Find your next learning step
If a calculator answer seems correct but a real result is different, start with the first four chapters. For trends and extrapolation, use the next sections. For data, probability and model uncertainty, read the related chapters before attempting the 22 mixed practice cases.
Two competing models · 22 solved interpretation checks · Mathematics hub
A mathematically correct model can make a poor prediction when its assumptions do not hold, its inputs are unsuitable or its relationship is extended beyond the range in which it has evidence. The correct question is not only ‘Did I calculate accurately?’ but also ‘Why should this formula describe the situation I am predicting?’
Two functions can even fit the same known observations and disagree later. At x = 0 and x = 1, the points (0,2) and (1,4) lie on both y = 2x + 2 and y = 2x² + 2. At x = 2 they predict six and ten. This example shows that fitting the past is not the same as establishing one unique future relationship.
This G1/G2/G3 Mathematics guide develops a common habit of checking conditions, appropriate units, interpretation, domain and evidence. It is not a claim that every extension belongs to all subject levels or that a mathematical forecast is guaranteed. Use the SEAB 2027 SEC subject syllabuses to identify the relevant assessed scope.
- Quick read
- 1. A mathematical model answers a conditional question
- 2. Start by naming what the model leaves out
- 3. Correct arithmetic can coexist with a wrong starting assumption
- 4. Full worked comparison: two models fit the same observations
- 5. Interpolation and extrapolation ask different things
- 6. A trend is not a universal law
- 7. Physical capacity can invalidate an otherwise smooth forecast
- 8. A model depends on correct units
- 9. Linear and exponential growth are different models
- 10. Different averages can tell different stories
- 11. An outlier may be real rather than an error
- 12. A correlation does not establish a cause
- 13. Selection bias can change what the model represents
- 14. A fitted curve can reproduce noise rather than structure
- 15. A model should be checked against new observations
- 16. Absolute error and percentage error serve different roles
- 17. Rounding can disguise a meaningful threshold
- 18. Probability is not a guaranteed event count
- 19. A confidence or uncertainty statement must have a basis
- 20. Models are useful for comparing scenarios
- 21. Sensitivity analysis reveals fragile predictions
- 22. Some predictions fail because conditions change over time
- 23. An exact formula may still describe only an idealised object
- 24. A model can be mathematically valid on the wrong domain
- 25. A contradiction can reveal a failed assumption
- 26. Choose what to measure before judging success
- 27. Separate a proof from evidence of model suitability
- 28. Improve a model by learning from error patterns
- 29. Connect modelling to G1, G2 and G3 learning
- 30. A final checklist for trustworthy conclusions
- 22 worked model checks
- Frequently asked questions
A model is a conditional mathematical description. Its arithmetic can be exact while its predictions are uncertain. Check the target quantity, input measurements, units, assumptions, domain, data selection, possible changes and any independent evidence. Correctly fitting earlier data or computing a precise value is not sufficient to guarantee a future outcome.
Use simple contrasts: a constant-speed journey may include stops; a line and a quadratic may fit the same two observations; expected frequency is not a promised event count; and a capacity limit may prevent a calculated demand from being met. The aim is responsible reasoning, not cynicism about models.
A mathematical model represents selected features of a situation using quantities, rules and relationships. The mathematics can be executed flawlessly while the model’s assumptions are incomplete or the surrounding circumstances change. Therefore ‘the calculation is correct’ and ‘the prediction will be correct’ are separate claims that must be evaluated independently.
Suppose a model assumes a cyclist travels at a constant speed of twelve kilometres per hour. It predicts six kilometres in thirty minutes. The arithmetic is exact under the assumption. Actual travel might be shorter because of stops, slopes or traffic signals. Those factors do not invalidate 12 × 0.5 = 6.
The distinction protects both mathematical confidence and practical honesty. A model can be useful without describing every detail of the world.
Whenever a result concerns a real event, finish the sentence ‘This prediction assumes …’ before claiming how certain it is.
No model includes every detail. A classroom word problem often ignores friction, loading delays, demand variation, measurement errors or changing behaviour so that one relationship can be studied clearly. That simplification is not automatically a flaw. It becomes a problem when an omitted factor materially changes the question being answered.
A uniform-speed model for an MRT journey may use distance divided by average speed. It can teach the relation among distance, time and speed, but it does not by itself include platform waiting, transfers or variable dwell times. Real trip planning requires more information.
The right response is not to reject all simple models. It is to use them inside the conditions they reasonably describe and avoid claims beyond that scope.
Write down which features have deliberately been held constant. Revisit them when a predicted result does not match observation.
Multiplication, percentages and equations do not test the factual truth of the inputs automatically. If an assumption is false, a correctly evaluated formula may give an unsuitable result. Students should learn to check the statement being modelled before celebrating a clean calculator output.
A fictional cafe predicts selling 120 drinks at five dollars each and calculates revenue of 600 dollars. This is the correct product of price and forecast quantity. If only ninety drinks are bought, actual gross revenue is 450 dollars at that unchanged price, not 600.
The mismatch concerns the demand assumption, not the multiplication table. Explaining the type of error is part of mathematical literacy.
Separate ‘forecast quantity’, ‘observed quantity’ and ‘calculated result’ in your working. Do not substitute one for another without changing the meaning.
A limited set of observations may be consistent with more than one mathematical rule. A model cannot be declared uniquely true merely because it reproduces a small dataset. The relevant question is what other evidence or assumptions distinguish the competing relationships and how their predictions perform on later observations.
Suppose measurements at x = 0 and x = 1 give y = 2 and y = 4. The straight line y = 2x + 2 fits both. The quadratic y = 2x² + 2 also fits both. At x = 2, however, the line predicts six while the quadratic predicts ten. The two models agree with all available observations but disagree outside them.
This example shows why matching past values is necessary for some purposes but not sufficient to establish a unique future relationship.
Before extrapolating, ask what evidence supports choosing a linear rather than curved rule. Two data points alone do not answer that question.
| x | Observed y | Line 2x + 2 | Quadratic 2x² + 2 |
|---|---|---|---|
| 0 | 2 | 2 | 2 |
| 1 | 4 | 4 | 4 |
| 2 | Not yet observed | 6 | 10 |
Interpolation estimates within the range of observed inputs, while extrapolation predicts beyond it. Both may use a model, but extrapolation often demands stronger assumptions about whether the relationship continues. Even interpolation is uncertain when the function’s shape is not reliably known.
If a simple temperature table records readings at noon and 2 pm, a linearly interpolated estimate for 1 pm lies between the observations. Using the same line to predict the temperature at midnight is a much larger extension and may be unjustified by the two daytime readings.
The calculation procedure may be identical in both cases, yet the evidential strength of the prediction differs because the input domain has changed.
Mark the smallest and largest observed x-values on a number line. Place the requested prediction point relative to them before using the model.
An increasing sequence can tempt a learner to continue the pattern indefinitely. A mathematical rule inferred from recent data remains a hypothesis unless the problem states it as a given exact relationship. Real systems may reach limits, change regime or respond to new conditions.
A shop records 10, 12, 14 and 16 units in four weeks. Extending the plus-two pattern predicts 18 the following week. That is a reasonable arithmetic continuation under a constant-increment assumption, but actual demand could be 14 or 20.
Within a number-pattern exercise the next term may be defined by the intended rule. Outside that controlled setting, observed values rarely guarantee a unique continuation.
State whether the rule is supplied as a definition or inferred from data. Those are different sources of mathematical certainty.
Mathematical expressions may grow indefinitely, while real processes have maximum capacities. A model that ignores a ceiling can be useful at smaller values and misleading near the limit. Recognising this difference is especially important with straight-line extrapolation and repeated percentage growth.
A library room has only thirty available seats. A linear forecast might predict thirty-four visitors needing a seat. The demand estimate is mathematically calculable, but it does not imply thirty-four seats exist. The practical outcome includes a capacity shortage.
A capacity constraint does not make the forecasted demand number impossible. It distinguishes requests or demand from fulfilled services or physically available resources.
Ask whether the output represents desired quantity, delivered quantity or maximum capacity. Do not merge them into a single number.
A formula can be structurally sensible while its numerical inputs use incompatible units. Such an error may produce a prediction with an unrealistic scale. Converting before substitution is not merely a presentation preference: it preserves the relationship that the formula is supposed to represent.
A speed of 60 kilometres per hour over thirty minutes should use half an hour, giving distance 30 kilometres under constant speed. Entering ’30’ as if it meant thirty hours gives 1,800 kilometres, a correct product for the wrong time unit.
This is a computational modelling error rather than uncertainty about what speed might be observed. It can be prevented before collecting more data.
Label each variable with its unit and test dimensional consistency before calculating. The unit should remain attached through to the answer.
A repeated fixed increase adds the same amount each interval, whereas repeated percentage growth multiplies by the same factor. Confusing them can produce forecasts that diverge rapidly over time. Even when both are calculated correctly, choosing the wrong model changes the implied mechanism.
A quantity starting at 100 and adding ten each year becomes 110, 120 and 130 after three years. A quantity rising ten percent annually becomes 110, 121 and 133.1 after three years. The first uses addition; the second uses compounding.
Neither pattern should be assumed to continue indefinitely in a real population or business simply because it fits a brief observation window.
Read whether the change is ‘ten units’ or ‘ten percent of the current amount’. The reference quantity decides the correct relationship.
A mean summarises a total shared across observations, while a median describes the middle of an ordered list and a mode describes frequency. A prediction based on an average may be poor for an individual case if the observations vary greatly. The statistic can be correctly calculated and still answer a different question from the one being asked.
Travel times of 10, 10, 10, 10 and 60 minutes have mean twenty minutes but median ten. Predicting that every journey takes twenty minutes is not justified. The data contain both frequent short journeys and one much longer journey.
The choice between mean and median depends on the goal, not on which number looks more convenient or optimistic.
Ask whether you need the total per observation, a central typical case or the most frequent outcome. Then choose the appropriate summary.
An extreme observation can change a fitted model or an average. Excluding it may sometimes be justified when there is evidence of measurement failure, but removing a genuine value because it is inconvenient can make the model misleading. Data cleaning and model selection require explicit rules.
A set of five delivery durations includes 12, 13, 13, 14 and 45 minutes. The forty-five-minute record may reflect a real disruption. Deleting it without explanation makes the average look shorter but also conceals an observed possibility.
The correct treatment depends on why the model is used. A routine duration estimate and an emergency planning analysis may reasonably emphasise different features.
Before discarding a value, determine whether it is a recording mistake, a rare genuine event or a consequence of changed conditions.
Two variables may move together because one influences the other, both share another driver, or the relationship is coincidental. A model that predicts one from the other does not by itself explain causality. This is a key limit when mathematical patterns are interpreted as policy or personal advice.
On warmer days a fictional town sells more cold drinks and sees more visitors at outdoor pools. The positive association does not show that buying drinks causes pool visits. Temperature and leisure patterns may affect both counts.
A model can still be useful for forecasting association without proving a causal mechanism. The question is whether a prediction or causal explanation is needed.
Use words such as ‘associated with’ when the evidence establishes correlation only. Reserve ’causes’ for stronger relevant evidence.
A dataset may reflect a particular group rather than the whole population relevant to a decision. Even accurately calculated percentages and trends may not transfer from the observed sample to everyone else. The missing information concerns who was measured and how they were selected.
Suppose a school survey about homework habits receives responses mainly from students who join an optional study club. The average study time of those respondents may be calculated correctly, but it does not necessarily estimate the average for all students at the school.
The relevant issue is not arithmetic; it is whether the data-gathering process supports the intended generalisation.
Write down the population of interest and the group actually observed. Compare them before reporting the result as a population prediction.
With enough flexibility, a mathematical expression can be made to match a limited set of observed points perfectly. This does not guarantee that it will perform well on future inputs. The model may be following accidental measurement fluctuations rather than a stable relationship.
Three points can determine one quadratic polynomial with a nonzero leading coefficient or sometimes a lower-degree special case, yet a different future value may depart sharply from that interpolating curve. Perfect fit to those three points is not proof of long-term accuracy.
This is the intuition behind overfitting. A strong model balances fit with a defensible structure and evidence beyond the examples used to build it.
Ask whether the model was checked on observations that were not used to choose its parameters. Independent checks give more meaningful information.
One way to test predictive usefulness is to compare forecasts with later results not used to build the model. The size and direction of differences reveal whether a model consistently underestimates, overestimates or varies unpredictably. A single successful prediction is not enough to establish lasting reliability.
A model predicts values 6, 8 and 10 for three later cases, but the observed values are 7, 8 and 13. The errors observed minus predicted are +1, 0 and +3. The model fits one case exactly but underpredicts the other two.
These simple residuals do not prove why the model failed, but they locate discrepancies that deserve further investigation.
Record both forecast and observed value in a table. Do not quietly replace an old forecast with the result and then call the forecast accurate.
An absolute difference states how many units a prediction missed by. A percentage error compares that difference with a reference value. The same absolute error may be large relative to a small quantity and minor relative to a large one. Choosing an error measure requires clarity about what matters.
A forecast of 12 when the observed amount is 10 misses by two units, a twenty-percent error relative to the observed amount. A forecast of 102 when the observed amount is 100 also misses by two units, but the relative error is only two percent.
If the observed reference is zero, this particular percentage-error calculation is undefined. A different metric is needed instead of dividing by zero.
Label whether an error is absolute, signed or relative and name the denominator used for any percentage.
Mathematical predictions often generate decimal values, while decisions require whole people, containers or passes. Rounding can change whether a threshold appears to be met. A correct decimal computation is not automatically a suitable practical decision, and rounding rules must be matched to the quantity.
A model predicts 8.2 full-size vans for a simplified number of passengers. A real count cannot be 8.2 vans. If every passenger must be accommodated and the capacity assumptions hold, nine whole vans are required. Rounding to eight would violate the model’s capacity requirement.
The situation is not solved by always rounding upward; other targets such as ‘complete groups only’ may require truncating a quotient.
Describe the action represented by the whole number before selecting a rounding convention.
A probability model describes likelihood under its assumptions. It does not say that a certain number of successes must occur in each short run. Turning an expected frequency into a promise confuses a distributional summary with a deterministic prediction.
If a trial has success probability 0.6 and ten independent trials are carried out, the expected number of successes is six. But obtaining four, seven or even zero is not logically ruled out by the expectation. The result of a particular run remains variable.
Independence is an assumption that needs justification in the model; it does not follow merely from repeating a trial.
Use ‘expected’ or ‘on average across repeated comparable runs’ rather than ‘guaranteed in the next ten cases’.
Predictions become more responsible when their uncertainty is expressed in a way supported by data and a suitable model. Simply adding a plus-or-minus number does not produce a trustworthy interval. Uncertainty may come from measurement, natural variation, model choice or insufficient observations.
A student gives a line-based forecast of twenty units and writes ’20 ± 1′ without any residual analysis, instrument tolerance or stated assumptions. That interval is not justified by the arithmetic alone. It might be much too narrow for the observed variability.
A school-level answer need not always calculate a formal prediction interval, but it should avoid invented precision or unwarranted confidence.
Where uncertainty cannot be quantified from the given information, say what is missing rather than fabricating a margin.
Even if a model is not a perfect forecast, it can help reveal how an output changes when an assumption changes. Scenario analysis can clarify which variables have strong influence and what trade-offs matter. The output is conditional on each scenario rather than a verified statement about future reality.
A simple cost model C = 100 + 5n predicts 150 units of cost at n = 10 and 200 at n = 20. If the per-unit amount were instead six, the corresponding totals would be 160 and 220. The calculations show sensitivity to a parameter, not a guarantee that these costs will be incurred.
Comparing scenarios is valuable precisely because the assumptions are made visible and changeable.
Present the input assumptions next to each output so that readers can see why the cases differ.
A prediction may change considerably when a particular input changes by a small amount. A model is fragile with respect to that input when the output is highly sensitive. Understanding sensitivity helps learners decide which measurements or assumptions need the greatest care.
Suppose C = 100 + 50p, where p is an uncertain proportion. If p changes from 0.20 to 0.21, C changes from 110 to 110.5. If instead C = 100 + 5000p, the same one-percentage-point change increases the output by fifty units.
These models have different sensitivities because the coefficient multiplying p differs. Neither result determines what the true p will be.
Vary one input at a time and record its effect. This is a transparent way to explore how much a forecast relies on uncertain quantities.
A fitted relationship may have worked under an earlier schedule, rule, price or technology and then stop working after that environment changes. Forecast errors can therefore reflect a changing process rather than a mistake in the original fitting step. Models need periodic reassessment when their conditions are not stable.
A bus frequency model built when vehicles leave every ten minutes will misestimate waiting times if the timetable later changes to a twenty-minute interval. The ten-minute calculation may have been accurate for the old policy and inappropriate for the new one.
Updating inputs, assumptions and dates is more important than defending a model that is no longer connected to its context.
Record the period and conditions of the data. When a prediction fails after a change, first check whether the underlying process changed.
Geometry provides useful examples of exact equations with limited real-world scope. The area of a rectangle is exactly length times width for an ideal rectangle. But a real plot may have an irregular boundary or unavailable interior area. Applying the rectangle formula to a roughly rectangular photograph without checking dimensions can produce a poor estimate despite correct multiplication.
A field is described as approximately thirty by twenty metres. The rectangular approximation gives six hundred square metres. If a corner is missing, the actual area is smaller, and the estimate needs the missing region’s dimensions for correction.
The formula itself does not become false. The error lies in representing a nonrectangular region as a perfect rectangle without accounting for what was omitted.
Label an estimate as such and state the geometric approximation. Avoid claiming survey-grade accuracy from a classroom sketch.
Functions and formulas come with domains of permitted inputs. A polynomial may be defined for all real x algebraically, while a context permits only nonnegative time or integer counts. Predictions outside the allowed domain can be mathematically calculated yet meaningless or irrelevant to the application.
Consider a model h(t) = −(t − 3)² + 10, intended for 0 ≤ t ≤ 6. Evaluating it at t = 100 is algebraically possible but produces a negative value vastly outside the intended period. It should not be presented as the real height of an object at that time.
The domain is part of the model’s definition. Extrapolating beyond it changes the question.
Write the allowed x-values alongside the formula. If a requested input lies outside them, state that the model does not support the prediction.
When a model predicts something that cannot occur under the stated definitions, the issue may be an incorrect assumption or inappropriate extension. A negative quantity of physical objects or a probability greater than one is a prompt to inspect the model and input range, not an invitation to report an impossible result without comment.
A linear model P(t) = 0.9 + 0.1t cannot be a valid probability forecast for all nonnegative t because it exceeds one once t is greater than one. The formula can still be a valid arithmetic line, but it cannot represent probability throughout that unrestricted range.
The mismatch reveals a domain or structural limitation rather than a failure of elementary algebra.
Check natural constraints such as nonnegative counts and probabilities between zero and one after computing a model output.
A model can predict the average well but individual outcomes poorly, or estimate a total accurately while misrepresenting its distribution. Predictive performance is inseparable from the target being evaluated. Evaluating the wrong target can make a model look successful while leaving the user’s actual question unanswered.
A model predicts total demand of one hundred items across two shops, and the real combined total is one hundred. But it assigns fifty to each shop while demand was ninety at one and ten at the other. The total forecast is exact and the location-level forecast is poor.
These two judgments are compatible. The model succeeded at one aggregation level and failed at another.
Name the target quantity and grouping before comparing predictions with outcomes. A correct total cannot validate every part of the distribution.
Within mathematics, a proof shows a claim follows from definitions and assumptions. In applied modelling, evidence is needed to support whether those assumptions describe the relevant situation. Students benefit from keeping these two forms of justification separate rather than declaring that an algebraic identity proves a real-world forecast.
The identity (x + 1)² = x² + 2x + 1 holds for every real x by expansion. It does not prove that an observed process follows this squared relationship. Measurements, theory or problem statements are needed to establish the connection.
Mathematical certainty about the equation is not the same thing as empirical certainty about its application.
Ask ‘what did the algebra prove?’ and then ‘why should the real process follow this equation?’ Do not answer the second question with only the first.
When a model repeatedly disagrees with new observations, investigate systematic error rather than adjusting numbers until a preferred result appears. Perhaps an omitted fixed cost, seasonal factor or capacity limit explains the pattern. A responsible improvement changes assumptions openly and checks whether the revised model performs better on independent data.
If a simple line consistently underpredicts outcomes as x increases, a curved relationship may be worth testing. But merely adding a high-degree polynomial to fit the old data perfectly could overfit instead of solving the problem.
The aim is a defensible model that remains useful beyond the examples already known, not the most complicated expression.
Keep a record of the original prediction, the mismatch, the proposed change and the next independent check.
At G1 level, modelling may begin with choosing quantities, percentages, ratios, measurement units and data representations. G2 students develop algebraic relationships and proportional reasoning. G3 learners extend interpretation of graphs, quadratic relationships, rates, probability and mathematical assumptions. The specific assessed details follow the relevant SEC subject syllabus, not a broad informal comparison.
The How Mathematics Works directory connects those explanations across subject levels. The Clementi Mathematics reference describes a first-principles teaching approach: diagnose the actual difficulty, build a clear model, test its conditions and teach students to interpret a calculation rather than merely obtain one.
Learning modelling habits does not guarantee better grades or perfect predictions, but it makes the boundary between known calculations and uncertain conclusions much clearer.
Use examples matched to the student’s current syllabus and then extend only when the foundations are secure.
Before presenting a model-based prediction, name the target, the inputs, the relationship and its assumptions. Check the units and allowed domain. Identify whether the prediction is interpolation or extrapolation, compare with relevant observations when possible, and communicate important limits. This makes the model’s usefulness and uncertainty visible.
A forecast of fifteen boxes from a quantity-and-capacity calculation needs a distinct check from a forecast of fifteen future customer orders. The first may follow exactly from stated capacity and an integer ceiling; the second depends on uncertain demand assumptions. Both can involve the number fifteen without having the same certainty.
A good final statement answers the question and describes the strength of the evidence. Do not hide a conditional model behind a categorical promise.
The best mathematical thinker can say both ‘the calculation is right’ and ‘that is not enough evidence to trust the prediction’ when the situation requires it.
For each example, separate the mathematical computation from the condition that makes it meaningful. Sometimes the result is exact under a stated model, sometimes it is an estimate, and sometimes it is outside the context’s domain. Give the computed answer and the limit of the claim rather than treating those as rival tasks.
Case 1. Constant speed 12 km/h for half an hour.
6 km under the constant-speed assumption.
Actual stops could reduce the travelled distance.
Case 2. Revenue at five dollars per drink for 120 forecast drinks.
$600 forecast revenue.
If only ninety are sold, actual gross revenue at that price is $450.
Case 3. Two data points (0,2), (1,4): must the next y be 6?
No. A line predicts 6 but y = 2x² + 2 predicts 10 at x = 2.
Both models match the two observed points.
Case 4. Week counts 10,12,14,16; next week guaranteed 18?
No. Eighteen follows only if the plus-two rule continues.
An observed trend is not a universal law.
Case 5. 60 km/h for thirty minutes.
30 km under constant speed.
Convert thirty minutes to 0.5 hour.
Case 6. Start 100, add ten each year for three years.
130 after three years.
This is additive, not percentage growth.
Case 7. Start 100, grow ten percent annually for three years.
133.1 after three years.
Compound the current amount each year.
Case 8. Times 10,10,10,10,60: mean.
20 minutes.
The mean does not claim an individual journey will last twenty minutes.
Case 9. Times 10,10,10,10,60: median.
10 minutes.
The median better describes the ordered middle observation.
Case 10. Prediction 12, observed 10: absolute error.
2 units.
Relative error is 20% using observed ten as the base.
Case 11. Prediction 102, observed 100: absolute error.
2 units.
Relative error is 2%, despite the same absolute miss.
Case 12. A model predicts 8.2 buses for all passengers.
Nine whole buses under capacity assumptions.
A count of whole vehicles cannot be 8.2.
Case 13. Success probability 0.6 in ten independent trials.
Expected successes equal 6.
A specific ten-trial run is not guaranteed to produce six.
Case 14. Line-based forecast 20 ± 1 with no uncertainty evidence.
The stated interval is unsupported.
A margin of error needs data or a justified model.
Case 15. Area estimate of a 30 m by 20 m rectangle.
600 m² for an ideal rectangle.
A missing corner would make the actual field area smaller.
Case 16. Probability formula P(t) = 0.9 + 0.1t for all t ≥ 0.
Not a valid unrestricted probability model.
It exceeds one when t > 1.
Case 17. Combined forecast 100, actual total 100, wrong shop split.
Correct total, inaccurate location predictions.
Evaluate the target at the required aggregation level.
Case 18. Observed 7 when predicted 6.
Signed error observed minus predicted is +1.
State the sign convention used for residuals.
Case 19. Compare costs 100 + 5n at n = 10 and n = 20.
150 and 200.
These are conditional scenarios, not guaranteed bills.
Case 20. Quadratic identity matches algebra exactly; does that prove the world follows it?
No. Algebra establishes an identity under its assumptions.
Empirical evidence is needed to connect the formula to a process.
Case 21. Interpolation or extrapolation for x = 8 when observations span 2 to 6?
Extrapolation.
The prediction lies outside observed x-values.
Case 22. A capacity ceiling is thirty seats but estimated demand is thirty-four.
Four people exceed capacity under the model.
Demand and fulfilled capacity are different quantities.
Return to the first case you misinterpreted. Was the problem a wrong unit, an unsupported pattern, a missing capacity constraint, a poor target metric or an unrealistic prediction guarantee? Match the next exercise to that cause. A single correction with a clear explanation can be more useful than repeating a dozen unrelated calculations.
Can a mathematical formula be correct but its prediction wrong?
Yes. The formula may be evaluated correctly while an input, model assumption, domain or real-world condition is unsuitable. Check model validity separately from arithmetic accuracy.
Does matching two observed points prove a linear trend?
No. Multiple functions can agree at the same two points yet give different later values. A linear relationship needs a stated assumption or supporting evidence.
What is the difference between interpolation and extrapolation?
Interpolation estimates within the observed input range. Extrapolation predicts outside it, often requiring stronger assumptions that the relationship continues.
Why is a model’s domain important?
It specifies where the model is intended to apply. A value that is algebraically calculable may be meaningless for the context if its input is outside that domain.
Can probability guarantee a particular number of successes?
Not generally. Expected frequency is an average under a probability model, not a deterministic outcome for each short run.
How do G1, G2 and G3 connect to mathematical modelling?
All levels develop mathematical reasoning and interpretation at different subject demands. Consult the relevant SEAB 2027 SEC syllabuses for level-specific assessed topics.
For subject-specific requirements consult SEAB’s 2027 school-candidate syllabus hub. The G1/G2/G3 Mathematics explainer directory connects related mathematical ideas and the eduKate Clementi Mathematics tutorial reference explains how careful diagnosis and first-principles instruction help learners distinguish a correct calculation from an unsupported conclusion.