How Problem Representation Fails | Why Students Solve the Wrong Problem Correctly

Problem representation is the hidden layer between reading an exam question and solving it. A student can know the mathematics, science, vocabulary, formulas, concepts and examination techniques, yet still solve the wrong problem correctly because the question has been represented incorrectly before the first line of working begins.

This is why some examination mistakes survive more revision. The learner is not always missing the technique. Sometimes the technique is executed perfectly on the wrong target, the wrong quantity, the wrong relationship, the wrong source boundary or the wrong set of constraints. The visible answer looks like a calculation error, comprehension error, science error or essay error. The upstream failure happened earlier: the mind built the wrong internal version of the task.

For students searching for how to understand exam questions, how to solve word problems, why students choose the wrong method, why correct working gives the wrong answer, how to interpret maths questions, how to read science questions, how to answer comprehension questions and how to improve problem solving, the central idea is simple: before a learner can choose the right method, they must construct the right problem.

Alicia, Tricia and Kai Kai return as the resident learners in this eduKateSG examination-performance series. Alicia is often technically accurate once she has chosen a route. Tricia is careful with evidence but can overcomplicate a representation. Kai Kai learns procedures quickly and therefore sometimes recognises a familiar surface before checking whether the new problem has the same structure. The examples involving them are illustrative teaching cases, not reports of identifiable pupils or claims of measured tuition outcomes.

This article is deliberately narrower than How Method Selection Fails. Method selection asks which approach should be used once the task has been understood. Problem representation asks what the task is in the first place. It is also separate from How Misconception Repair Works, which examines wrong underlying models, and from the newer integrated-question and multi-step failure pages, which own failures after several pieces of knowledge must be coordinated. Here the owner is the representation layer itself: how wording, quantities, diagrams, evidence, variables, constraints and requested outputs become the learner’s internal problem.

1. The student can solve the wrong problem perfectly

Suppose a question says that a jacket costs 96 dollars after a twenty per cent discount and asks for the original price. Alicia multiplies 96 by 1.20 and obtains 115.20. Every calculator key is pressed correctly. The multiplication is accurate. The answer is still wrong because the twenty per cent discount was applied to the original price, not to the discounted price. The learner has represented 96 as a base to which twenty per cent should be added, when 96 is actually eighty per cent of the original base.

The correct representation is not “add back twenty per cent.” It is “the final amount equals eighty per cent of the original amount.” If P is the original price, 0.80P = 96, so P = 120. Once the relationship is represented correctly, the arithmetic is routine.

This example matters because it changes the repair. If we call the error “weak percentages,” the learner may receive more forward percentage questions. Those may become faster while the reverse relationship remains confused. If we call it “careless,” we add no mechanism at all. The useful diagnosis is that the role of the known quantity was represented incorrectly.

Now notice how easily the same pattern travels. In a speed problem, a learner may treat the final speed as the starting speed. In a science graph, the learner may treat the dependent variable as the cause because it is plotted vertically. In a comprehension passage, a learner may treat the writer’s example as the writer’s main claim. In an essay, the learner may treat the topic as the task and write everything known about the topic while ignoring the command. The surface subject changes. The representation error is structurally similar.

Problem representation therefore deserves its own place in examination training. It is not simply “reading carefully.” Reading is one input. Representation is the constructed result.

2. A useful pipeline: words → task model → structure → method → execution → verification

A compact way to locate failure is to separate six stages.

1. Words and symbols: what the question literally presents.
2. Task model: what the learner thinks the question is asking.
3. Structure: the relationships, constraints, variables, evidence or logical form inside that task.
4. Method: the procedure or reasoning route selected to work with that structure.
5. Execution: the calculations, sentences, transformations or steps.
6. Verification: whether the result answers the original task and respects its conditions.

The stages interact. A method can change a representation, and verification can expose an earlier misunderstanding. But keeping the stages separate long enough for diagnosis prevents a common teaching error: repairing the first visible wrong line even when the wrong line was downstream of a mistaken task model.

Kai Kai provides a clear example. She is asked to compare two sources. She summarises Source A accurately and then summarises Source B accurately. Her writing is grammatical and her factual extraction is strong. Yet she has not actually compared them. The execution is competent inside the wrong task model: “tell me what each source says.” The requested task was “identify relationships between the sources along a shared dimension.” The repair must therefore begin before sentence polishing.

Tricia offers another example. A probability question describes drawing objects without replacement. She correctly remembers that multi-stage probabilities often involve multiplication. But she represents the second draw as occurring in the same sample space as the first, so she keeps the original denominator. Her multiplication is correct relative to the wrong state representation. The denominator error is not merely a fraction slip. She has not updated the world after the first draw.

The pipeline also explains why checking only the final answer can miss the source of failure. A plausible number can survive a wrong representation. Two wrong steps can even cancel. A polished paragraph can remain irrelevant. Verification should eventually compare the result back to the original words and constraints, not only inspect local arithmetic or grammar.

3. What research says about representation

Problem representation has a long history in research on expertise and problem solving. In the classic work of Chi, Feltovich and Glaser, expert and novice physics solvers categorised problems differently: experts tended to organise them around underlying principles, while novices relied more strongly on literal surface features. The point is not that experts ignore surfaces or that novices cannot think structurally. The useful implication is that what a learner notices first can influence which knowledge becomes available for solving.

Alibali, Phillips and Fischer later reported that learning a new problem-solving strategy could change children’s representations of mathematical-equivalence problems, with effects differing by strategy. This is important because representation is not a fixed pre-solution photograph. Learning new ways to solve can change what the learner sees in the problem, and what the learner sees can in turn change which method becomes available.

Research on worked examples adds another warning. Worked examples can reduce unnecessary search and support learning, but learners may process their surface details rather than the underlying structure. Work on subgoal-labelled examples and self-explanation attempts to direct attention toward the functional structure of a solution rather than only its local steps. That research does not imply one universal exemplar format. It does support the broader teaching question: what is the learner being encouraged to encode?

The Institute of Education Sciences has also supported research connecting algebra performance with problem representation and feature knowledge. The general lesson for this article is modest: conceptual features and how problems are encoded can matter to later equation solving; practice volume alone does not guarantee that the learner encodes the right structure.

For reality-based mathematics tasks, research on mathematical modelling similarly highlights stages such as understanding the situation, simplifying or structuring it, mathematising, working mathematically, interpreting and validating. A learner can therefore fail before computation begins, while translating between a described situation and a mathematical model.

This article uses those research traditions as a foundation while keeping the teaching tools below explicitly practical. The “representation audit,” “problem fingerprint” and diagnostic routines in this guide are proposed instructional devices rather than validated clinical instruments. They should be judged by whether they expose useful distinctions and improve later independent performance.

4. Representation is a compression decision

An examination question contains more information than the learner can keep equally active. Representation is partly a compression process: decide what matters, what relates to what, what can be temporarily ignored and what must remain protected.

This means a representation can fail in at least two opposite ways. It can discard a decisive condition. Or it can preserve so much irrelevant surface detail that the important relation becomes difficult to see.

Imagine a word problem describing a school charity event, three prices, two discounts, a transport fee and a final budget. One number is decorative context. Another is essential to a ratio. A learner who translates every noun and number into an equation can become overwhelmed. A learner who uses a keyword rule and ignores a small phrase such as “including the fixed fee” can build the wrong equation quickly. Better representation is not maximal capture. It is selective fidelity.

In a comprehension question, the same principle applies. A passage may contain descriptions, examples, quotations, counterarguments and the writer’s main claim. A summary task requires a subset. An evaluation task requires a different subset plus criteria. An inference question may depend on two small textual cues that are not the most memorable sentences. Representation decides which information enters the active problem.

In science, an experimental description may contain measured outcomes, controlled variables, uncontrolled variables, instrument limits and proposed explanations. Treating all details as equal can produce a long answer with no causal structure. Ignoring one uncontrolled variable can produce a confident causal claim the design cannot support.

A strong representation therefore preserves the task’s invariants and constraints while compressing irrelevant detail. That is a difficult skill because “irrelevant” is not a property of a fact alone. It is relative to the question being asked.

5. The ten-part problem fingerprint

Before discussing failure modes, it is useful to define a compact representation tool. The problem fingerprint asks ten questions. Learners do not need to answer all ten aloud during an examination. In teaching, however, the list helps reveal where a representation diverges.

1. Target: What must be produced?
2. Knowns: What information is explicitly given?
3. Unknowns: What must be found, decided, proved or explained?
4. Roles: What does each important quantity, statement or source component represent?
5. Relationships: How are the important elements connected?
6. Constraints: What limits values, interpretations, methods or evidence?
7. State: What can change as the problem unfolds?
8. Allowed evidence: What information may legitimately support the answer?
9. Output form: Exact value, estimate, explanation, comparison, proof, recommendation, summary or something else?
10. Sanity checks: What would make a proposed answer obviously inconsistent with the original task?

A representation does not need a separate written box for each component. The fingerprint is a diagnostic lens. If a learner’s answer is wrong, ask which fingerprint element was misbuilt. If the learner repeatedly fails the same element across different topics, the pattern may justify targeted practice.

Alicia’s reverse-percentage error is a role failure: she misidentified what 96 represented. Tricia’s without-replacement probability error is a state failure: she did not update the sample space. Kai Kai’s source-comparison response is a target and output-form failure: she represented comparison as two summaries.

The fingerprint also protects against overdiagnosis. If the learner identifies the target and constraints correctly but cannot execute the required algebra, the representation may be adequate. Do not keep teaching interpretation when the bottleneck has moved.

6. Failure Mode 1: the topic replaces the task

The learner sees “photosynthesis,” “quadratic,” “Macbeth,” “percentage” or “climate change” and begins retrieving everything associated with the topic. Topic recognition is useful for orienting memory, but an examination question is rarely asking for the entire topic.

Suppose an English essay asks whether a particular policy is likely to reduce inequality. The learner has memorised an excellent essay about inequality and reproduces its causes. The writing is relevant to the topic but may not evaluate the policy’s mechanism, conditions or likely effects. The internal task became “write about inequality.”

A repair begins with a target sentence: “This question requires me to judge whether this policy changes inequality, not to explain inequality in general.” Then list what evidence would bear on that judgement. Only after the target is stable should topic knowledge be retrieved in depth.

In mathematics, the topic trap appears when a learner sees trigonometry and immediately searches for a trigonometric formula. The actual problem may ask for a geometric proof where a different relation is simpler. In science, seeing “enzyme” can trigger a memorised temperature explanation even when the graph is about substrate concentration.

The training contrast is to keep the topic constant while changing the task. Use four questions on the same content: define, explain, compare and evaluate. If the learner produces the same response architecture every time, the topic is controlling representation too strongly.

7. Failure Mode 2: a keyword replaces a relationship

Keyword strategies are attractive because they compress interpretation. “Altogether” means add. “Difference” means subtract. “Average” means add and divide. “Explain” means give a reason. Sometimes these shortcuts work. The danger is treating them as deterministic.

Consider “twenty per cent less than.” The relevant operation depends on which quantity is the base. Consider “average speed.” The arithmetic mean of speeds is not always the journey’s overall average speed. Consider “because.” Its presence in a passage does not guarantee that the clause gives the causal explanation the question asks for.

A stronger keyword routine asks what relationship the word is signalling. “Difference” may indicate subtraction, but which quantity from which? “Average” prompts the question: average of what units, weighted by what? “Explain” prompts: what mechanism, evidence or connection must be made visible?

Training should therefore include near pairs where the same keyword appears but the relationship differs. The purpose is not to make students distrust language. It is to make language the beginning of interpretation rather than the end.

8. Failure Mode 3: the first number becomes the base

In quantitative tasks, the first prominent number can become psychologically privileged. Learners use it as the denominator, starting amount or reference point before checking what it represents.

This explains many percentage, ratio and rate errors. A problem may state a final amount before the original. It may state a subgroup before the total. It may give one group’s mean before its size. The order of presentation is not the mathematical hierarchy.

A representation repair labels roles before operations: original, final, part, whole, elapsed time, distance, group size, rate. Units help, but role labels add another layer. Two quantities can share units and still play different roles.

Do not require verbose labelling forever. During diagnosis, ask the learner to annotate only the ambiguous quantities. When the distinction stabilises, the annotation can disappear.

9. Failure Mode 4: the requested unknown is replaced by an easier unknown

Students sometimes solve for something that is easier to compute and then stop. A question asks for the original quantity; the student computes the amount changed. A geometry problem asks for an angle; the learner finds a length and leaves it. A passage asks for the writer’s purpose; the learner identifies the topic.

The intermediate result may be correct and useful. The failure is closure: the learner’s internal target changed during execution.

A simple verification question catches this: “What noun phrase completes the sentence ‘I have found…’?” If the answer does not match the noun phrase in the original request, more work is required.

This is especially valuable in multi-step tasks where several subgoals are legitimate. The learner should know whether the current result is a destination or a waypoint.

10. Failure Mode 5: the learner solves the diagram instead of the stated geometry

Diagrams are powerful representations because they make relations visible. They are also dangerous when appearance supplies information that has not been stated or derived.

A triangle looks right-angled. Two lines look parallel. A point looks like a midpoint. A graph looks linear. If the question or established reasoning does not provide that relation, using it can create a visually plausible but invalid route.

The repair is not to distrust diagrams. It is to distinguish encoded information from apparent information. Marks, labels, coordinates, stated properties and derived relationships are evidence. Visual impression is a cue to investigate.

A training exercise can show the same diagram twice, once with a right-angle marker and once without. Ask which theorem becomes licensed by the marker. This teaches that the visual shape is not the condition; the established property is.

11. Failure Mode 6: the graph shape is represented without its axes

A rising straight line can mean constant speed on a distance-time graph, constant acceleration on a velocity-time graph, proportional growth when it passes through the origin, or something entirely different depending on the variables and scales.

Students sometimes recognise the shape before reading the axes. The visual pattern triggers a memorised interpretation. This is efficient only when the axis roles match the stored pattern.

A robust graph representation names the variables and units first. Then it asks what the gradient, area, intercept or curvature would mean for those variables. Dimensional reasoning can reject some wrong interpretations cheaply.

For example, area under a velocity-time graph has units of distance. Area under a distance-time graph has units of distance multiplied by time, so it cannot represent distance travelled in the same way. Units do not solve every graph question, but they can expose a representation mismatch.

12. Failure Mode 7: the denominator disappears from the story

Ratios, rates, probabilities, means and proportions depend on denominators. Learners often focus on the numerator because it carries the visible event: successes, red counters, books read, kilometres travelled.

The denominator defines the reference set. “Three red” means little for probability without knowing how many possible items are in the relevant draw state. “Ten cases” means little for a rate without exposure, time or population. “An average of eight” is incomplete without knowing the unit being averaged.

Representation training should therefore ask, “Out of what?” before calculation. In probability: out of which current sample space? In percentages: out of which base? In a rate: per what time or quantity? In a mean: which observations receive weight?

This single question repairs many superficially different errors because it reconstructs the reference class.

13. Failure Mode 8: state changes are not represented

Some problems describe a system that changes after each action. Without-replacement probability is a simple example. Inventory, population, account balances, chemical conditions and algorithmic states can also change.

Tricia draws one counter from a bag and mentally leaves the bag unchanged. She then applies the original denominator again. The arithmetic is internally consistent with a static world. The problem describes a dynamic one.

A useful representation explicitly marks the state transition: before → action → after. What changed? What stayed invariant? Which quantities must be recalculated?

Do not apply this machinery to every trivial task. Use it when an action changes the conditions for the next step. The representation should make the dynamic feature visible, not add unnecessary notation.

14. Failure Mode 9: the learner assumes every number must be used

Word problems and data questions can include redundant information. Students trained on tightly constructed exercises may develop an expectation that every number has a computational role.

When an irrelevant number appears, the learner forces it into the calculation. The resulting answer can look sophisticated because more information has been “used.” In reality the representation failed to filter.

Training should include some questions with extra information and require a justification for exclusion: “This number does not affect the requested quantity under the stated relationship.” The aim is not suspicion. It is relevance control.

The converse problem also matters: dismissing a small number because it looks incidental when it defines a constraint or fixed cost. Relevance is determined by structure, not visual prominence.

15. Failure Mode 10: missing information is silently invented

Learners often prefer a complete problem to an incomplete one. When information is missing, they may unconsciously assume a value, linearity, independence, equal probability, equal group size or a familiar causal direction.

Sometimes such assumptions are intended and conventional. Sometimes the task explicitly invites modelling assumptions. But an unstated assumption should not be mistaken for given evidence.

A strong response can be conditional: “If the relationship is linear, then…” or “This cannot be uniquely determined without knowing…” Identifying underdetermination is not refusing to solve. It is an accurate representation of what follows from the available information.

Training should include cases where the correct answer is “not enough information” and cases where a conventional assumption is explicitly stated. Students need the boundary, not a blanket rule to distrust every problem.

16. Failure Mode 11: the learner treats a picture as a complete model

Pictures simplify reality. That is their strength and their danger. A bar model, particle diagram, circuit sketch, map, timeline or force diagram highlights selected relations and omits others. Students can forget that the picture is a representation with conventions rather than the object itself.

A bar model may preserve relative quantities but not physical shape. A particle diagram may show count and spacing symbolically rather than to scale. A circuit diagram shows connectivity, not literal wire placement. A map compresses distance and direction according to a scale and projection. If the learner imports physical properties from the drawing that the representation was never meant to encode, the model becomes misleading.

The repair is a representation legend: what information does this format preserve, what does it suppress, and which visual features are conventional? This should be brief and domain-specific. The aim is not to make pupils suspicious of diagrams but to teach them what each diagram is authorised to mean.

17. Failure Mode 12: the learner represents a causal problem as a correlation problem

A graph shows that two variables move together. The learner immediately represents the task as “explain why A causes B.” The original evidence may support only association, or the direction of influence may be uncertain.

The representation error occurs before the explanation. Once the learner assumes a causal arrow, every later sentence is recruited to defend it. More scientific vocabulary can make the overclaim sound stronger without improving its evidential basis.

A representation-first routine separates observation from explanation: “What is measured?” “What relationship is observed?” “What causal claim, if any, is the question asking me to evaluate?” “What alternative causes or design features matter?”

This is not a demand for sophisticated causal inference terminology in every school answer. It is a discipline of keeping the evidence state separate from the explanatory hypothesis until the task permits the connection.

18. Failure Mode 13: the learner treats the writer’s example as the writer’s claim

In reading comprehension, vivid examples are memorable. A reader can therefore build the passage representation around the example rather than around the proposition the example is intended to support.

Suppose a writer argues that public transport reliability affects access to work and then gives one commuter’s story. A learner summarising the passage writes about the commuter’s missed interview. The detail is accurate but narrower than the writer’s claim. The representation has elevated evidence into thesis.

A useful annotation distinguishes claim, reason, example, evidence, counterargument and conclusion. Students do not need to label every sentence forever. During training, however, these roles prevent a memorable anecdote from hijacking the argument model.

19. Failure Mode 14: the learner treats the command word as the whole question

“Compare,” “evaluate,” “explain” and “justify” help specify response functions, but they do not determine content by themselves. A student can know that evaluate requires judgement and still fail because the criteria, evidence or object of evaluation are misrepresented.

For instance, “Evaluate the claim that the experiment proves X” requires judgement about evidential support. It does not invite a general essay on whether X is scientifically plausible. The command word and the object together define the job.

The repair is to paraphrase the complete request, not merely the command: “I must judge whether this experiment’s evidence supports this causal claim.” That sentence is much more specific than “evaluate means give both sides.”

20. Failure Mode 15: the learner drops small constraint words

Words such as exactly, at least, at most, only, except, not, integer, positive, using the passage, to the nearest, and for all can change an entire solution.

Because these words are short, they receive less visual weight than nouns, formulas and numbers. A student reads the “main” content and represents the constraint as secondary decoration. The resulting method may be perfectly valid for a nearby but different problem.

One practical training method highlights constraint words only after the learner has attempted the problem. Ask which word would have changed the answer if removed. This encourages sensitivity without teaching pupils to colour-code every question indefinitely.

21. Failure Mode 16: the learner ignores units while building the model

Units are not merely labels attached at the end. They often reveal what a quantity means. Dollars per hour is not dollars. Metres per second is not metres. Grams per cubic centimetre is not grams.

When units are ignored during representation, students can add incompatible quantities, use the wrong denominator, or misinterpret a graph operation. Adding five metres to three seconds is not merely an arithmetic impossibility; the quantities belong to different dimensions.

Dimensional analysis is especially useful as a representation check because it can expose a wrong operation before the final answer. If multiplying two quantities produces units that cannot match the requested output, the route deserves inspection.

Units are not infallible. A dimensionally consistent expression can still be conceptually wrong. They are a filter, not a proof.

22. Failure Mode 17: the learner confuses rate with amount

A rate tells how one quantity changes relative to another. An amount tells how much. Students often switch between them because the same words or numbers appear nearby.

A tap flows at four litres per minute for six minutes. The rate is four litres per minute; the amount is twenty-four litres. If the question asks how long it takes to collect twenty litres, dividing amount by rate is relevant. If it asks the flow rate after a change in pressure, the problem is different.

In graphs, slopes often represent rates while coordinates represent amounts or states. A student who reports a y-value when the question asks for a rate has selected the wrong object from the representation.

Training should alternate “how much?” and “how fast?” questions on the same context. The learner must state which quantity is sought before choosing the operation.

23. Failure Mode 18: the learner confuses stock with flow

Stock-flow confusion is a more advanced version of amount-rate confusion. A stock is a quantity at a point in time; a flow is a quantity per unit time entering or leaving.

A bank balance is a stock. Income per month is a flow. The number of people currently in a room is a stock. Arrivals per minute are a flow. Water volume in a tank is a stock; inflow and outflow rates are flows.

A student may represent “increasing inflow” as “increasing stock” even when outflow is larger. Or interpret a falling growth rate as a falling total when the total may still be increasing more slowly.

Representation-first teaching asks what is accumulated and what changes the accumulation. This distinction supports mathematics, science, economics and data interpretation without requiring advanced systems terminology.

24. Failure Mode 19: the learner assumes a linear relationship because two points can be joined

Any two distinct points determine a straight line. That does not mean the underlying relationship is linear. If the problem states or justifies linearity, interpolation may be appropriate. If not, the line is one possible model among many.

Students often represent “points on a graph” as “a line to extend” because school exercises frequently use linear relationships. A third point can sometimes expose the error. So can a context where theory predicts curvature.

The repair is to ask what licences the model family. Is linearity stated? Is it an assumption to be evaluated? Is it only a local approximation? Representation includes the form of the relationship, not just its parameters.

25. Failure Mode 20: the learner assumes independence because probabilities are multiplied

Multiplication does not by itself prove independence. The probability of two events occurring together can be expressed through a conditional relationship: P(A and B) = P(A)P(B given A). Independence is the special case where P(B given A) equals P(B).

Students may remember “and means multiply” and reuse the original probability for the second stage even when the first event changes the second. The representation has erased dependence.

At school level, the repair can remain concrete: after the first event, did the situation change? If yes, what is the new relevant probability? This question reconstructs dependence without requiring notation beyond the learner’s syllabus.

26. Failure Mode 21: the learner treats an estimate as an exact quantity

Measurements, graph readings and rounded data have limited precision. A student can build a representation that treats them as exact and then report an answer with unjustified precision.

For example, a length reported as 12.4 cm to the nearest tenth does not establish that the true length is exactly 12.4. In a school task, the required treatment depends on the syllabus and question. The broader representation principle is that the information’s precision is part of its meaning.

The opposite error also occurs: refusing to calculate because data are approximate when the task clearly expects an estimate. Representation should preserve the requested level of certainty rather than demanding impossible exactness.

27. Failure Mode 22: the learner mistakes a bound for a measured value

A statement such as x ≤ 5 does not mean x = 5. “At least three” does not mean “exactly three.” A limit, threshold or allowable range defines a set of possibilities.

Students sometimes collapse a range to its boundary because the boundary is the most visible number. This can break optimisation, inequalities, probability and data interpretation.

A representation repair draws the allowable set explicitly: interval, number line, cases or verbal range. Once the space is visible, the method can reason over it.

28. Failure Mode 23: the learner treats a representative example as an exhaustive rule

One worked example can illustrate a method without defining all cases where it applies. If the learner represents the example as the rule itself, surface features become compulsory.

A student sees a percentage increase followed by decrease and memorises the exact sequence. Another sees a proof with an even number and assumes the same decomposition is required for every parity proof. Another sees one high-quality paragraph with two quotations and assumes every paragraph must contain exactly two.

The repair extracts the invariant: what property of the example mattered? Which features were incidental? This connects directly with concept boundaries and generalisation, but here the emphasis is on the internal problem model: the learner must recognise which parts of the new problem correspond to the invariant.

29. Failure Mode 24: the learner overgeneralises a representation from the previous question

Sequential papers create local priming. If Question 7 used a particular representation, the learner may carry it into Question 8 without rebuilding the new task.

This is often efficient because examinations group related items. But it becomes risky when the second item deliberately changes a condition. A previous graph may use distance against time; the next uses velocity against time. A previous probability problem used replacement; the next does not.

A compact reset at question boundaries can help: “new target, new conditions.” The learner does not need to forget previous work. They need to confirm which representation survives.

30. Failure Mode 25: the learner represents difficulty as evidence of a trick

When a familiar method becomes awkward, students sometimes assume the examiner has hidden a trick. They begin searching for a clever shortcut instead of questioning the representation.

Difficulty can indeed signal that a different method or representation is useful. But it can also reflect ordinary algebra, weak fluency or a genuinely demanding question. “There must be a trick” is not evidence.

A better response asks whether the current representation is producing contradictions, excessive complexity or a mismatch with the requested output. If so, rebuild the model. If not, continue a valid route without chasing novelty.

31. Failure Mode 26: the learner treats every unfamiliar context as a new concept

Transfer fails in the opposite direction when surface change is over-weighted. A ratio appears in a recipe, a map scale, a mixture and a probability. The stories feel different, so the learner searches for separate methods.

Representation training asks what relation remains invariant. Are two quantities being compared multiplicatively? Is a fixed scale connecting representations? Is a part-whole relation preserved?

This does not mean every superficially different task is secretly the same. The learner must check the conditions. The goal is structural recognition without careless overgeneralisation.

32. Failure Mode 27: the learner treats every familiar context as the same problem

Surface familiarity can be just as misleading. Two journey problems may differ in whether time or distance is fixed. Two growth problems may differ in whether change is additive or multiplicative. Two source questions may concern the same topic but ask different response functions.

A useful contrast holds the story constant and changes one structural feature. Students should articulate which part of the representation changes and which method consequence follows.

33. Failure Mode 28: the learner confuses a constraint with a suggestion

“Use information from both sources,” “without a calculator,” “give an exact answer,” “assume air resistance is negligible,” or “write for a formal audience” are not optional decorations. They modify the admissible solution space.

A correct result obtained by violating an explicit condition may not satisfy the assessment task. Representation should therefore include procedural constraints as part of the problem, not as postscript.

At the same time, teachers should distinguish assessment-specific requirements from universal truth. A command to use a particular method in one exercise does not make alternative methods mathematically invalid elsewhere.

34. Failure Mode 29: the learner treats all given information as equally reliable

Some tasks deliberately present claims from different sources, measurements with uncertainty, estimates, opinions, model outputs and direct observations. The learner may flatten them into one pool of facts.

Representation should preserve provenance. Who said this? Was it measured, inferred, predicted or asserted? What uncertainty or bias is relevant? In a source-based essay, evidence quality can affect the judgement. In science, an instrument reading and a student’s explanation occupy different evidential roles.

This does not require cynical distrust. It requires keeping information type attached to information content.

35. Failure Mode 30: the learner represents uncertainty as ignorance

Some questions do not support complete certainty. A learner can represent that as “I do not know” and either guess or stop. Yet the information may support bounds, probabilities, ranked hypotheses or a qualified conclusion.

A source may make one interpretation more plausible without proving motive. Experimental evidence may support an association while leaving causal uncertainty. A graph may allow an estimate within a range. A mathematical problem may have several possible values until an additional condition is supplied.

A mature representation can preserve uncertainty in structured form. “Not uniquely determined” is different from “nothing can be said.”

36. The mathematics laboratory: five ways to represent the same surface badly

Mathematics makes representation failure unusually visible because a mistaken model often leaves a symbolic trace. The following examples are constructed for this guide. They are not official examination questions.

Case 1: percentage base

A town’s hypothetical recycling total falls by twenty per cent to 4,800 tonnes. What was the earlier total? If E is the earlier total, 0.80E = 4,800, so E = 6,000. Adding twenty per cent of 4,800 gives 5,760, which would fall by twenty per cent to 4,608, not 4,800.

The representation test is forward consistency. Apply the stated process to the proposed original. If the result does not recreate the given final amount, the model is wrong even if the arithmetic looked plausible.

Case 2: weighted mean

A class has ten students with mean score 70 and twenty students with mean score 80. The combined mean is (10×70 + 20×80)/30 = 76⅔. Averaging 70 and 80 gives 75 because it treats the two groups as equally weighted objects rather than thirty equally weighted students.

The problem fingerprint asks: what are the observational units? Once the unit is “student,” group size must enter the representation.

Case 3: distance-time versus speed-time

A straight line from (0,0) to (10,50) on a distance-time graph represents a constant average gradient of five distance units per time unit over the interval. The same coordinates on a speed-time graph would make the area under the line relevant to displacement in an idealised interpretation. Identical geometry, different semantic roles.

The first representation move is not “this is a triangle.” It is “what do the axes mean?”

Case 4: inequality direction

Suppose −2x > 6. Dividing by −2 gives x < −3. A learner who represents division as “do the same operation to both sides” without preserving order reversal for a negative factor gets the wrong set.

Here the method is not simply arithmetic division. It is an order-preserving or order-reversing transformation depending on the sign of the multiplier. The representation includes the relation, not just the expressions.

Case 5: two points and model family

Two points, (0,1) and (2,5), fit the line y = 2x + 1. They also fit infinitely many nonlinear functions. If the problem states that y depends linearly on x, interpolation is justified. If it does not, a unique value between the points is not determined by the points alone.

The representation error is treating a convenient model as given fact.

37. Mathematics laboratory: equations can be equivalent, implications or approximations

Students often represent every line-to-line transformation as if it preserves exactly the same solution set. That is not always true.

Adding the same quantity to both sides of an equation is reversible. Multiplying by a known nonzero constant is reversible. Squaring both sides can introduce extra solutions. Dividing by an expression that could be zero can remove solutions. Replacing a function with a linear approximation changes the problem rather than preserving exact equivalence.

This matters because the symbol “=” in working can hide different logical relationships. If √(x+6)=x is squared to x+6=x², every solution of the original satisfies the squared equation, but not every solution of the squared equation satisfies the original. Verification against the original is part of the method.

A representation-first learner asks: what set of possibilities does this transformation preserve? The answer need not be phrased in formal logic for younger pupils. “Can this step create an extra answer?” is often enough.

38. Mathematics laboratory: a word problem is a model, not a word-to-symbol dictionary

Consider: “A service charges a fixed fee of 12 dollars plus 3 dollars per item. The total bill is 42 dollars. How many items were purchased?”

A keyword approach sees “plus” and “total.” A model approach identifies a fixed component and a variable component: 12 + 3n = 42, so n = 10.

Now change the wording: “A service charges 3 dollars per item, and a customer’s bill including a fixed fee is 42 dollars. The customer bought ten items. Find the fixed fee.” The same relationship is present, but the unknown changes: F + 3×10 = 42, so F = 12.

A third version asks which of two pricing plans is cheaper at different quantities. Now the representation involves two functions and a comparison, not merely one equation.

The words recur. The problem changes because the role of the unknown and the requested decision change.

39. Mathematics laboratory: ratio direction is part of the object

If the ratio of red to blue counters is 3:5, reversing it to 5:3 changes the relation. Students sometimes remember the numbers and lose the direction.

Suppose there are forty counters in total. Eight equal parts represent forty, so each part is five. Red = 15 and blue = 25. If the ratio were blue:red = 3:5, the counts would reverse.

Representation should keep labels attached to parts. A bare “3:5” written without role labels is fragile when several ratios occur in one problem.

This principle extends to rates such as dollars per kilogram versus kilograms per dollar. Reciprocals can both be meaningful, but they answer different questions.

40. Mathematics laboratory: coordinate geometry can hide a change of object

A slope is a ratio of changes, not simply “y divided by x.” For two points, the relevant quantities are differences: (y₂−y₁)/(x₂−x₁), provided x₂≠x₁.

A learner who represents slope as the vertical coordinate divided by the horizontal coordinate will be correct for points on a line through the origin and wrong elsewhere. The shortcut accidentally works on a special case.

Training should therefore include a pair of lines with the same slope, one through the origin and one not. Ask why y/x changes while Δy/Δx remains constant. This makes the invariant explicit.

41. Mathematics laboratory: area and perimeter can share numbers but not meaning

A rectangle with length eight and width five has perimeter 26 units and area 40 square units. Students who retrieve formulas mechanically may still benefit from a representation question: what is being accumulated?

Perimeter measures boundary length; area measures two-dimensional extent. Their units encode different objects. If a word problem gives fencing cost per metre, perimeter is relevant. If it gives tiling cost per square metre, area is relevant. The same rectangle and same dimensions produce different operations because the requested quantity differs.

A strong representation ties the physical or abstract meaning to the mathematical operation before formula retrieval.

42. Mathematics laboratory: a correct formula can be applied to the wrong triangle

Pythagoras’ theorem relates side lengths in a right-angled triangle. A triangle may look right-angled without being stated or established as such. If the condition is absent, applying the theorem is not licensed merely by appearance.

Conversely, a diagram may look distorted while a right-angle marker clearly establishes the condition. The representation should privilege encoded information over drawing style.

Near-nonexample practice is valuable here: show two similar-looking triangles, one with the required condition and one without. Ask which relation is justified and what extra evidence would be needed in the second case.

43. Mathematics laboratory: the sample space must match the event

A fair six-sided die is rolled twice. There are 36 equally likely ordered outcomes. If the question asks for a sum of seven, six ordered pairs satisfy it. If the learner represents outcomes only by possible sums 2 through 12 and assumes those sums are equally likely, the sample space is wrong.

The numbers 2 through 12 are possible sums, but they are not equally probable categories. Representation therefore includes the probability distribution over the representation, not merely the labels.

A table of ordered pairs can make the underlying multiplicity visible. Later, the learner may compress the representation once the structure is understood.

44. Mathematics laboratory: “at least” changes the event

When two fair coins are tossed, “exactly one head” consists of HT and TH, probability 1/2. “At least one head” consists of HT, TH and HH, probability 3/4. The phrase “at least” defines a set including the boundary and everything above it within the sample space.

A learner who represents both as “one head” loses the inclusion structure. Number-line and set representations can make the difference visible before probability rules are applied.

45. Mathematics laboratory: a constant in the story may be a variable in the model

A taxi hypothetical might say that the booking fee is fixed for one journey but asks how total cost changes across different journey lengths. The fee is constant with respect to distance; total cost is variable.

Representation requires choosing which quantities vary in the question’s comparison. A number can be fixed in one analysis and variable in another. This is why algebraic modelling cannot be reduced to replacing unknown nouns with x.

Ask: across the cases I am considering, what changes? What stays the same? That distinction often determines the model form.

46. Science laboratory: observation, inference and explanation are different representational layers

Science questions often appear difficult because several layers are compressed into one paragraph. A graph or table reports observations. The learner then infers a pattern, proposes a mechanism, evaluates a design or predicts what could happen next. Problems arise when one layer is mistaken for another.

Suppose a graph shows that plant height is greater in Group A than Group B. “Group A plants are taller” is a description of the observed result. “They are taller because they received more light” is an explanation. “More light caused the difference” is a causal claim requiring a design capable of supporting that conclusion. If the groups also differed in water, temperature or starting size, the causal representation is incomplete.

Representation-first science therefore begins by naming the evidential object: observation, pattern, inference, causal explanation, model prediction, limitation or proposed method. This stops students from using a true scientific statement to answer a different evidential question.

47. Science Failure Mode 1: the independent variable is treated as the cause before the design is checked

School investigations often teach learners to identify independent and dependent variables. That vocabulary is useful, but it can encourage a shortcut: the independent variable is automatically represented as the cause.

In a well-controlled experiment, changing one factor while holding relevant others sufficiently comparable can support a causal interpretation under the design’s assumptions. In an observational comparison or a confounded experiment, merely naming an independent variable does not establish causation.

A representation check asks: what was manipulated, what was merely observed, what differed between groups, and what alternative explanations remain? The learner does not need advanced causal inference language. They need to keep the design attached to the claim.

48. Science Failure Mode 2: a control variable is represented as something that can never change

“Keep all variables the same” is a useful beginner instruction and an impossible literal demand. Some conditions fluctuate, some are irrelevant, and some need only be controlled within a meaningful range.

Students can therefore represent experimental quality as a checklist of identical values rather than as a design that prevents plausible alternative explanations from dominating the comparison.

A better question is: which variables could affect the outcome enough to confuse the interpretation, and how will the method manage them? This turns “control variables” from ritual vocabulary into a causal representation.

49. Science Failure Mode 3: repeated trials are represented as a cure for every weakness

Repeating measurements can reduce the influence of random variation on an average under appropriate conditions. It cannot automatically repair systematic bias, confounding, a wrong instrument scale or a method that measures the wrong construct.

If a thermometer reads two degrees high every time, one hundred readings can give a very stable wrong average. If two treatment groups differ in both light and water, repeating the same confounded comparison does not isolate either factor.

Students should therefore represent “repeat” as a tool matched to a particular uncertainty, not as a universal improvement. The question becomes: what source of uncertainty is repetition intended to reduce?

50. Science Failure Mode 4: precision is represented as accuracy

A set of measurements tightly clustered around the same value can be precise and still systematically offset from the true value. A single value can be close to the true value by chance and yet come from a noisy process.

At school level, the exact terminology used depends on syllabus and context. The deeper representation point is that consistency among readings and correctness relative to a target are different properties.

A learner who sees “three measurements agree” and concludes “therefore they are correct” has represented repeatability as validation. A calibration or independent reference may be needed to assess a different question.

51. Science Failure Mode 5: a model prediction is represented as an observation

Simulations, diagrams and equations can generate predictions. Those predictions may be valuable and highly constrained by evidence. They remain outputs of a model rather than direct observations of the world.

If a simulation shows a future temperature curve, the curve is not an observed future. If a particle diagram shows a molecular mechanism, it is not a photograph unless the representation is explicitly built from such data. If an equation predicts a trajectory, measurement is still needed to test how well the model describes the system.

The learner should keep provenance attached: observed, calculated, simulated, inferred or assumed. This protects scientific reasoning and helps with source-based questions beyond science.

52. Science Failure Mode 6: the learner represents a mechanism as a correlation label

“Temperature increases reaction rate” may be a description of a pattern. A mechanistic explanation needs an account of why, within the model being taught, the change in temperature affects the frequency or effectiveness of collisions. Merely repeating the direction of the relationship does not expose the mechanism.

Conversely, a long mechanism can be unnecessary when the question asks only for the pattern. Representation determines response depth.

Train pairs: one question asks “describe the relationship”; another asks “explain the relationship.” Students should identify the additional reasoning demanded by the second before writing more words.

53. Science Failure Mode 7: the learner represents a graph as evidence of mechanism

A graph can show what changes with what. It rarely displays the complete mechanism by itself. The same trend may be consistent with several explanations.

For example, a decline in measured performance with increasing load may reflect one mechanism, several interacting mechanisms, a measurement ceiling or selection effects. The exact possibilities depend on the domain. The general rule is to avoid treating visual regularity as a complete causal model.

A strong answer separates “the graph shows…” from “one explanation is…” and then evaluates whether additional evidence supports the mechanism.

54. Science Failure Mode 8: the learner ignores boundary conditions

Scientific relationships often hold within ranges. A proportional approximation may work at low concentration and fail at high concentration. A model may assume constant temperature, negligible resistance or ideal behaviour.

Students sometimes learn the central relation and represent it as universal. The next unfamiliar question pushes outside the trained range and the same formula is applied automatically.

Teaching should therefore occasionally include a boundary case. Ask what assumption the relationship depends on and what observation would signal that the model is no longer adequate. This is not an invitation to teach every exception at once. It is an invitation to attach conditions to high-value rules.

55. Science Failure Mode 9: the learner represents uncertainty as error bars with no meaning

When uncertainty appears graphically, students may treat it as decoration. The task may ask whether two results differ meaningfully, whether estimates overlap, or how confident a comparison should be. The learner needs to know what the specific uncertainty representation means in that assessment context.

Different error bars can represent different quantities; one should not assume a universal interpretation. The representation must be read from the stated definition or method.

This is a general lesson: graphical uncertainty has semantics. Read the legend before drawing conclusions.

56. Science Failure Mode 10: the learner treats one mechanism as the only possible mechanism

A familiar explanation can dominate interpretation. If a system changes, the learner retrieves the mechanism most recently taught and represents the observation through it, even when the question leaves alternatives open.

Good science reasoning asks what evidence distinguishes candidates. The learner does not need an exhaustive list of every imaginable cause. They need to know whether the proposed explanation is supported uniquely, supported provisionally or merely plausible.

A useful exam phrase is often conditional: “This result is consistent with…” rather than “This proves…” when the evidence does not isolate the mechanism.

57. Science worked case: collision representation

Consider an idealised collision. A two-kilogram cart moving at three metres per second collides with a stationary one-kilogram cart, and the carts stick together. Assume external horizontal impulse during the collision is negligible.

The representation identifies a system, a brief interval and an external-influence assumption. Initial horizontal momentum is 2×3 + 1×0 = 6 kg m/s. The combined mass is 3 kg, so the shared velocity is 2 m/s.

If a learner instead represents the collision as one in which kinetic energy is conserved, the calculation gives a different result. The failure is not arithmetic. It is the process model. Sticking together signals an inelastic collision, and the stated external-impulse approximation supports momentum conservation for the selected system over the collision interval.

After the collision, if the carts slide over a rough surface and stop, the representation changes again. Friction now provides the external horizontal force that changes momentum. A relation useful during the collision is not automatically valid during the later stage.

The important skill is stage representation: what system, what interval, what influences, what relation?

58. Science worked case: identical graph, different meaning

Imagine a straight line through the origin. In a distance-time graph, constant gradient can represent constant speed. In a velocity-time graph, constant gradient can represent constant acceleration. In a force-extension graph within a linear elastic region, the gradient has another physical interpretation.

Students who memorise the visual phrase “straight line means constant speed” will fail when axes change. The structural representation is variable-specific.

A training sequence should keep the shape constant while changing the axes. Ask what the gradient’s units are in each case. Units force the learner to rebuild meaning.

59. Science worked case: conclusion exceeds design

Two fictional groups of seedlings are compared. Group A receives more water and more light than Group B. Group A grows taller. A learner writes, “More water caused the greater growth.”

The evidence supports a difference between groups under combined differences in water and light. It does not isolate water as the cause. A redesign to test water alone would hold other relevant conditions sufficiently comparable or use an appropriate multi-factor design depending on the question.

The representation failure is causal attribution, not biology vocabulary. The learner may know perfectly well how water affects plant processes and still overclaim what this design establishes.

60. Science worked case: systematic offset

A hypothetical instrument reports every true value plus two units. Repeating measurements gives stable readings around the same shifted value. The representation “more repeats = more accurate” fails because the dominant problem is systematic offset, not random scatter.

A calibration check against a known reference targets the offset. Repetition can still be useful for random variation once the measurement model is understood.

This is a broader scientific habit: match the repair to the failure model.

61. Language laboratory: reading creates a problem representation too

English and language examinations can hide representation failure because there are no equations showing the first wrong move. The learner produces a paragraph. The paragraph may be grammatical, fluent and knowledgeable while answering a nearby question.

Reading comprehension requires building a representation of what the question requests and how the source supports it. Writing requires representing audience, purpose, task, evidence range and success criteria. Oral work requires representing the conversational demand and responding at an appropriate scope.

The internal question “What am I doing with this text?” is therefore as important as “What does this text say?”

62. Language Failure Mode 1: topic is represented as purpose

A passage is about urban trees. The question asks why the writer includes an anecdote about a hot playground. The learner answers, “The writer wants to talk about trees.” That identifies topic, not rhetorical function.

A stronger representation distinguishes subject from purpose. The anecdote may illustrate the human experience of heat, make an abstract argument concrete, create urgency or support a claim about shade. The exact answer must be tied to the passage.

Train students to complete two separate sentences: “This passage is about…” and “In this sentence/paragraph, the writer is trying to…” The separation prevents topic from swallowing function.

63. Language Failure Mode 2: evidence is represented as explanation

A learner quotes a sentence and assumes the quotation explains itself. Evidence is present, but the connection between evidence and claim remains implicit.

For an analysis question, the representation should include the inferential bridge: what does this wording, action or detail suggest, and why does it support the proposed interpretation?

Conversely, an answer can contain an explanation with no textual support. Strong representation keeps claim, evidence and reasoning as distinct but connected roles.

64. Language Failure Mode 3: inference is represented as imagination

When the text does not state an answer directly, some learners feel free to invent plausible background. A character pauses, so perhaps they are ashamed, frightened, tired, guilty or distracted.

An inference is constrained by evidence. The learner should ask which interpretation accounts for the available cues with the fewest unsupported additions.

Multiple interpretations can be defensible. The standard is not matching one hidden model sentence. It is evidential fit and appropriate scope.

65. Language Failure Mode 4: summary is represented as shorter retelling

A summary task defines a focus. The learner must select principal points relevant to that focus and compress them without importing new judgement.

A shorter retelling can still include irrelevant events and omit the central argument. Representation begins with the summary’s target, not with the passage sequence.

One diagnostic uses the same passage with two different summary instructions. If the learner writes nearly the same summary both times, relevance selection may be weak.

66. Language Failure Mode 5: comparison is represented as two separate descriptions

Two paragraphs—one about Source A and one about Source B—can contain excellent summaries and very little comparison.

Comparison needs a shared axis: viewpoint, evidence, method, purpose, tone, reliability, criteria or another dimension specified by the task. The learner should align claims along that axis.

A simple planning grid can support this during training. Later, the comparison can be written directly without the scaffold.

67. Language Failure Mode 6: evaluation is represented as “pros and cons”

Listing one positive and one negative point does not automatically produce evaluation. Evaluation requires criteria and judgement: how important is each consideration, under what conditions, and what conclusion follows?

“Both sides” can be useful when relevant, but it is not the representation of evaluation itself. A task may contain several benefits but one decisive limitation, or vice versa.

Teach the learner to ask: “What standard am I judging against?” Without a criterion, evaluation becomes opinion collection.

68. Language Failure Mode 7: the source boundary is lost

Some questions ask for information from the passage. Others invite wider knowledge. Others require synthesis across sources. If the learner does not represent the evidence boundary, true outside knowledge can become irrelevant or even disallowed.

A response can therefore be factually correct and task-invalid. The issue is not knowledge quality but source scope.

Underline or paraphrase the boundary during training: “using the passage,” “based on Source B,” “using both sources,” “drawing on your own knowledge.” These phrases determine what evidence can enter the response.

69. Language Failure Mode 8: audience is represented as decoration

In writing, audience influences register, explanation depth, assumed knowledge and persuasive strategy. Students may treat “write to your principal” as a greeting choice rather than as a constraint on the whole communication.

A proposal to a school leader should make decisions easy to understand, support recommendations and use an appropriate tone. The same topic in a message to a friend would be represented differently.

Audience is therefore part of the problem, not an afterthought.

70. Language Failure Mode 9: purpose is inferred from genre alone

A speech can inform, persuade, commemorate, explain or combine purposes. An article can argue, report, review or instruct. Genre narrows possibilities but does not finish representation.

The prompt and context determine the communicative job. Students who memorise one “speech structure” may produce a polished answer that performs the wrong purpose.

71. Language Failure Mode 10: the learner answers a stronger claim than the question asks

A question asks whether a policy can help. The learner argues that it will solve the problem. A passage suggests a character may be worried; the learner declares that the character is terrified.

Modality matters. Can, may, likely, always, only, sufficient and necessary define different claim strengths. Representation should preserve that strength.

A useful check is to compare the certainty of the answer with the certainty of the question and evidence. Overstatement is often a representation error before it becomes a style problem.

72. Language worked case: same passage, four tasks

Consider this original micro-passage:

The community centre had placed three new water dispensers beside the sports hall. During the first week, the floor around the busiest dispenser was repeatedly wet. Staff first assumed that users were spilling water, but a volunteer noticed that the puddle reappeared before the evening crowd arrived. A maintenance check found a loose connection behind the unit. After it was tightened, the floor remained dry.

Retrieval question: What did the maintenance check find? Answer: a loose connection behind the unit.

Inference question: Why was the volunteer’s observation important? Because it challenged the assumption that crowds of users were causing the puddle; the water reappeared before the evening crowd.

Explanation question: How did the sequence of observations help identify the likely cause? The timing evidence weakened the user-spillage explanation and prompted inspection of the unit, where a loose connection was found.

Evaluation question: Does the passage prove that every puddle around a dispenser is caused by a loose connection? No. It supports that conclusion for this incident after inspection and repair; other incidents could have different causes.

The source remains unchanged. The representation of the task changes. A learner who uses one generic comprehension template will flatten these distinctions.

73. Language worked case: source comparison needs a shared question

Source A argues that a school reading programme succeeds if participation rises. Source B argues that success should be judged by changes in comprehension and sustained reading outside the programme.

A weak comparison says, “Both sources discuss reading, but Source A talks about participation while Source B talks about comprehension.” This is a beginning.

A stronger comparison identifies the shared axis: they use different criteria for success. Source A prioritises access and participation; Source B demands evidence of deeper or durable outcomes. The disagreement is not merely about different facts. It is partly about what should count as success.

That representation can then support evaluation: which criterion is appropriate for the decision being made, and can both be measured?

74. Language worked case: rhetorical analysis begins with local function

Suppose a writer describes a crowded bus as “a moving room in which nobody had agreed to meet.” A learner identifies metaphor. That label alone does not explain function.

The representation should ask what relationship the image constructs. It frames forced physical proximity as an accidental social gathering, potentially highlighting discomfort, anonymity or urban density. The exact effect depends on surrounding context.

Device names are tools for noticing. They are not substitutes for explaining what the wording does here.

75. Language worked case: “discuss” is not a licence for balance without judgement

In some assessment systems, “discuss” invites consideration of relevant perspectives. The exact expectations depend on the syllabus. Students can nevertheless make a general representation error by assuming that discussion means equal paragraph counts for two sides.

Evidence may be asymmetric. One position may have stronger support. A discussion can acknowledge alternatives while reaching a reasoned judgement.

The learner should check the actual assessment guidance rather than applying a generic command-word slogan across boards and subjects.

76. Representation failure is often invisible in marked work

Marked scripts show final products. They do not always show the learner’s internal problem model. A wrong equation may expose the representation directly, but a paragraph can hide it. A correct answer can hide it too if the learner guessed, matched a pattern or relied on a tutor prompt.

Diagnosis therefore needs more than the answer key. Preserve the original attempt, then ask a small number of questions that reconstruct the decision without turning every review into an oral examination.

Useful questions include: “What did you think the question was asking?” “What did this number represent?” “Which sentence in the passage made you think that?” “What changed after the first event?” “Which condition allowed this formula?” “What would have made you choose a different method?”

These are not truth machines. Learners can rationalise after the event. The explanation should be tested against fresh tasks. A diagnosis becomes stronger when the same representation pattern recurs under new wording.

77. The representation audit

A representation audit is a short review of the problem before examining detailed execution. It asks five questions:

1. Target: Did the learner represent what had to be produced?
2. Roles: Did they identify what the important quantities or source elements represented?
3. Relations: Did they connect the elements appropriately?
4. Constraints: Did they preserve conditions, domain, evidence scope and state changes?
5. Output: Did the result return to the requested form?

If all five are sound, move downstream. The issue may be retrieval, method selection, execution, checking or pacing. If one fails, repair that layer before adding more speed or volume.

The audit is deliberately smaller than the ten-part fingerprint. The fingerprint helps teaching and analysis. The audit helps routine review.

78. Alicia’s representation audit

Alicia’s marked mathematics paper contains a wrong answer to a reverse-percentage problem. Her working is:

96 + 20% = 115.2

The arithmetic is not the first issue. Alicia’s representation audit asks:

• Target: original price — correct.
• Role of 96: she says “current price” — partially useful.
• Relationship: she says “the original is twenty per cent more than the current price” — this is the divergence.
• Constraint: the twenty per cent reduction was from the original base — omitted.
• Output: dollars — correct form.

The repair is a pair of tasks: one forward percentage, one reverse percentage, same numbers. Alicia must identify which quantity is the base before calculating. If she succeeds independently across a later mixed set, there is little value in continuing to discuss percentages abstractly.

79. Tricia’s representation audit

Tricia writes a strong comprehension paragraph but receives limited credit. The question asks why a writer includes a statistic. Her answer explains what the statistic says.

The audit shows:

• Target: Tricia represented it as “explain the statistic.”
• Role: statistic represented as information, not rhetorical evidence.
• Relationship: no connection to the writer’s claim.
• Constraint: use the local context — weakly attended.
• Output: explanatory paragraph — fluent but misdirected.

The repair is not more vocabulary. Tricia compares three questions about the same statistic: What does it show? Why does the writer include it? How convincing is it? The changing task forces three different representations of the same source element.

80. Kai Kai’s representation audit

Kai Kai correctly solves a without-replacement probability task when the teacher draws the bag after the first draw. On a fresh task without the diagram, she keeps the original denominator.

The audit suggests that the external picture carried the state update. The next teaching step is not another probability formula. Kai Kai must construct the before/after state herself.

A simple table works:

Stage	Total items	Relevant items
Before first draw	5	3 red
After a red draw	4	2 red

Then the table is faded. The goal is internal state updating, not permanent dependence on a table.

81. Do not confuse a representation aid with the representation skill

Diagrams, bar models, tables, colour-coding and annotation can improve a task representation. They can also become external prosthetics that the learner never learns to construct or choose.

A bar model supplied by the teacher demonstrates one representation. A student who can use it but cannot decide when a bar model is useful has a different capability. A learner who can draw one for every problem may overuse it when a simpler equation is clearer.

Representation instruction should therefore include choosing and fading representations. Ask: does this diagram reveal a relationship that is hard to hold mentally? Does the table make state changes visible? Does the number line represent an inequality better than a verbal sentence? If the representation adds no useful information, it may be unnecessary.

82. Translation practice: move between representations, not only within one

One of the strongest ways to test understanding is to ask learners to translate the same structure across words, diagrams, tables, graphs, equations and verbal explanations.

For the fixed-fee problem, show:

• Words: 12-dollar fixed fee plus 3 dollars per item.
• Equation: C = 12 + 3n.
• Table: n values paired with C values.
• Graph: line with intercept 12 and gradient 3.
• Verbal interpretation: each extra item adds 3 dollars; even zero items retains the fixed fee.

The representations are not identical in information density. The graph makes trend and intercept visible. The equation makes symbolic manipulation easy. The table provides discrete examples. Translation helps learners identify what each format preserves.

Do not assume that translating once creates flexible understanding. Use changed contexts and ask the learner to choose the representation that makes the requested operation easiest.

83. Contrast pairs: change one thing that matters

Random variation is not the same as instructional contrast. To teach a representation boundary, keep most features stable and change one decisive condition.

Pair:

• “The original price is 120; find the discounted price.”
• “The discounted price is 96; find the original price.”

Or:

• “Two equal time intervals at different speeds.”
• “Two equal distances at different speeds.”

Or:

• “What does the statistic show?”
• “Why does the writer include the statistic?”

The learner should name the feature that changes the representation. This is more powerful than doing ten randomly different questions because the comparison isolates the decision boundary.

84. Same structure, different surface

The complementary exercise changes the story while preserving the relation. A fixed fee plus variable cost can appear as a phone plan, delivery service, club membership or equipment rental. A part-whole ratio can appear as recipe ingredients, map scale or class composition.

The learner’s task is to identify what stays invariant despite the new nouns. This supports transfer without pretending that all real-world contexts are mathematically interchangeable.

Ask for a sentence such as, “The story changed, but the total still consists of a fixed component plus a constant amount per unit.” This makes structural recognition explicit.

85. Near nonexamples teach where the representation stops

A concept is often clearer when paired with a case that looks similar but does not belong. If every percentage problem in a set is reverse percentage, the learner can solve them through worksheet context. Add a forward percentage problem. If every graph asks for gradient, include one asking for a coordinate value.

The nonexample should be near enough that the distinction matters. A triangle versus a poem teaches little about when Pythagoras applies. Two similar triangles, one confirmed right-angled and one not, teach the condition directly.

Near nonexamples should not dominate practice to the point that every question feels adversarial. Their job is to establish boundaries, then ordinary cases should remain common enough for fluent execution.

86. Representation mutation

Mutation takes one problem and changes a single representational dimension. It is a practical way to convert one solved example into a family of transfer tests.

Possible mutations include:

• reverse which quantity is known and unknown;
• change “exactly” to “at least”;
• change with replacement to without replacement;
• change a line graph’s axes while keeping its shape;
• add a domain restriction;
• remove the assumption of linearity;
• change “describe” to “explain”;
• change source-only evidence to source-plus-own-knowledge;
• change equal time to equal distance;
• change a measured value into an estimate with stated uncertainty.

After mutation, ask what part of the previous representation remains and what must be rebuilt. This is more diagnostic than simply asking whether the learner can solve another unrelated problem.

87. Problem sorting can reveal representation patterns

Give learners a small set of questions and ask them to group those that would be approached similarly. Then ask why.

A surface-based grouping may place all problems containing a triangle together. A structural grouping may separate right-triangle length problems from similarity problems and optimisation problems involving triangles.

Do not interpret one sorting task as a stable cognitive profile. The value is conversational: it reveals which features the learner currently treats as important.

After teaching, repeat with new problems. If the grouping becomes more aligned to relevant structural conditions, representation may be changing.

88. The first-line test

On a set of examination questions, ask only for the first line or first move. No full solution.

Examples:

• Define the unknown and write the governing relation.
• Name the variable represented on each axis.
• State the comparison axis for two sources.
• Identify the evidence boundary.
• Write the first equation without solving it.
• State whether the problem is uniquely determined.

This isolates representation and method entry from long execution. If the first line is wrong, a full answer may only automate the downstream consequence. If the first line is correct but the full answer fails, the bottleneck has moved.

The first-line test must eventually reconnect to whole-task performance. It is a diagnostic and training tool, not a substitute for solving.

89. The wrong-problem test

Give the learner a worked solution that is internally correct but answers a subtly different question. Ask them to identify the mismatch.

For example, provide a correct calculation of the final price when the question asks for the original price. Provide a well-written summary when the prompt asks for evaluation. Provide a graph gradient when the question asks for area.

This exercise is powerful because local correctness cannot be used as a shortcut. The learner must compare the solution’s object with the task’s object.

It also supports checking: a correct-looking answer should still be validated against the original problem.

90. The representation-first marking code

A teacher can use a light code during selected practice:

• T: target represented incorrectly;
• R: roles or quantities misrepresented;
• L: relationship or logical structure wrong;
• C: constraint lost;
• S: state not updated;
• E: evidence boundary wrong;
• O: output form mismatched;
• X: execution error after a sound representation.

This is not an official assessment taxonomy. It is a teacher shorthand. Use it only if it changes the next practice decision. If codes accumulate without targeted follow-up, they become decorative analytics.

91. Representation training should become quieter over time

Early teaching may involve explicit questions: What is known? What is unknown? Which quantity is the base? Which evidence is allowed? What changes after this action?

As skill stabilises, those prompts should compress. The examination version might be a two-second internal check rather than a written checklist.

This is an application of support fading. A learner who still needs a ten-question representation audit on every routine task may understand the audit without owning the problem model. The aim is increasingly automatic but still defensible construction.

92. Representation speed should not be trained before representation quality

Timed practice can encourage quick pattern matching. If the learner’s representations are still fragile, the clock may reward the very surface cues that need to be replaced.

First stabilise accurate representation on selected contrasts. Then reduce time gradually. The target is not leisurely interpretation forever; examinations impose time. The target is fast enough construction that preserves decisive constraints.

See How Timed Practice Fails for the broader speed-accuracy transition.

93. Representation under pressure: why familiar surfaces take over

Under time pressure, the learner has an incentive to reduce interpretation cost. Familiar cues become attractive because they allow a rapid route into action.

This can be adaptive when expertise has built reliable pattern recognition. It can be dangerous when the cue is only loosely connected to the governing structure.

Training should therefore create fast recognition from valid conditions, not from chapter labels. Mixed practice, near contrasts and delayed transfer can help move the cue from “this looks like yesterday’s worksheet” to “this condition establishes the relationship I need.”

94. The ten-second exam representation protocol

For many questions, the live version can be compressed to four moves:

1. Target: What exactly must I produce?
2. Structure: What relation, evidence or state matters?
3. Constraint: What small condition could change the route?
4. Return: Will my final answer visibly satisfy the requested form?

Not every question needs ten literal seconds. Routine items may require less. Complex unfamiliar questions may need more. The protocol is a mental organisation, not a mandatory stopwatch rule.

95. The one-minute reset when the representation feels wrong

Sometimes execution becomes strangely difficult because the task model is wrong. Rather than adding more effort to the same representation, use a reset:

1. Stop local manipulation.
2. Restate the target without using the current method’s language.
3. Relabel known quantities or evidence.
4. List one or two constraints.
5. Ask what relationship would connect the givens to the target.
6. Only then reconsider methods.

This is not guaranteed to rescue every question. It is a way to prevent ten minutes of flawless work on the wrong problem.

96. Checking should return to the representation, not only the calculation

Many students check by repeating arithmetic. If the original equation represented the wrong problem, the repeated calculation can reproduce the same wrong answer.

An independent check asks a different question: does the answer behave correctly in the original situation?

For reverse percentage, apply the stated discount forward. For a rate, inspect units. For probability, ask whether the answer lies between zero and one and whether extreme cases make sense. For a source answer, compare each claim with the evidence boundary. For an essay, reread the exact command and ask whether each paragraph contributes to it.

Checking is strongest when it attacks the model as well as the mechanics.

97. Problem representation and exam recovery

A bad representation can produce a stall. The learner keeps manipulating because the task seems familiar but nothing resolves. If the stall persists, the recovery system should include representation rebuilding rather than only method switching.

Ask: “Am I stuck because this method is hard, or because I have built the wrong object?” A wrong object may need a new diagram, variable definition, source scope or target sentence before another method becomes useful.

This connects to How Exam Recovery Fails. Rebuilding the representation is one possible bounded recovery action, not a reason to restart every difficult question from zero.

98. Problem representation and model answers

Model answers can make a correct representation visible without teaching how it was constructed. A student sees the right equation, diagram or paragraph plan and understands it after the fact.

The hidden training question is: what feature of the original problem led to that representation? If the model answer begins with a bar diagram, why was a bar diagram useful? If it begins by defining a variable, why that variable? If it organises an essay by criteria, what in the prompt made that structure appropriate?

How Model Answers Fail owns the broader exemplar dependency problem. Here the specific danger is borrowed representation.

99. Problem representation and method selection

Representation and method selection form a loop. A representation can suggest a method; trying a method can reveal that the representation was incomplete.

If a learner cannot choose between methods, do not assume the problem is simply a large repertoire. Ask whether the representations distinguish the cases. Two methods can feel equally plausible because a decisive condition has not been encoded.

Conversely, a strong representation can make method choice obvious. Once a learner represents a reverse percentage as “final = remaining fraction × original,” division is not a memorised trick. It follows from the relation.

100. Problem representation and misconception repair

A misconception is a wrong or incomplete underlying model. A representation failure can occur even when the underlying concept is understood, because the learner misreads the current task. The two can also interact.

If a student believes the equals sign always means “the answer comes next,” that misconception affects how equations are represented. If the student understands equality but misreads which quantity an equation should represent, the local problem is different.

Do not treat every representation error as a deep misconception. Test the boundary with fresh cases.

101. Problem representation and cognitive load

Long questions, unfamiliar layouts and multiple representations can overload working memory. A learner may know how to represent the task but fail to keep all constraints active.

External supports such as short annotations, diagrams and tables can reduce this burden during learning. The aim is not to prove a permanent memory limitation. It is to build a manageable representation and gradually increase independence.

When a support dramatically improves performance, ask what information it made easier to maintain. Then test whether the learner can construct a similar support themselves.

102. Problem representation and vocabulary

Sometimes representation fails because a key word is not understood. “Consecutive,” “residual,” “contrast,” “approximate,” “justify,” “proportional,” “minimum,” “evaluate” or “corresponding” can determine structure.

This is not a reason to reduce all examination problems to vocabulary lessons. It is a reason to distinguish language access from conceptual failure. If explaining one word repairs the representation, targeted vocabulary work may have high value.

Then test the concept with the word absent or replaced by an equivalent phrase. The learner should understand the relation, not only one lexical cue.

103. Problem representation and background knowledge

Reading and real-world problem solving often require background knowledge to build a coherent situation model. A reader who does not know what a service hatch is, how a public consultation works or why a control group matters may misrepresent a passage even when decoding is accurate.

Teachers can supply essential background knowledge before or during learning. In an examination, the assessment should be interpreted according to its intended knowledge demands.

The important diagnostic question is whether the learner lacks the target reasoning or the world model needed to understand the prompt.

104. Problem representation and digital interfaces

Digital examinations add interface representation. Information may be split across tabs, scrolling panels, pop-ups, reference sheets and answer boxes.

A learner can lose a condition because it is off-screen, misread which source belongs to which question, or treat a flag as a completed answer. Interface fluency becomes part of preserving the problem state.

Where possible, practise with a representative interface. The skill is not only clicking faster. It is maintaining the task model while navigating.

105. Problem representation and multi-step questions

Multi-step questions can contain several local representations. A result from part (a) becomes given information for part (b). A variable changes meaning. A condition introduced later modifies the model.

The dedicated multi-step failure page owns that broader chain problem. The representation contribution is to mark state transitions and role changes. Do not keep solving part (a) when part (b) has introduced a new target.

106. Problem representation and integrated questions

Integrated questions require knowledge across topics or domains. The learner may know each component yet fail to construct a single representation that connects them.

The newer integrated-question owner covers that synthesis. The representation lens asks which object or relationship can serve as the bridge: a shared variable, conservation relation, evidence criterion, common timeline or causal chain.

107. Problem representation and proof

Proof questions require representing the logical scope of the claim. “For all integers” differs from “for these examples.” “If P then Q” differs from “P if and only if Q.” A learner can perform correct algebra and prove a weaker statement than required.

The proof/derivation owner covers the larger issue. Representation contributes by preserving domain, quantifier, assumptions and target conclusion.

108. Problem representation and uncertainty

Advanced problems often ask not for a single answer but for what can be known under uncertainty. The representation may be a range, confidence statement, scenario set or conditional conclusion.

Students trained mostly on deterministic exercises can feel that uncertainty means the problem is incomplete. In many real systems, representing uncertainty accurately is the solution.

The school-level version begins with honest qualifiers: “under this assumption,” “within this range,” “the evidence suggests,” or “cannot be uniquely determined from the information given.”

109. The representation repair ladder

Repairs should move from the smallest useful support back to independent whole-task performance.

1. Restate the target.
2. Relabel roles.
3. Make the relationship visible.
4. Add or recover constraints.
5. Choose a representation format.
6. Attempt the method.
7. Check against the original task.
8. Try a near variant.
9. Try a delayed mixed variant.
10. Return to timed integrated performance.

The learner can move backward when a new case exposes a weak boundary. The ladder is a control structure, not a fixed curriculum sequence.

110. A representation error ledger

For recurring errors, keep a compact record with five fields:

• Visible wrong answer or first move;
• Wrong internal representation;
• Decisive feature that was missed;
• Corrected representation;
• Fresh task that tests the boundary.

Example:

Wrong move: added twenty per cent to final price.
Wrong representation: treated final price as the base.
Missed feature: discount was applied to original.
Corrected representation: final = 80% of original.
Retest: new reverse-percentage problem mixed with forward percentage.

Retire entries after the distinction survives appropriate variation and delay. Do not maintain a museum of every past mistake.

111. How parents can help without supplying the representation

At home, adults often help by translating the question immediately: “This is a ratio question,” “You need to divide,” “They are asking for the original amount.” The child then succeeds, but the representation decision belonged to the adult.

A lower-support prompt is, “What is the question asking you to find?” Next: “What does this number represent?” Next: “What relationship connects them?” Only later, if necessary, name the method.

Do not withhold explanation indefinitely. If the child genuinely lacks the concept, teach it. The aim is to preserve the child’s share of the decision where possible, not to turn homework into a guessing test.

112. How teachers can design representation-rich lessons

A representation-rich lesson need not involve more materials. It needs tasks that make representation consequential.

Useful design moves include:

• same surface, different structure;
• different surface, same structure;
• extra information;
• missing information;
• changed axes;
• reversed known and unknown;
• near nonexamples;
• multiple valid representations;
• incorrect representations to diagnose;
• translation between words, diagrams, tables and symbols.

After representation stabilises, execution practice can become extensive. The point is not to turn every lesson into metacognitive discussion. It is to make the hidden choice visible long enough for the learner to own it.

113. How tutors can avoid becoming the representation engine

Tutors see problems quickly. That expertise creates a risk: the tutor translates the problem before the student has had to. A small hint such as “draw a bar model” can already contain the representation choice.

During acquisition, modelling is appropriate. During transfer, fade. Ask the learner to choose the representation and justify it briefly. If the learner chooses a different valid representation, do not force the tutor’s preferred one unless the assessment or learning purpose requires it.

Independent performance means the representation travels with the learner.

114. Independent practice bank: build the problem before solving it

The following tasks are original. They are designed to expose representation decisions, not to imitate any specific examination paper. Work each task in two stages. First, write or think the representation: target, roles, relationship and decisive constraint. Second, solve or answer. The commentary explains the representation, not merely the final answer.

Practice 1: reverse percentage

A membership fee is reduced by fifteen per cent and becomes 68 dollars. What was the original fee?

Representation. Sixty-eight dollars is the final amount after the reduction. Therefore it represents eighty-five per cent of the original amount.

Solution. If F is the original fee, 0.85F = 68, so F = 80 dollars.

Boundary check. If the problem instead said the original fee was 68 dollars and asked for the reduced fee, the role of 68 would change and the operation would become 0.85×68.

Practice 2: weighted mean

Group A has twelve students with a mean score of 65. Group B has eighteen students with a mean score of 75. Find the combined mean.

Representation. The observational unit is a student, so the group means must be weighted by group sizes.

Solution. Total score = 12×65 + 18×75 = 780 + 1350 = 2130. Total students = 30. Combined mean = 71.

Boundary check. Averaging 65 and 75 would answer a different question: the mean of the two group means with equal group weight.

Practice 3: without replacement

A bag contains five black counters and three white counters. Two counters are drawn without replacement. Find the probability that both are white.

Representation. The sample space changes after the first draw. If the first counter is white, seven counters remain, two of them white.

Solution. (3/8)(2/7) = 3/28.

Boundary check. With replacement, the second factor would remain 3/8.

Practice 4: exactly versus at least

Three fair coins are tossed. Find the probability of at least two heads.

Representation. “At least two” includes exactly two heads and exactly three heads.

Solution. Outcomes with exactly two heads: HHT, HTH, THH. Outcome with three heads: HHH. Four of eight equally likely outcomes, so probability = 1/2.

Boundary check. “Exactly two” would exclude HHH.

Practice 5: line or curve?

A relationship passes through the points (0,2) and (2,6). The problem asks for the value when x = 1 but does not state the relationship is linear. Can the value be uniquely determined?

Representation. Two points do not determine an arbitrary function. A model family is missing.

Answer. No. If the function is linear, the value would be 4. But nonlinear functions can pass through the same two points and take a different value at x = 1.

Boundary check. If linearity were explicitly stated, interpolation would be justified.

Practice 6: graph semantics

A straight line on a graph runs from (0,0) to (4,20). The horizontal axis is time in seconds. In Version A, the vertical axis is distance in metres. In Version B, the vertical axis is velocity in metres per second. What does the gradient mean in each version?

Representation. Same geometric shape, different variables.

Answer. In Version A, gradient is 5 metres per second, a speed. In Version B, gradient is 5 metres per second squared, an acceleration.

Boundary check. Area under Version B has units of metres and can represent displacement in the appropriate interpretation; area under Version A does not represent distance in the same way.

Practice 7: geometry diagram

A triangle in a sketch appears right-angled, but the question provides no right-angle marker and no information from which a right angle can be deduced. Can Pythagoras’ theorem be applied merely from the appearance?

Representation. Visual appearance is not an established condition.

Answer. No. Additional evidence that the triangle is right-angled is needed.

Boundary check. If a right-angle marker or derived perpendicular relation is present, the condition becomes established.

Practice 8: source boundary

A comprehension question says, “Using only Source B, explain why the proposal was delayed.” The learner knows a relevant outside fact. Should it be used?

Representation. The evidence boundary is explicitly restricted to Source B.

Answer. The response should be supported by Source B. The outside fact may be true but does not belong in this answer unless the assessment guidance says otherwise.

Boundary check. If the question invited own knowledge, the evidence boundary would change.

Practice 9: rhetorical purpose

A writer argues that school meals should be improved and includes a short account from a student who skips lunch because the queue is too long. Why might the writer include the account?

Representation. The question asks for function, not merely content.

Answer. A defensible response might explain that the account makes the access problem concrete and provides an individual example supporting the broader claim that current arrangements deter students from eating.

Boundary check. Simply restating that the student skips lunch describes the example but does not fully address why it is included.

Practice 10: experimental causality

Plant Group A receives more light and more water than Group B and grows taller. Can the difference be attributed uniquely to light?

Representation. Two potentially relevant conditions differ.

Answer. No. The comparison does not isolate light from water. It shows a difference under combined conditions.

Boundary check. A redesign aimed specifically at light would need to manage water and other relevant variables appropriately.

Practice 11: systematic bias

A scale always reads 0.5 kg too high. A learner proposes taking twenty readings and averaging them. Will that remove the offset?

Representation. The problem describes a fixed systematic shift, not only random variation.

Answer. Averaging repeated readings will not by itself remove the 0.5 kg offset. Calibration or correction relative to a known reference is needed.

Practice 12: proof scope

A learner tests n = 1, 2, 3 and 4 and concludes that n(n+1) is even for every integer n. Is that a proof?

Representation. The claim is universal over all integers, while the evidence covers four examples.

Answer. No. A general argument is needed. Of two consecutive integers n and n+1, one is even; therefore their product is even.

Boundary check. Exhaustive checking can prove a claim over a finite explicitly bounded domain if every case is covered.

Practice 13: minimum versus exact

A venue requires at least 120 seats. A design has 120 seats. Does “at least 120” mean the design must have exactly 120?

Representation. At least defines a lower bound.

Answer. No. Any number of seats greater than or equal to 120 satisfies that condition, assuming no other constraints.

Practice 14: stock and flow

A reservoir contains 1,000 cubic metres of water. Water enters at 20 cubic metres per minute and leaves at 30 cubic metres per minute. Is the water level increasing because there is an inflow?

Representation. The stock changes according to net flow, not inflow alone.

Answer. No. Net flow is −10 cubic metres per minute, so the stored volume is decreasing while those rates remain constant.

Practice 15: same topic, different command

A passage describes a city’s new cycling lanes. Question A asks, “Summarise the reasons given for building the lanes.” Question B asks, “Evaluate whether the reasons are convincing.” Can the same answer be used?

Representation. Same content domain, different response operation.

Answer. No. A summary represents the source’s reasons concisely. Evaluation judges their strength using relevant criteria and evidence.

Practice 16: unknown drift

A question asks for the angle of elevation. The learner correctly calculates the vertical height and stops. Is the answer complete?

Representation. The calculated height may be an intermediate quantity, not the target.

Answer. No, unless the height itself is also requested. The learner must continue from the relevant geometry to the angle.

Practice 17: irrelevant information

A store sells notebooks for 4 dollars each. A customer buys five notebooks. The store opened in 1998. How much do the notebooks cost before tax?

Representation. Opening year is irrelevant to the stated cost relation.

Answer. 5×4 = 20 dollars.

Boundary check. Extra information should be excluded because it has no structural role, not because it is numerically old or visually separate.

Practice 18: insufficient information

A rectangle has perimeter 20 units. What is its area?

Representation. Perimeter alone does not uniquely determine side lengths.

Answer. Not uniquely determined. If the side lengths are x and 10−x for positive x, the area is x(10−x), which varies with x.

Boundary check. If the rectangle were stated to be a square, the area would be uniquely determined as 25 square units.

Practice 19: estimate versus exact value

A graph allows a root to be read as approximately 1.4. The question asks for an exact solution to x² = 2. Is 1.4 sufficient?

Representation. The requested output is exact.

Answer. No. The exact real solutions are ±√2. A graph can support an estimate but does not by itself supply the requested exact form.

Practice 20: representation of uncertainty

A source states that a policy “may reduce waiting times if staffing remains stable.” A learner writes, “The policy will reduce waiting times.” What changed?

Representation. A conditional, uncertain claim became an unconditional prediction.

Repair. Preserve the modality and condition: the source suggests a possible reduction under stable staffing; it does not guarantee the result.

115. How to review the practice bank

Do not turn the practice bank into another answer collection. The important record is not the twenty final answers. It is the set of distinctions that controlled them.

After attempting the tasks, sort mistakes into representation families:

• wrong target;
• wrong role or base;
• wrong reference set or denominator;
• state not updated;
• constraint lost;
• evidence boundary lost;
• surface cue overused;
• model family assumed;
• uncertainty collapsed;
• output form mismatched.

Choose the two most recurrent families and create fresh near contrasts. Do not immediately practise all twenty again. The purpose of the long article is diagnostic breadth; the purpose of a learner’s next session is targeted change.

116. A seven-day representation repair cycle

This is an example schedule, not a universal prescription.

Day 1: diagnose. Use a small mixed set and preserve first moves. Identify one or two recurring representation failures.

Day 2: contrast. Work with near pairs that change only the decisive feature.

Day 3: translate. Move among words, diagrams, equations, graphs or source roles.

Day 4: mutate. Change the known/unknown, evidence boundary, state or constraint.

Day 5: mix. Remove topic labels and combine neighbouring task types.

Day 6: delay. Reattempt without reviewing the earlier model first.

Day 7: integrate. Use a timed section or realistic task where representation is only one component of performance.

If performance is still weak, revisit the diagnosis. The problem may be content knowledge, execution or language rather than representation alone.

117. A thirty-minute tutor lesson

A short focused lesson can work like this:

1. Five minutes: three independent first-line tasks.
2. Five minutes: compare the first moves and locate the earliest divergence.
3. Eight minutes: teach one representation boundary using two near cases.
4. Seven minutes: new variants without the original cue.
5. Five minutes: one whole problem and one delayed target to revisit next lesson.

The times are illustrative. The architecture matters more: evidence before explanation, explanation matched to the earliest weak link, and transfer after support.

118. A three-student tutorial can exploit comparison

With three learners, representation becomes visible when different internal models produce different first moves. Alicia, Tricia and Kai Kai can each write the first line privately, then compare.

If all three choose the same valid representation, the tutor can move on. If one differs, ask that learner to explain the cue. If two valid representations exist, compare their affordances rather than forcing consensus. If all three make the same wrong assumption, the shared error becomes a useful teaching target.

The small group should not become public ranking. The pedagogical value lies in comparing representations while preserving individual accountability.

119. Representation and AI-assisted learning

Generative AI can instantly translate a problem into equations, outlines or diagrams. That can be useful for explanation, but it can also outsource the exact representation skill being trained.

A productive workflow is:

1. Learner builds an initial representation.
2. AI or teacher provides an alternative representation or critique.
3. Learner compares the two and identifies the decisive difference.
4. Support is closed.
5. Learner represents a fresh problem independently.

When accuracy matters, verify generated explanations against authoritative sources or the relevant syllabus. Fluent representation is not automatically correct.

120. Frequently asked questions

Is problem representation just reading comprehension?

No. Reading comprehension contributes to representation, especially in worded and source-based tasks, but representation also includes mathematical roles, state changes, diagrams, model assumptions, evidence boundaries, output form and constraints. A student can decode every word and still build the wrong relational model.

Is problem representation the same as method selection?

No. Representation constructs the problem; method selection chooses an approach to that problem. The stages interact. A wrong representation can make the wrong method look correct, while attempting a method can reveal that the representation is incomplete.

Why can a student get easy chapter exercises right but mixed exam questions wrong?

Chapter exercises often announce the relevant representation and method through headings, recent examples and repeated structure. Mixed papers remove some of those cues. The learner must infer the structure from the problem itself. Other factors such as retrieval, fatigue and time can also contribute, so the pattern should be diagnosed rather than assumed.

Should students always draw a diagram?

No. Draw a diagram when it makes a relevant relationship, state or constraint easier to see. A diagram that adds no information can consume time or introduce misleading visual assumptions. The learner should build a repertoire of representations and choose among them.

Are bar models only for primary mathematics?

No representation format belongs permanently to one age. A bar model can remain useful for certain part-whole and comparison relationships, though older learners may prefer equations once symbolic fluency makes them more efficient. The representation should serve the task and learner, not age-based fashion.

Should students underline keywords?

They can, if underlining directs attention to meaningful conditions. Keyword marking becomes weak when it substitutes for understanding the relation. Underline “at least” or “using only Source B” because these change the problem, not because every question must be decorated.

What if the student understands the question but still chooses the wrong method?

Then representation may be adequate and method selection is the more relevant owner. Check whether the learner can state the target, roles and constraints correctly before reteaching interpretation.

What if the student picks the right method but makes many calculation errors?

Move downstream. Execution, fluency, working-memory load, checking or time pressure may be the bottleneck. Representation training should not become the explanation for every mistake.

How do I know whether the representation is actually internal?

Fade the support. Use a fresh problem with changed surface features. Ask the learner to build the diagram, equation, source map or target statement independently. Then test again after delay and within mixed work. Same-day success with a supplied scaffold is weaker evidence.

Can a correct final answer come from a wrong representation?

Yes. A shortcut may work accidentally, two errors may cancel, or the problem may be forgiving. Correctness is important but does not always reveal the quality of the generator. If the route looks fragile, use a nearby case where the wrong representation predicts a different answer.

Should every correct answer be interrogated?

No. That would be inefficient and demoralising. Audit strategically: new concepts, recurring error families, unusual correct answers, high-stakes boundaries and tasks where transfer is uncertain.

Can representation be taught directly?

It can be supported through explicit modelling, comparison, translation, diagrams, self-explanation, near nonexamples and mixed practice. Research suggests that strategy learning and example design can influence representation, but no single routine guarantees transfer for every learner or domain.

Why do students sometimes get worse when representation practice becomes mixed?

Because mixed tasks remove cues that blocked practice supplied. The learner now has to identify the structure and choose the representation. Short-term performance can fall while the practice becomes more diagnostic. The important question is whether later independent transfer improves.

Does a longer question always require a more complex representation?

No. Some long questions contain much contextual detail around a simple relation. Some short questions hide a subtle quantifier or boundary condition. Length and structural complexity are different properties.

What is the fastest representation strategy for exams?

There is no universal fastest format. The efficient representation is the smallest one that preserves the target, decisive relationships and constraints. Experts often compress these features rapidly because they have practised them. Novices may need explicit external representations first.

How can a parent help with word problems without giving the method?

Ask what the problem wants, what each important number represents and what relation connects them. If the child still lacks the concept, teach it. Avoid turning every difficulty into “try harder to think” when knowledge is missing.

How can a teacher tell whether a language error is actually a representation error?

Ask the learner to paraphrase the task and state what evidence the answer must use. If that paraphrase is wrong, fix the task model. If it is accurate but the learner cannot express the reasoning, language production may be the more immediate barrier.

How can representation training avoid overthinking?

By fading. Explicit analysis belongs in teaching and difficult boundary cases. Once common distinctions stabilise, compress them into quick cues. The goal is better first moves, not endless conscious narration.

Can representation change during solving?

Yes. New results can reveal structure, invalidate assumptions or suggest a better model. Flexible problem solving updates representation when evidence changes. The danger is changing randomly rather than because the current model has produced a meaningful signal.

What is the final test of representation quality?

The learner can construct an appropriate model from a fresh problem, use it to guide a valid solution, adapt when a decisive condition changes, and verify the final result against the original task without relying on the teacher’s translation.

121. Evidence notes and further reading

The research cited here supports specific components of the article rather than validating this guide as a single packaged intervention.

Chi, Feltovich and Glaser’s classic expert-novice work is relevant to differences in how problem categories and underlying principles can organise representation. The original research concerned physics problems and should not be converted into a universal claim that all experts or novices behave identically.

Alibali, Phillips and Fischer examined how learning new strategies can be associated with changes in children’s representations of mathematical equivalence problems. The study also illustrates that different strategies can have different representational effects.

The Institute of Education Sciences project on algebraic concepts lists work on problem representation, feature knowledge, self-explanation and example learning in algebra.

Research on correct and incorrect algebra examples provides evidence that example design and explanation can influence conceptual learning. This supports using errors and contrasts carefully rather than assuming that only flawless models teach.

Research on subgoal-labelled worked examples in programming discusses how learners can process surface details instead of solution structure and how subgoal labels are intended to direct attention toward structural components.

Research on heuristic worked examples for reality-based mathematics tasks examines worked examples designed to display strategic and metacognitive aspects of mathematical modelling.

These sources support the article’s concern with representation, structural processing, examples and transfer. The specific ten-part fingerprint, audit code, repair ladder, fictional students and practice bank are eduKateSG teaching constructions. They should remain open to revision when classroom evidence shows that a simpler or different representation works better.

122. The complete problem-representation operating protocol

1. Read for the requested output. What exactly must be produced?
2. Label ambiguous roles. What does each important quantity, statement or source element represent?
3. Identify the relation. What connects the knowns to the target?
4. Protect small constraints. At least, exact, only, without replacement, positive, using the source, within the interval.
5. Track state changes. What changes after each action or stage?
6. Choose a representation. Equation, diagram, table, graph, timeline, source map, causal chain or concise verbal model.
7. Check provenance and evidence scope. What is observed, assumed, inferred, predicted or supplied?
8. Choose the method after the structure is visible.
9. Monitor for representation warnings. Impossible units, unexplained complexity, lost cases, unsupported claims, mismatched output.
10. Return the answer to the original task. Does the final result answer the exact question?
11. After an error, diagnose upstream. Was the task model wrong, or did the failure occur later?
12. Repair with contrasts. Change one decisive feature.
13. Fade the scaffold. The learner should construct the representation independently.
14. Delay and mix. Test whether the distinction survives without worksheet cues.
15. Reintegrate under time. Representation must eventually become fast enough for the actual examination.

123. A final scene: the calculation is still correct

Kai Kai reaches the last third of a mixed examination paper. The question describes two groups and asks for the combined average. Two means are printed in bold. A year ago, those numbers would have become the problem. She would have added them and divided by two.

Now she does something smaller and more important before calculating.

“Average of what?”

The groups contain different numbers of students.

She writes the totals each mean represents, combines them, and divides by the total number of students.

Two questions later, a graph appears. The shape looks familiar. She does not name the graph from the shape. She reads the axes. The vertical variable is velocity, not distance. The meaning of the gradient changes.

Then comes a source question about a proposal. She knows a great deal about the topic. The instruction says “using Source C.” She leaves the outside knowledge where it belongs and works inside the stated evidence boundary.

None of these moments looks dramatic.

There is no spectacular new formula.

No motivational breakthrough.

No perfect model answer copied from a page.

The improvement is quieter.

The problem she solves is now closer to the problem that was actually asked.

Alicia checks the final answer to a reverse percentage by applying the discount forward. Tricia reads the source comparison again and notices that both paragraphs now use the same comparison axis. Kai Kai finishes the graph question and writes the units that make the interpretation visible.

Their techniques were useful before.

The change is that the techniques are now attached to better internal objects.

That is why problem representation belongs before method selection and beside examination technique, checking, error analysis and transfer. A learner cannot reliably solve what they have not yet represented.

The first line of working is not always the beginning of the solution.

Often, the solution begins a few seconds earlier, when the learner decides what problem is actually on the page.

124. Advanced representation: when the same evidence supports several candidate models

As questions become more advanced, the learner is less often choosing between one obviously correct model and one obviously wrong model. Several representations may fit part of the evidence. The task becomes comparative: which model explains more, assumes less, preserves the known constraints and produces testable consequences?

Suppose a data set shows a rise followed by a plateau. A linear model may fit the early region. A saturating model may better capture the whole pattern. A student who represents “trend” as “draw the best straight line” has collapsed model choice into a visual habit. A student who automatically selects a complicated nonlinear model may overfit a pattern that the data do not justify.

At school level, the right response may be qualitative rather than statistical. Ask what features each candidate can and cannot represent. Does the model preserve a maximum? Does it predict impossible negative values? Does it require an assumption the question never supplied? Does it fit the range being studied, even if it would fail outside that range?

The educational value is not to turn every learner into a model-selection specialist. It is to teach that representation is sometimes provisional. A useful model can be locally adequate without being universally true.

125. Advanced representation: the hidden role of scale

Scale changes what patterns are visible. A small variation can look flat on one axis range and dramatic on another. A map can preserve local geometry while distorting global area. A graph can hide or exaggerate change depending on the interval shown.

Learners sometimes represent the visual magnitude as the real magnitude. A steep-looking line is treated as a large effect even when the vertical axis spans a very narrow numerical range. A map distance is read directly without attending to scale. A microscopic image is treated as literal size.

A representation audit should therefore ask what transformation sits between the object and its display. Is the axis truncated? Is the image magnified? Is the map projected? Has a variable been normalised, logged or converted to a percentage?

This is especially important in source evaluation. Visual evidence carries rhetorical force. A graph can be numerically correct and still invite a misleading interpretation if the reader does not reconstruct its scale.

126. Advanced representation: category boundaries can be the problem

Some questions fail because the learner has represented a category too broadly or too narrowly. An organism is classified by one visible feature while ignoring a defining structural condition. A literary text is labelled “persuasive” because it contains an opinion. A mathematical object is treated as a function even though one input maps to more than one output in the representation being used.

Category knowledge is therefore representational knowledge. To know what something is, the learner must know which features are defining, which are typical and which are incidental.

Near nonexamples are powerful here. Ask why one case belongs and a highly similar case does not. The boundary itself becomes the learning target.

This prevents a common exam failure: the student remembers a prototype and forces every similar-looking case into it.

127. Advanced representation: a missing negative case can create false certainty

When learners study only positive examples, they can form a rule that is too broad. If every successful use of a method also shares an irrelevant feature, that feature can become part of the learner’s representation without anyone noticing.

For example, if every proportion problem shown in class contains a table, the learner may associate proportionality with tables rather than with the multiplicative relationship. If every evaluative paragraph uses the word “however,” the learner may treat the transition word as evidence of evaluation.

A negative case breaks the accidental association. Show a table with a non-proportional relationship. Show an evaluative paragraph without the expected transition word. Ask what truly determines category membership.

The instructional question is not “Can the student imitate the example?” It is “Can the student separate essential structure from coincidental packaging?”

128. Advanced representation: a correct model can be too coarse

A representation can be directionally correct and still too coarse for the question. “The population increased” may be true when the task asks whether the rate of increase accelerated or slowed. “The writer is critical” may be true when the question asks how the criticism changes across the passage. “The object moved” may be true when the task asks for displacement, distance and direction.

Coarse representations are especially tempting under time pressure because they reduce complexity. The learner recognises the general idea and begins writing. Marks are then lost because the requested distinction remains unresolved.

A practical check is granularity: does my representation distinguish the categories the question distinguishes? If the prompt separates rate from amount, cause from association, description from evaluation or exact from approximate, the internal model must preserve that difference.

129. Advanced representation: a model can be too detailed

The opposite problem is overrepresentation. The learner introduces every known variable, every possible mechanism and every caveat. Working memory becomes crowded and the central relation disappears.

In mathematics, a simple ratio problem is turned into several algebraic variables. In science, a school-level causal question becomes a catalogue of molecular detail that never returns to the measured outcome. In writing, every counterargument becomes its own branch until the thesis is lost.

Good representation is economical. It preserves what the task requires and no more than the learner needs to reason reliably. Expertise often looks simple because irrelevant detail has been suppressed.

Training should therefore include compression after understanding. Ask the learner to produce the smallest complete model: the fewest variables, the shortest evidence map, the simplest diagram that still protects the decisive constraints.

130. Advanced representation: changing the representation can change the apparent difficulty

A problem can be hard in one representation and straightforward in another. A verbal comparison may become obvious in a table. A geometric relation may become manageable in coordinates. A list of events may become clear on a timeline. A dense passage may reveal structure when claims and evidence are separated.

This does not mean the new representation has solved the problem automatically. It has changed the cognitive workspace.

A powerful training question is: “What is hard to see in the current form?” If the difficulty is order, use a sequence. If it is part-whole structure, use a bar or ratio. If it is simultaneous relationships, use equations or a table. If it is comparison, align dimensions. If it is causal sequence, use a chain.

The representation should be selected for the information it makes cheap to inspect.

131. Advanced representation: verification can expose representation rather than execution

Verification is often introduced as a final stage, but it can function as a representation detector. If the answer has the wrong units, impossible magnitude or incompatible relationship, the learner should not merely fix the last line. They should ask whether the task model was wrong.

Suppose a learner calculates a probability greater than one. The arithmetic may be locally correct. The impossible result signals that events, denominators or overlap were represented incorrectly. Suppose a calculated original price is lower than the discounted price. The sign of the relationship should trigger a representation review.

In writing, a paragraph can be factually true yet fail a simple verification: “If I remove the question and read only this paragraph, can I tell what judgement it is contributing?” If not, the representation of the task may have drifted.

Verification therefore closes the loop: original problem → internal model → solution → comparison with original problem. The last arrow is essential.

132. Advanced representation: learners should know when to stop representing and start solving

Representation can become another form of delay. A perfectionistic learner keeps redrawing diagrams, relabelling variables and listing constraints long after a workable model exists.

The stopping condition is functional: is there now a valid relation between knowns and target? Are the decisive constraints visible? Is the next action justified? If yes, begin execution.

A representation does not need to be beautiful. It needs to be sufficient. The learner can update it later if execution reveals a contradiction or missing condition.

This protects the article’s central argument from becoming its own failure mode. The goal is not to insert a large planning ritual before every answer. The goal is to prevent costly execution on a misconstructed task.

133. A compact teacher decision tree

When a learner gives a wrong answer, the teacher can ask:

1. Could the learner state the exact target? If no, repair task interpretation.
2. Could the learner label the important quantities, source roles or variables? If no, repair roles.
3. Could the learner state the governing relationship or evidence connection? If no, repair structure or prerequisite knowledge.
4. Could the learner preserve the decisive condition? If no, use near contrasts and boundary cases.
5. Was the selected method valid for that representation? If no, move to method selection.
6. Was execution accurate? If no, practise execution.
7. Did the learner verify against the original task? If no, strengthen checking.

Do not insist on moving through the tree mechanically if the evidence is already clear. The purpose is to stop the teacher from jumping to arithmetic, careless mistakes or more practice before checking whether the learner was solving the intended problem.

134. A compact learner decision tree

The learner’s version is shorter:

1. What exactly am I finding, proving, explaining or judging?
2. What does each important piece of information mean?
3. What relation connects it to the target?
4. What condition could change the route?
5. What is the simplest useful representation?
6. Now solve.
7. Does the answer fit the original question?

This sequence can be trained slowly and then compressed. The destination is not a checklist-dependent learner. It is a learner whose internal problem is increasingly faithful to the one on the page.

135. The reason this matters beyond examinations

Problem representation is not only an exam technique. Everyday decisions fail for the same reason. A budget problem is represented as “earn more” when the real issue is irregular cash flow. A timetable problem is represented as “not enough time” when transitions and forgotten tasks create the actual bottleneck. A disagreement is represented as a conflict over values when two people are using the same word differently. A technical fault is represented as one broken component when the interaction between components is responsible.

Education should therefore train people to ask what problem they are actually facing before optimising a solution. Examinations provide a compressed environment in which this capability can be practised with explicit evidence and feedback.

The wider value is intellectual humility. A person who can revise a problem representation can say, “The solution was internally correct, but I was solving the wrong problem.” That is a stronger response than defending the procedure simply because the procedure was executed well.