose two small checks: identify the direction of energy transfer; show how conduction in the solid links the heated end to the handle. The tutor formalises conduction and clarifies that this example does not require the class to invent particle physics beyond its syllabus.
Now change the situation: why does a saucepan handle with an insulating grip stay cooler than an exposed metal handle under comparable conditions? If the criteria still help, the learners can explain differences in energy transfer without memorising a single spoon answer. The method has succeeded because it makes a quality judgement transferable. The tutor did not ask three children to design a national assessment scheme. They participated in understanding what the explanation must accomplish.
A parent listening to a summary might ask, “Why not just give them the answer?” Sometimes direct explanation is absolutely necessary. Yet here the learners already possessed pieces of the science; the missing capability was recognising which pieces combined into a causal explanation. The criteria conversation made that gap visible and gave them a check to use independently.
The evidence: strong foundations, narrower claims
The Education Endowment Foundation’s feedback guidance, published in June 2021, stresses high-quality instruction, formative assessment, clear learning intentions and opportunities for pupils to use information about their performance. The EEF’s feedback toolkit synthesises a substantial body of studies and finds positive average effects for feedback, but its estimates concern varied feedback interventions. They do not supply a causal effect size for the narrow routine of students co-constructing task criteria.
The What Works Clearinghouse guide on secondary writing, released in 2016 and revised in 2019, recommends explicit writing strategy instruction through modelling, practice and reflection, as well as assessments used to inform feedback. This supports the broader design: make quality observable, provide guided practice and ensure the learner gets another attempt. It does not establish that every participatory rubric exercise will raise scores. Evidence for formative-assessment practices is stronger as a family than evidence for every small named classroom technique.
Research on feedback also shows why implementation deserves attention. Information is useful only when learners can interpret it, have enough knowledge to act, and receive a chance to revise or reattempt. Student voice may increase buy-in in some settings, but enthusiasm is not a reliable substitute for improved work. A rigorous evaluation would compare not only student enjoyment, but also blind-rated quality on new tasks, accuracy of self-assessment and transfer after a delay.
Do not promise that co-construction saves time every lesson. At the beginning it may take longer than distributing a sheet. Its potential efficiency appears later if learners can diagnose recurrent weaknesses without needing the teacher to translate the same adjective repeatedly. That is a plausible mechanism to test locally, not a universal guarantee.
Failure mode: the loudest students define what everybody must do
A classroom vote on criteria may look participatory while privileging the confident. A pupil who speaks English fluently can propose polished phrases that quieter classmates cannot yet use. Another may argue that a criterion is unnecessary because they personally find it easy. If the teacher adopts each suggestion by applause, the final product can reflect social hierarchy rather than assessment validity. Protect individual thinking before discussion, collect anonymous suggestions where useful, and check every criterion against actual examples. Invite a student who has not spoken to explain whether the proposed wording would help them revise a draft.
There is a deeper equity question. Some students know the hidden conventions of schooling because their families have discussed model answers for years. Others are encountering the language of criteria for the first time. Co-construction should reduce that hidden curriculum, not reward those who already know it. Provide understandable examples, define technical terms and ensure the teacher names important standards even when the class does not spontaneously identify them. A student should not need to win a debate to receive access to what good work requires.
If multilingual learners offer a clear observation in informal language, preserve its conceptual contribution and teach the disciplinary equivalent. “It tells you the reason, not just what happened” can be refined into “the response connects observed evidence to a causal mechanism.” Students should be able to see the intellectual continuity between their words and the formal version. That is a more empowering outcome than presenting polished teacher language as though it emerged from everybody simultaneously.
Failure mode: the checklist becomes a second assignment
Co-construction can exhaust the available lesson if every task requires a fresh poster, a debate over wording and a four-level marking grid. Teachers have limited time and students have limited working attention. The remedy is selective use. Choose criteria for tasks where quality is genuinely misunderstood and where the resulting discrimination will recur. A single fifteen-minute comparison around “evidence versus explanation” may serve several weeks of writing better than new criteria for every worksheet. A short oral check may suffice for routine arithmetic accuracy.
Limit the initial criteria to the few properties that matter most. For a Primary 5 science explanation, perhaps name the relevant process, connect it to the observation and use accurate terms. For a Secondary 3 essay paragraph, perhaps answer the precise question, select persuasive evidence and explain how the evidence supports the claim. More dimensions can be added as students develop. A rubric with sixteen rows may be technically precise but cognitively unusable during writing.
Avoid treating the jointly written criteria as immutable. Put a date on a class version and revise it after a new challenge case shows that it is ambiguous. If students discover that a concise answer can satisfy all three criteria, remove the unnecessary word-count expectation. If a creative response succeeds through an unexpected structure, distinguish structural conventions that are essential from preferences that were merely familiar. The point is improved judgement, not permanent paperwork.
Learner route: from reading a criterion to testing your own work
Before beginning an answer, choose one criterion that you frequently misread. Translate it into a question with an observable answer. “Explain clearly” might become, “Have I named the process and shown how it causes the outcome?” “Use evidence” might become, “Which exact line or value supports my claim, and why?” Write the question beside the task, then hide the exemplars and complete the work. This protects the attempt from becoming an exercise in copying the model.
After writing, read once as the marker might, without fixing anything. Put a small mark where each criterion is visibly satisfied. Leave gaps where the claim depends on your intention rather than words on the page. Now choose one high-value revision. Change the actual answer, not merely the checklist. If your paragraph now links an observation to a mechanism, underline the new sentence and explain why it is better. Do not mark your own work perfect just because you have invested effort in it.
A day or a week later, repeat the judgement on an unfamiliar piece of work. If you can discriminate better and revise independently, you are learning the standard rather than memorising the original example. If you still need the teacher to tell you where evidence ends and explanation begins, ask for one more contrasting pair and make the separating feature explicit. The problem is specific enough to practise.
Parent route: three questions that help without turning home into an examiner’s office
Parents often receive vague instructions: “Your child needs to be more analytical,” “work on presentation,” or “develop ideas.” Instead of demanding another hour of undirected practice, ask the child to show two answers and explain which better meets the task. Follow with: “What exactly makes that answer stronger?” and “What one change would improve yours?” These questions make quality visible without requiring the parent to invent a substitute marking scheme.
If the child says, “The teacher wants more detail,” ask what kind. Does the question need another example, a link in reasoning, a definition, a unit, a qualification or a check? A child who adds adjectives may not be addressing the problem. It is appropriate for a parent to say, “I cannot judge the Science itself; can we use the teacher’s model or the textbook to check?” Honest limits are better than confident but inaccurate feedback.
Avoid ranking siblings or turning self-assessment into a character verdict. “You are careless” tells a child little about the work. “This explanation states the result but does not link it to the process” identifies a repairable difference. If anxiety rises, shorten the conversation and return to one small task. A successful home routine leaves the learner more able to check alone, not more dependent on a parent hovering beside every sentence.
Teacher and tutor route: design the next twenty minutes
First, choose a recent assignment where the same feedback phrase appears repeatedly. Second, find or write three brief response samples that differ on that exact criterion. Third, plan one question that students can answer privately before seeing the formal rubric. Fourth, use disagreement to surface the hidden quality distinction. Fifth, guide learners toward concise, technically accurate wording. Sixth, ask them to test the criterion on a sample not previously discussed. Seventh, return to their own work for an immediate revision.
The seventh step is often missing. A lively discussion can end with the learner having acquired vocabulary about quality but no changed performance. Reserve at least as much time for use as for discussion. If the lesson is short, cut the number of samples before cutting the independent revision. For a three-student tutorial, ask each learner to defend a different judgement, then rotate. Be alert to the possibility that the seemingly strongest student is overconfident about why an answer is successful.
Record one simple observation for the next lesson: could students name the relevant criterion without prompts, locate where a sample failed it, and improve a fresh response? These are better early measures than how attractive the final criteria poster appears. A delayed task is the stronger test. A learner who understands criteria in a discussion but cannot use them during an examination needs rehearsal under more realistic conditions.
A one-week sequence that separates enthusiasm from transfer
On Monday, begin with individual rankings of three anonymous responses and collect reasons. On Tuesday, teach the missing content and revise the criteria using a deliberately awkward borderline example. On Wednesday, give students a short independent task and ask them to annotate the point where their work meets the most important criterion. On Thursday, return one piece of focused feedback and allow a real revision, ideally without the teacher dictating the sentence. On Friday, offer a new prompt and ask students to retrieve the criteria from memory before they start.
That sequence is not a validated package with a universal effect size. It is a practical design for asking whether shared criteria can survive beyond the original comparison. Teachers can shorten or stretch it. A Primary 3 class might use oral examples and two criteria. A Secondary 4 class might test evidence, counterargument and qualification across two unseen essay prompts. In both cases, the point is the same: learners should use quality language to direct decisions.
Look for a surprisingly powerful signal: fewer requests for “What do you want us to write?” alongside more precise questions such as, “Does this measurement actually support that claim?” The latter question shows that students have begun interrogating evidence. It does not imply they no longer need teachers. It means teacher support can move to the next unresolved level.
Frequently asked questions
Is co-constructing criteria the same as letting pupils mark themselves?
No. Co-construction is about clarifying and testing what quality means. Self-assessment is one possible later use. Learners can participate in developing criteria and still receive teacher marking; they can self-assess with a teacher-provided rubric without having co-constructed it. The mechanism here is making standards interpretable through evidence and discussion.
What if students suggest criteria that are wrong?
Treat the suggestion as a hypothesis to test, not a binding vote. Compare it with credible samples, curricular requirements and subject knowledge. Explain why a proposed rule fails. In science, claims must respect evidence; in mathematics, criteria must accept valid equivalent methods; in writing, criteria should distinguish effectiveness without demanding one formulaic voice.
Can this help with PSLE or secondary examination answers?
It can support examination preparation if the criteria are faithful to the assessed skills and if learners apply them to unseen questions. Do not invent official marking rubrics or pretend a classroom poster reproduces every examiner decision. Use publicly available syllabus expectations and school feedback where appropriate, and practise transfer rather than memorising a sample answer.
What is the difference between a success criterion and a learning intention?
A learning intention identifies what is being learned, such as explaining a scientific mechanism. A success criterion identifies features that would make that learning visible in performance, such as a correct causal link supported by the relevant observation. Students need both, but confusing them can produce goals too vague to guide revision.
Should criteria be written in student language or formal subject language?
Both have a role. Begin with understandable wording, then explicitly connect it to precise disciplinary terms. A criterion should become more exact as knowledge grows, not remain permanently informal. Check that learners can explain the phrase using a real answer.
How many criteria are enough?
As few as can discriminate the central quality problem. Two or three often make a useful starting point for younger learners or short tasks. Complex projects may need more, but students should know which are priorities and how to act on them. More boxes do not automatically produce more learning.
Does this replace modelling and explicit instruction?
No. The teacher should model the relevant knowledge and reasoning, correct misconceptions and demonstrate good work. Co-construction helps the learner recognise and use those standards. Without instruction, a class may become skilled at discussing a rubric while remaining unable to produce the underlying performance.
How can we tell whether the method worked?
Ask students to judge unfamiliar samples, justify differences and revise a fresh answer. Repeat after a delay and compare with earlier performance. Engagement and confidence are useful observations but not adequate evidence by themselves. Improvement should appear in the quality of work and the independence of judgement.
Sources, evidence boundaries and useful continuation
The Education Endowment Foundation guidance on teacher feedback sets out why learning goals, good teaching, formative evidence and opportunities to use feedback matter. Its feedback toolkit summarises an extensive but diverse evidence base; the positive overall synthesis must not be mislabelled as an effect estimate for the specific co-construction routine. The US Institute of Education Sciences’ secondary writing guide supports modelling, guided practice, reflection and informed feedback. The UK Department for Education’s early-career teacher evidence guide provides accessible guidance on clear, specific, actionable feedback.
Continue to How Feedback Works for the whole correction loop, How Metacognition Works for self-monitoring, the Education Hub for parent and teaching routes, and the existing How X Works Hub for the wider mechanism library. These are connections, not new ownership claims. No existing route needs to move.
The return question is simple: after a student has helped sharpen the criteria, can they see what their own work needs next? A poster full of admirable adjectives is not enough. A learner who changes one sentence, one equation or one explanation for a reason they can defend is getting closer. Co-constructed success criteria work when the class stops merely possessing standards and starts being able to use them.