Secondary Mathematics Tuition | How to Check Maths Answers: A Complete System for Verification, Independent Checking and Answer Validation

Secondary Mathematics Tuition · Verification, independent checking and answer validation

Clara has finished the question. Her algebra is neat, the calculator agrees with the last line and the final answer is 14. She looks up and asks the question students ask every day: “Is this right?”

Ben does something different. He reads the original question again. The answer was supposed to be an area, but his final line says 14 cm. The number may or may not be numerically useful, but the unit already tells him the solution is not complete.

Mira checks another question by substituting her value into the original equation. It fails. She does not yet know which line is wrong, but she now knows the answer has not earned confidence.

Checking mathematics means gathering evidence that a result still satisfies the original problem. Strong checking does not merely reread the same arithmetic. It tests the answer through constraints, inverse operations, substitution, bounds, units, alternative representations, special cases or genuinely different methods. The best check is the cheapest one that can detect the kind of failure most likely here.

This guide is written for secondary learners across school systems. It owns the integrated learner-facing checking system: when to check, what to check, how strong a check is, how to choose an independent route, how to validate a final answer, and how to allocate checking time under load.

It deliberately does not replace three existing BTT owners. How Mathematical Verification Works | From Answer to Confidence remains the broad mathematical theory owner across proof, modelling, computation and the wider subject. Check an Answer Through an Independent Route remains the focused leaf on independent checking. Singapore School Mathematics: Redundancy, Independent Checks and Mathematical Fault Tolerance remains the Singapore operating-manual owner for fault tolerance and redundancy. This page is the global Secondary Mathematics tuition layer that turns those ideas into one usable checking system.

Clara, Ben and Mira are fictional recurring learners. Their scenes are explanatory examples, not reported student cases or fixed ability types.

Choose a route: if “check your work” feels vague, start with what checking actually means. If you want fast checks, use the cheap-check ladder. If you need a second route, go to independent checking. If you are unsure whether an answer is complete, use the whole-problem check. If examination time is limited, go to risk-weighted checking. For independent practice, use the verification checkpoint.

This guide connects to the wider global lane. Use How to Improve Maths Accuracy for preventing avoidable errors while working; use How to Stop Repeating Math Mistakes when a failed check reveals a recurring error that needs diagnosis and repair; use How to Solve Math Problems when the main issue is route construction rather than verification.

1. Checking is evidence, not repetition

Students are often told, “Check your answer.” Many respond by reading the same working from top to bottom. That can catch a missing sign or copied digit, but it is weak if the original reasoning itself is wrong.

Suppose Clara solves 2(x + 4) = 18 and incorrectly expands it to 2x + 4 = 18. She then solves accurately and gets x = 7. If she rereads the same page without questioning the expansion, every later line still looks internally consistent.

Substitution into the original is stronger. Put x = 7 into 2(x + 4): the left side is 22, not 18. The candidate fails immediately. The check uses the original condition instead of trusting the route that produced the answer.

This gives the first principle: a check is useful to the extent that it can fail differently from the original route.

Repeating 38 × 27 on the same calculator in the same way may reproduce the same key-entry error. Estimating 40 × 30 ≈ 1,200 uses a different signal. If the calculator says 102.6, the order of magnitude check rejects the result quickly.

Checking can test different layers. A working check asks whether a transformation was legal. An answer check asks whether the final result satisfies the original problem. A model check asks whether the chosen representation matches the situation. A completeness check asks whether all required solutions, units or cases have been included.

These layers matter because a solution can be correct at one level and wrong at another. The algebra can be flawless inside the wrong model. The numerical answer can satisfy the last equation but violate a domain restriction. One root can be valid while a second required root is missing.

Take a rectangle problem whose algebra gives widths 5 and −8. The quadratic work may be exact. The whole-problem check rejects −8 because a physical width is positive in the stated model. Verification returns mathematics to meaning.

Ben uses a simple distinction. While solving, he asks, “Is this line still true?” At the end, he asks, “Does this answer still belong to the question?” The first is local; the second is global.

A check does not need to prove everything. Estimation may only show that a decimal placement is plausible. Units may only show that the quantity type is correct. Substitution may show that a candidate solves an equation but not that every possible solution has been found. Each check has a scope.

Good verification therefore asks two questions about the check itself: What error could this catch? and What does it still not prove?

Once learners understand that checking is evidence rather than ritual, the instruction “check your work” becomes actionable.

2. Verification begins before the first exact calculation and continues after the final line

The strongest checking systems do not wait until the end. They create expectations before calculation, local gates during calculation, and final validation afterward.

Before calculating 19% of 310, estimate 20% of 300 ≈ 60. The exact result should be in that neighbourhood. This does not produce the exact answer, but it creates a reference independent of later calculator entry.

During a long geometry question, check a high-leverage intermediate height before using it in two later area calculations. If a right triangle has hypotenuse 13 and one leg 5, the other leg is 12. A value of 18 should be rejected before it spreads.

After solving an equation, substitute the candidate into the original. After a percentage reversal, replay the percentage change forward. After a factorisation, expand. After a probability tree, check that exhaustive outcome probabilities sum to one where appropriate.

This produces a three-part rhythm:

PREDICT → SOLVE WITH LOCAL GATES → VALIDATE GLOBALLY.

Clara predicts sign and scale. Ben checks one or two risky transitions. Mira validates the final result against the starting conditions. None of them redoes every line twice.

Prediction is especially useful because it is generated before the exact answer can bias judgment. Once a calculator displays 742.83, learners are surprisingly good at finding reasons why 742.83 “looks about right”. An earlier estimate of roughly seventy provides a stronger independent reference.

Local gates prevent propagation. If a mixture problem finds a mass of −4 kg, there is little value in using −4 through three more steps and checking only at the end. The domain failure should stop the chain.

Global validation protects against correct local work answering the wrong problem. A learner may calculate radius accurately when the question asks for diameter. Only a return to the target catches that.

The checking system should therefore be designed around transitions. Before a risky transition, know what should stay true. After it, test the preserved property.

For a unit conversion, preserve physical quantity. For algebraic rearrangement, preserve the intended solution relationship. For approximation, preserve enough precision for the final decision. For a graph, preserve scale and domain. For a model, preserve the stated assumptions.

Verification becomes efficient when it is embedded where information is vulnerable rather than added as a large final chore.

3. Start with cheap checks before expensive ones

Checking has a cost. Under ordinary homework, time still matters. Under examination conditions, it matters even more. A useful system begins with low-cost checks and escalates only when the result remains uncertain or the stakes justify stronger evidence.

Use a verification ladder:

1. Target: Did I answer the quantity actually requested?

2. Type: Is the answer the right kind of object—length, area, probability, coordinate, count, interval or proof?

3. Sign: Is positive or negative plausible?

4. Magnitude: Is the scale reasonable?

5. Units: Do the dimensions match?

6. Bounds/domain: Is the value allowed?

7. Original condition: Can I substitute, reverse or replay the relationship?

8. Alternative representation: Does a graph, table or diagram agree?

9. Independent method: Is a second solution route worth the time?

This ladder is not a mandatory sequence for every question. It is a menu ordered roughly from cheap to expensive.

Suppose a probability answer is 1.3. Stop at the bounds check. There is no need to solve the entire tree a second way before knowing the result is impossible.

Suppose an equation solution is x = 4. The value is plausible, sign is allowed and units do not apply. Substitution into the original equation is cheap and strong. Use it.

Suppose a six-mark geometry model ends in a surprising area. A diagram check, unit check and second decomposition may all be worthwhile because the answer carries more downstream importance.

Checking should therefore be risk-weighted. High-mark, multi-step, unusual or domain-sensitive answers deserve stronger evidence than one-step arithmetic already supported by a clear estimate.

Mira does not check every multiplication twice. She checks the steps most likely to create large downstream damage: unit conversions, sign-changing inequalities, values passed into later parts, and final candidate filtering.

A cheap check is valuable even when it cannot certify the answer completely. If it rejects the result, it has saved time. If it passes, the learner decides whether the question deserves another layer.

This is mathematical quality control under limited resources: use inexpensive filters early and reserve expensive redundancy for the places where it buys real confidence.

4. The fastest check is often: what kind of thing should the answer be?

Before looking at the number, look at the object.

A question asks for area. The answer should normally use square units. A question asks for the number of buses. The answer should be a nonnegative whole number under an ordinary counting model. A question asks for a probability. The answer must lie between zero and one.

These type checks are powerful because they do not depend on the same arithmetic used in the solution.

Suppose Ben finds the area of a rectangle and writes 42 cm. The multiplication may be correct, but the answer type is wrong. The quantity should be 42 cm².

Suppose Clara solves a capacity problem and obtains 4.6 vans. The raw quotient may be correct, but “number of vans required” demands a whole number, and the direction of adjustment depends on the capacity condition.

Suppose Mira solves an equation for a physical time and gets −3 seconds. Negative time might be meaningful in some coordinate models, but if the story asks how long after an event something occurs, the context may reject it.

Answer types can be symbolic. “Find all real solutions” asks for a set, not one convenient root. “Find the interval” requires inequalities or interval notation. “Find the coordinates” requires an ordered pair. “Prove” requires a general argument, not a numerical example.

Use the requested object as a final filter. If the work ends in a different object, the question is not finished even when the arithmetic is right.

This check is especially valuable because learners often stop at the first mathematically interesting intermediate result. A radius appears while the question asks for diameter. A gradient appears while the question asks for the line equation. A total cost appears while the question asks for change remaining.

Ben writes a short target at the top of long questions: “all real roots”, “max whole count”, “area cm²”, or “x-coordinate”. The note becomes a checking device at the end.

Type checking is the mathematical equivalent of asking whether the answer is even the right species before judging whether it is the right individual.

5. A final answer is valid only if it satisfies the whole problem, not merely the last equation

Long solutions often compress the original problem into intermediate equations. By the final line, some original conditions may no longer be visible. Whole-problem validation restores them.

Consider a rational equation whose manipulation produces x = 2 and x = 5, but the original denominator is x − 2. The candidate x = 2 cannot be used because the original expression is undefined there. The final polynomial alone does not remember this unless the restriction was carried forward.

Consider a geometry problem that produces two algebraic lengths, one negative. The algebra can be correct while the physical model admits only the positive one.

Consider a probability question “without replacement”. If a later calculation accidentally uses the original denominator on the second draw, the final number may still lie between zero and one. Only returning to the state-change condition exposes the model failure.

Whole-problem checking uses the original givens as a checklist. Which conditions defined the solution? Have all been satisfied?

For the thirteen-box type problem, a proposed pair must satisfy both total number of boxes and total contents. Checking only one condition is insufficient.

For simultaneous equations, the pair must satisfy both original equations. A point on one line is not automatically the intersection.

For optimization, the candidate must lie in the allowed domain and genuinely provide the requested maximum or minimum, not merely be a stationary point or algebraic critical value.

For an inequality problem with integer counts, the real-valued boundary must be translated back into the allowed discrete set.

Mira uses a simple final question: “What conditions would another person use to reject this answer?” She checks those conditions directly.

This is stronger than asking whether the final number feels familiar. The original problem remains the authority. Every transformation was only a route to a candidate.

Whole-problem verification closes the loop from representation back to meaning.

6. Substitution checks whether an equation answer survives the original relation

Substitution is one of the strongest low-cost checks in secondary algebra because it returns the candidate to the equation that started the problem.

Suppose 4x − 7 = 21. Solving gives x = 7. Substitution gives 28 − 7 = 21, so the original relation is satisfied.

Now suppose a learner makes an arithmetic slip and reports x = 6. Substitution gives 24 − 7 = 17, not 21. The check rejects the candidate without needing to inspect every rearrangement first.

Substitution becomes even more valuable after transformations that may create extra candidates. Consider √(x + 6) = x. Squaring gives x + 6 = x², or x² − x − 6 = 0, so the algebraic candidates are x = 3 and x = −2.

Return both candidates to the original equation. For x = 3, √9 = 3, so it works. For x = −2, √4 = 2, not −2. The second candidate is extraneous.

This demonstrates an important principle: the transformed equation can generate candidates; the original equation decides which ones are solutions.

For simultaneous equations, substitute the pair into both originals. Suppose x = 4, y = 3 is proposed for 2x + y = 11 and x − y = 1. The first gives 8 + 3 = 11; the second gives 4 − 3 = 1. Both conditions hold.

Checking only one equation would be incomplete. Many points lie on one line; a simultaneous solution must satisfy both.

For parameter problems, substitution can also reveal excluded cases. If a formula was obtained by dividing by a parameter, test the zero parameter separately because the derived formula may not apply there.

Substitution is not always enough to prove completeness. If you find one value that works in a quadratic equation, you have verified that candidate. You have not proved that there are no other solutions. Completeness requires analysis of the solving method and domain.

Clara uses substitution when the cost is small and the gain is large: equation roots, simultaneous pairs, function values, coordinate points and model conditions.

The narrow skill is simple: put the answer back where the problem began. The broader verification skill is knowing what that check establishes and what it does not.

7. Inverse operations can reconstruct the original quantity

Many school mathematics operations come in inverse pairs. This makes backward reconstruction a natural checking route.

If 84 ÷ 7 = 12, multiply back: 12 × 7 = 84. If a price after adding 18 becomes 73, subtract eighteen from seventy-three to recover fifty-five. If a quantity was multiplied by 1.2, divide the final amount by 1.2 to recover the starting amount.

Inverse checks are useful because they reverse the direction of the original calculation.

Take a percentage increase. Original price 120 increased by 15% gives 138. To check the result, divide 138 by 1.15. The result returns to 120. This is stronger than repeating 0.15 × 120 in exactly the same way.

For ratios, after splitting 64 in the ratio 3:5 to obtain 24 and 40, add the parts to recover 64 and simplify 24:40 to 3:5. Two defining conditions are reconstructed.

For averages, if the mean of twelve values is claimed to be 17, the implied total is 204. If the original total is known, this provides a reverse check.

In geometry, if area of a rectangle is 96 and one side is 12, dividing area by that side should recover the other side as eight. This does not prove every geometric assumption, but it checks the multiplicative relationship.

Be careful with operations that are not one-to-one. Squaring loses sign information. If x² = 25, taking the principal square root gives five, but solving the equation requires x = ±5. Inverse checking must respect the information lost by the forward operation.

Absolute value behaves similarly. |x| = 7 gives two real solutions, seven and negative seven. Reversing a many-to-one operation requires branches.

Rounding cannot usually be inverted to a unique original value. If a number rounds to 12 to the nearest integer, the original lies in an interval rather than being exactly recoverable.

Ben therefore asks whether the forward operation was reversible over the stated domain. If yes, reconstruct. If not, reconstruct a set or range of possibilities rather than forcing one answer.

Inverse checking is strongest when it respects the information content of the operation, not when it simply “does the opposite” mechanically.

8. Units provide an independent check on quantity type

Units are among the cheapest verification channels because they test structure rather than numerical detail.

If distance is divided by time, the result should have speed units. If two lengths are multiplied, area units appear. If area is multiplied by a length, volume units appear.

Suppose a learner calculates the area of a 7 cm by 9 cm rectangle and reports 63 cm. The numerical product is correct, but the unit type is wrong. The result should be 63 cm².

Consider speed. A car travels 150 kilometres in 2.5 hours. Speed = 150/2.5 = 60 kilometres per hour. If the written answer is 60 hours per kilometre, the division relation has been inverted.

Units can also detect incompatible addition. Five metres plus three seconds is not a meaningful ordinary sum because the quantities are of different types.

In scale problems, dimensional checking prevents linear factors from being applied directly to area. If a length scale factor is four, area scale factor is sixteen. A one-dimensional factor cannot remain unchanged after two independent length dimensions are multiplied.

Probability is dimensionless in ordinary school treatment. If a probability calculation ends with centimetres, the model has mixed quantities.

Financial rates and percentages also carry implicit reference units. “5% per year” is not the same as “5% per month”. The time basis is part of the rate.

Mira keeps units through conversions and key intermediate lines. She does not write units beside every pure algebra symbol. The check is used where physical quantity matters.

Dimensional consistency does not prove a formula correct. Many wrong formulas can still have correct units. But dimensional inconsistency proves something is wrong.

That asymmetry makes units a strong early filter: cheap to apply, powerful when they fail.

9. Bounds and magnitude can reject impossible answers before full recomputation

Many problems contain natural limits. Verification becomes faster when those limits are used explicitly.

A probability must lie between zero and one. A weighted mean with positive weights lies between the smallest and largest contributing values. A length in an ordinary geometric model cannot be negative. A count of objects is a nonnegative integer.

Triangle sides satisfy triangle inequalities. If two sides are 5 and 9, the third side x must satisfy 4 < x < 14. A result of twenty is impossible regardless of how neat the algebra looks.

Trigonometric values create bounds. For real angles, sine and cosine lie between −1 and 1. A calculator result sin θ = 1.4 signals an entry, mode or model error.

Order of magnitude is a softer bound. Estimate 51 × 198 as about 50 × 200 = 10,000. An exact result near 1,000 or 100,000 should be investigated.

For percentages, a 12% increase on about 500 should change the value by about sixty, not six hundred. The exact result should remain near the original scale.

For area, a rectangle roughly 20 by 30 has area around 600. A result of 60 may indicate a dropped zero or unit conversion problem.

Ben writes a rough interval before high-risk calculations: “between 50 and 70”, “positive under 1”, “larger than original but under 200”. The interval is not the answer; it is a verification envelope.

Hard bounds come from the mathematics itself. Soft magnitude expectations come from estimation and number sense. Both can be useful.

When a result violates a hard bound, reject it. When it violates a soft expectation, investigate the expectation and the calculation rather than automatically discarding one.

Bounds are powerful because they are often independent of the exact solving route. They test whether the answer belongs to the feasible region at all.

10. An independent route is strong only if it does not share the same hidden failure

Solving a problem twice sounds safe, but independence matters more than repetition.

If you enter the same wrong formula into two calculators, agreement proves little. If two algebraic methods both begin from the same misread equation, they can agree on the wrong problem.

A stronger check changes representation, direction or relationship.

For a quadratic, factorisation can be checked by expansion. Roots can be checked by substitution. A graph can inspect the number and approximate location of real roots. These methods share the same polynomial but expose different failure modes.

For simultaneous equations, solve by elimination and check the pair by substitution. You do not necessarily need a second full solution by substitution; direct replacement into the originals is cheaper and more independent.

For probability, calculate directly and compare with a complement when the complement is structurally simpler. Agreement between two routes increases confidence because they partition the event differently.

For geometry, calculate an area by adding pieces and compare with subtracting a cut-out from a larger shape. If both decompositions use genuinely different partitions, agreement is useful.

For a word problem, algebra can be checked against a reference-case argument or systematic table. The surface methods differ while the underlying constraints remain the same.

Independence is weakened when routes share a copied input error. If the original length was 18 but both methods accidentally use 13, they may agree falsely.

This is why provenance matters: ask what inputs and assumptions the two checks share.

Clara chooses independent checks by asking, “What could be wrong in my first route that this second check would still catch?”

If the answer is “nothing different,” the second route may be redundant in appearance only. If it tests a different relation, representation or direction, it provides stronger evidence.

Independent checking is not about performing the maximum number of methods. It is about selecting a second source of evidence with a genuinely different failure pattern.

11. Switch representations when another form exposes a different property

The same mathematical object can often be represented in several ways. An equation can become a graph. A word problem can become a table. A geometric relationship can become coordinates. A probability event can become a tree or counting argument.

Representation switching is useful for verification because different forms make different errors visible.

Suppose algebra gives the line y = −2x + 7. A graph should fall from left to right and cross the vertical axis at seven. If the drawn graph rises, either the graph or the algebra needs inspection.

Suppose simultaneous equations produce the pair (3,5). Plotting the two lines should show an intersection near that point. A graph is not always the most efficient exact solution route, but it can check whether the algebraic pair is structurally plausible.

A table can verify a linear relationship. If y = 4x + 3, then x-values 0, 1 and 2 should produce y-values 3, 7 and 11. A copied coefficient error becomes visible as the pattern breaks.

Geometry can be checked analytically. If two segments are claimed perpendicular and coordinates are available, gradients or vectors can test the relationship. Conversely, an algebraic coordinate result should still fit the shape and relative position shown by the diagram.

Probability can move between tree and counting representations. A two-draw same-colour probability can be calculated branch by branch or through combinations. Agreement increases confidence because the organisation differs.

Do not confuse representation agreement with proof of every claim. A graph at one viewing window cannot prove no other roots exist outside the window. A finite table cannot prove a general formula for every integer. The check has a scope.

Ben asks what property the second representation exposes. Graphs expose shape, sign and intersections. Tables expose repeated change. Diagrams expose spatial constraints. Algebra exposes exact symbolic relationships.

The strongest switch is purposeful. Do not redraw an equation as a graph merely because graphs exist. Use another representation when it can test a vulnerability in the first.

The learner-facing question is: “What would this answer look like in another mathematical language?”

If the same relationship survives translation, confidence rises. If it does not, the disagreement becomes diagnostic evidence.

12. Special and boundary cases stress-test formulas and general claims

A formula can work on ordinary values and fail at an edge. Verification becomes stronger when a few carefully chosen cases push the structure.

Suppose a proposed formula for the cost of a service is C = 5d + 12, where d is distance. Test d = 0. The formula gives twelve, which matches the idea of a fixed charge. If a proposed formula were 5(d + 12), the zero-distance cost would be sixty, revealing a modelling problem.

Zero is often useful because it removes terms. One is useful because it simplifies powers and products. Equal values can test symmetry. Maximum and minimum allowed values test endpoints.

For a sequence formula, check n = 1 against the stated first term. For a geometric formula, check a square or other familiar special shape when it belongs to the family. For probability, check extreme probabilities 0 or 1 when the model allows them.

Boundary cases deserve special attention in inequalities and domains. If the solution is x < 5, test a value below five and consider whether x = 5 should be included. If a denominator restriction says x ≠ 2, inspect what happens as x approaches or equals two.

For piecewise rules, check the boundary between pieces. If one formula applies for t ≤ 3 and another for t > 3, evaluate the behaviour near t = 3. A gap or double-counted condition may appear.

In geometry, degenerate cases can reveal hidden assumptions. A triangle condition may fail as one side approaches the sum of the other two. A rectangle formula should reduce sensibly when the rectangle becomes a square.

Clara does not test ten random values. She chooses cases that simplify the structure or push a condition. The goal is falsification power, not volume.

Passing several special cases does not prove a universal formula. Failing one valid case disproves it.

This asymmetry makes boundary and special-case checking efficient. A small number of well-chosen inputs can expose broad structural errors quickly.

Use them especially when the result is a general formula, parameter relation, recurrence or model rather than one isolated numerical answer.

13. Counterexamples are the fastest check for universal claims

If a statement claims something is true for every member of a class, one valid counterexample is enough to show it is false.

Consider the claim: “If x² > 4, then x > 2.” Try x = −3. Then x² = 9 > 4, but x is not greater than two. The universal claim fails.

Or consider: “The sum of two irrational numbers is irrational.” Take √2 and −√2. Their sum is zero, which is rational. One allowed example destroys the claim.

Counterexample search is a form of verification because it tests whether the scope of a rule is too broad.

Use boundary values, zero, one, negatives where allowed, equal values, symmetry cases and familiar special shapes as candidate counterexamples. These often expose hidden assumptions.

Make sure the example belongs to the stated domain. If the claim concerns positive integers, a negative decimal cannot refute it.

After finding a counterexample, repair the statement. “When denominator increases, fraction decreases” becomes more accurate when restricted to a fixed positive numerator and positive denominators.

Mira uses counterexamples to check memorised slogans. “Division makes smaller” fails for 6 ÷ 1/2 = 12. “A straight line means direct proportion” fails for y = 3x + 5.

This prevents shortcuts from expanding beyond the conditions that make them useful.

A counterexample can also verify that a proposed proof strategy is insufficient. If the argument would imply a known false statement, a hidden assumption has entered.

For general mathematical claims, checking through examples is not the same as proof. But searching for disproof is often the fastest first test.

The verification habit is to ask: “What is the simplest allowed case that would break this if it were false?”

14. Verification must ask whether every required answer or case has been found

A candidate can be correct and the solution still be incomplete.

From x² = 49, x = 7 is valid. It is not the complete real solution set because x = −7 also works.

For |x − 3| = 5, the solutions are x = 8 and x = −2. Reporting one branch verifies one candidate but misses the other.

For a trigonometric equation on a specified interval, a calculator may display one angle while periodicity creates others. The interval and function structure decide completeness.

For an optimization problem, finding a stationary point is not enough if endpoints can produce larger or smaller values. Check the whole allowed domain.

For a probability event split into cases, confirm the cases are exhaustive and not accidentally overlapping. Missing one branch produces an incomplete probability; overlapping branches can double-count.

For a proof by cases, every possible case must belong to one branch. Leaving a boundary out creates a logical gap.

Ben uses a completeness question at the end: “What mechanism in my method tells me there cannot be another answer?”

Factorising a quadratic into two linear factors provides a complete root mechanism over the relevant domain. Systematically enumerating every positive integer pair in a finite range can be complete. Checking one graph window may not be.

Completeness is therefore tied to the solving method. A valid candidate is evidence of existence; a complete method provides evidence that the set is finished.

This distinction matters whenever the question says “all”, “every”, “maximum”, “minimum”, “number of solutions”, “list the values” or asks for a proof.

Verification is not complete until the answer set itself has earned confidence.

15. Check whether exactness, approximation and reported precision match the task

A result can be numerically close and still be reported inaccurately.

Suppose the exact value is 7π. Writing 21.99 may be appropriate if the question requests a decimal to two decimal places. It is not equivalent as an exact mathematical statement.

Use ≈ when representing an approximation. Preserve = for exact equality unless the context has clearly defined rounded values.

Premature rounding can alter later results. If a length is √52 ≈ 7.211102 and a later threshold is 7.21, rounding to 7.2 too early can change a yes/no conclusion.

Keep exact forms or extra precision through intermediate stages when practical. Round at the final requested stage.

Check whether the instruction asks for decimal places, significant figures, nearest unit, exact form or an interval. These are different reporting contracts.

For 0.006784 to three significant figures, the answer is 0.00678. To three decimal places it is 0.007. The same underlying number produces different outputs because the instructions differ.

Data precision matters too. A calculator may show ten digits, but measurements given only to the nearest unit do not suddenly become known to ten decimal places.

Bounds can be the correct verification language for rounded data. A length recorded as twelve to the nearest unit represents a range under the usual rounding convention, not a single exact original value.

Clara marks the first moment approximation enters the solution. That makes later rounding traceable.

Precision checking therefore asks two things: did the computation preserve enough information, and does the final representation claim no more precision than the task supports?

A trustworthy answer has the right value and the right epistemic status: exact when exact, approximate when approximate.

16. Algebra checking: verify structure, candidates and reversibility

Algebra gives learners many natural checking routes because equations, expressions and functions can often be reversed or substituted.

Linear equations. Solve, then substitute. If 7x − 4 = 31 gives x = 5, check 35 − 4 = 31. This is cheap and direct.

Factorisation. Expand the factors. If x² + x − 12 is written as (x + 4)(x − 3), expansion returns x² + x − 12. This checks the factorisation rather than the roots.

Quadratic roots. Substitute roots into the original polynomial. If the method could have missed or added roots, also check completeness.

Rational expressions. Keep restrictions. Simplifying (x² − 9)/(x − 3) to x + 3 requires x ≠ 3. A check at x = 3 must refer back to the original domain.

Square-root equations. Substitute every algebraic candidate into the original because squaring can introduce extraneous solutions.

Inequalities. Test one value from the claimed solution region and one outside. This does not replace the derivation, but it can catch a reversed sign or wrong boundary.

Functions. Verify function inputs carefully. If f(x) = 3x − 2, then f(4) should equal ten. If an expression for f(x + 1) is derived, substitute a simple x-value and compare both routes.

Identities. Numerical tests can falsify an incorrect identity but cannot prove a universal identity. After a few spot checks, use algebraic reasoning or proof as required.

Ben uses different checks for different algebraic objects. Equation candidates get substitution. Factorisations get expansion. General identities get structure and possible counterexample search.

The checking method should match the claim. There is no single universal algebra check.

When a check fails, move into the error-analysis lane: locate the first invalid transformation rather than patching only the final line.

17. Geometry checking: test the whole figure, not one local calculation

Geometry contains networks of relationships that can cross-check one another: angle sums, congruence, similarity, perpendicularity, parallelism, coordinate relationships, area and perimeter.

Suppose a triangle has angles claimed to be 52°, 68° and 70°. The total is 190°, so the set cannot describe an ordinary Euclidean triangle. A global figure constraint rejects the answer.

Suppose a right-triangle calculation gives a hypotenuse shorter than one of the legs. The result is impossible before exact recalculation begins.

For an isosceles triangle, equal sides should correspond to equal opposite angles. If the computed angles violate that symmetry, inspect the route.

For similar figures, corresponding side ratios should agree. If one pair gives scale factor 2 and another gives 2.5, either the correspondence or the calculation is wrong.

For a rectangle with area 48 and side length 6, the other side should be 8. Then perimeter should be 28. Area and perimeter provide distinct checks on the same dimensions.

Composite figures benefit from alternative decomposition. Calculate area by adding pieces, then compare with a larger bounding shape minus cut-outs. Agreement between independent partitions is strong evidence.

Coordinate geometry adds analytical checks. Perpendicular finite nonvertical lines have gradient product −1. Distance formulas can check claimed equal lengths. Midpoints can check bisection.

But coordinate checks themselves depend on copied coordinates and correct formulas. A diagram can provide a qualitative signal if the analytic result places a point in an impossible quadrant.

Mira asks whether the answer fits the whole figure. A local equation may be satisfied while another stated relationship is violated.

Geometry verification therefore moves between local calculation and global consistency. A trustworthy result belongs everywhere in the figure, not just in the line that produced it.

18. Ratio and percentage checking: reconstruct the base, total and direction of change

Ratio and percentage errors often come from reference quantities. Their strongest checks rebuild the original relationship.

Suppose 64 is split in the ratio 3:5 to give 24 and 40. Check the total: 24 + 40 = 64. Check the ratio: 24:40 simplifies to 3:5. Both defining conditions hold.

Suppose an original price 150 is discounted by 20% to 120. Check forward: 20% of 150 is 30; subtracting gives 120. Check backward: 120/0.8 = 150.

Suppose a learner claims that increasing 80 by 25% gives 90. A magnitude check already suggests the increase should be twenty, not ten. Reconstruct the percentage: 90/80 = 1.125, which is only a 12.5% increase.

Compound percentage changes should be replayed in order. A 10% increase followed by a 10% decrease does not return to the original because the second percentage acts on a different base. Starting at 100 gives 110, then 99.

For direct proportion y = kx, check whether the ratio y/x is constant for nonzero x. A straight line with nonzero intercept fails that test even if it looks visually simple.

For scale drawings, reverse the scale. If 6 cm on a map at 1:25,000 represents 150,000 cm = 1.5 km in reality, converting back should return six map centimetres.

Clara uses direction as a quick check. A discount should reduce the price. A positive percentage increase should raise it. A ratio part should not exceed the whole total unless the context defines signed or nonstandard quantities.

Strong percentage checking is about the base. Strong ratio checking is about the part structure. Both return the numerical answer to the relationship that gave it meaning.

19. Probability checking: use bounds, totals, complements and alternative counting

Probability has built-in verification because events have hard bounds and often admit complementary or counting routes.

The first check is range: 0 ≤ P(A) ≤ 1. A result of 1.2 is impossible. A result of 0.4 is possible but not necessarily correct.

If a set of mutually exclusive outcomes is exhaustive, their probabilities should sum to one. A tree whose terminal branches total 0.82 has missing or miscalculated probability mass.

Complements are powerful. If P(at least one success) is calculated directly, compare with 1 − P(no successes) when the complement is simpler.

For two fair dice, P(at least one six) can be checked as 1 − (5/6)² = 11/36. Direct counting of the 36 ordered outcomes also gives eleven successful outcomes.

Without-replacement models should show changing denominators. If a bag begins with seven counters, the second draw comes from six unless replacement occurs. A repeated denominator of seven should trigger inspection.

Conditional probabilities require the correct conditioning set. A denominator should represent the world in which the condition is already known.

Combinatorial probability can be checked by comparing ordered and unordered counting when both are manageable. If both routes use the same event correctly, agreement provides strong evidence.

Ben also checks direction. Removing a red counter should usually reduce the chance of another red draw when there is no replacement. If the computed probability increases, the state update deserves inspection.

Probability verification therefore has layers: range, state, event definition, total probability, complement, and alternative count.

Because wrong probability answers often look numerically plausible, the event and denominator deserve as much checking as the final decimal.

20. Statistics checking: verify weighting, range, summaries and what the data can actually support

Statistical calculations can be arithmetically perfect while the interpretation is wrong. Verification must test both the number and the claim made from it.

For a mean, check that it lies between the minimum and maximum observations. A mean outside the observed range is impossible for ordinary arithmetic mean with positive weights.

When combining group means, reconstruct totals. Ten observations with mean 12 contribute 120; twenty with mean 18 contribute 360. Combined mean is 480/30 = 16. Directly averaging 12 and 18 would incorrectly give equal group weight.

For a median, sort order matters. A correct sum or mean does not verify the middle position.

For frequency data, frequencies should sum to the total count. Relative frequencies should sum to one or approximately one after rounding when categories are exhaustive and non-overlapping.

Graphs need scale checking. A truncated vertical axis can exaggerate differences even when every plotted point is numerically correct.

Check whether the summary can answer the question. A mean alone does not determine range, median or individual values. If the information is insufficient, the correct conclusion may be “cannot be determined uniquely”.

For samples and models, calculation is only one layer. A perfectly computed sample statistic does not automatically justify a broad causal claim or population conclusion.

Mira asks, “What did the calculation actually establish?” This prevents a numerical result from being given more interpretive power than the data support.

Verification in statistics therefore includes arithmetic, weighting, scale, sufficiency and inference.

A trustworthy statistical answer is not merely a correct number. It is a claim whose strength matches the evidence available.

21. Calculator and digital checks are useful only when the tool receives the intended mathematics

A calculator can verify arithmetic, test candidates, compare values and graph relationships. It can also reproduce a wrong expression perfectly.

Suppose the intended calculation is (18 + 6)/3. If the learner enters 18 + 6 ÷ 3, the display shows twenty. Repeating the same entry confirms only the same misrepresentation.

A tool-based verification should therefore begin with expression fidelity. Compare the written mathematical expression with the machine input.

Check brackets, angle mode, stored values and exact-versus-decimal state. In trigonometry, a degree-mode question entered in radians can produce a numerical result that looks plausible but describes a different angle convention.

Estimate before relying on the display. If a product should be near one thousand, a calculator answer near ten thousand deserves investigation regardless of how many decimal places are shown.

Use a calculator to substitute roots into an original equation, but make sure the original equation is entered accurately. The machine cannot detect that you copied x − 3 as x + 3.

Graphing tools are useful for shape, intersections and sign. They are limited by viewing window and numerical resolution. A hidden root outside the window is not absent from the function.

Spreadsheets can check repeated calculations, but copied formulas can spread a wrong reference across hundreds of rows. Hand-check one simple row or known case before trusting the whole column.

General AI tools can propose explanations or alternative methods. Their outputs should be treated as mathematical claims that require the same checks: domain, arithmetic, transformation validity, completeness and fit to the original task.

Clara uses tools as accelerators, not authorities. The tool reduces computation cost; mathematics still decides whether the expression, conditions and interpretation are correct.

A digital check is strongest when it uses a different representation or independent recomputation, not when it merely repeats the same mistaken input faster.

The verification principle is simple: first verify what you asked the tool to do; then verify what the tool returned.

22. Long problems should be checked at high-leverage intermediate nodes

In a long solution, one early error can feed several later steps. Waiting until the final answer to check everything makes the search expensive.

Instead, identify high-leverage intermediate results: values that will be reused, decisions that change the route, or transformations that could alter the domain.

Suppose a multi-stage geometry problem first finds a height, then uses it for area, then uses area inside a cost model. The height is a high-leverage node. Check it before the later stages depend on it.

Suppose a percentage problem finds the original price, then applies tax and delivery. If the recovered original price is wrong, every later amount inherits the error. Replay the discount once before continuing.

Suppose a probability tree branches after the first event. Check the updated state at the branch point. A wrong denominator there corrupts every downstream probability.

Use labels so checkpoints remain interpretable: “height = 12 cm”, “original price = 150”, “remaining red = 4/7”. A bare number is easier to misuse.

Checkpoints should be chosen by dependency. A value used once in an easy final addition may need little checking. A value reused four times deserves stronger evidence.

Ben thinks of the solution as a dependency graph. Some nodes have many arrows leaving them. Those nodes deserve early verification because their failure has a large blast radius.

This is the student-facing version of fault tolerance. It does not require industrial terminology or multiple full solutions. It means protecting the parts of the route whose errors would spread furthest.

Local verification also makes later diagnosis faster. If the height was already checked and the final area is wrong, the search can begin after the height rather than at line one.

Do not place a formal check after every line. Choose the places where uncertainty, downstream impact or prior error history makes the check worthwhile.

Strong long-form mathematics alternates progress and validation rather than running an unchecked chain from top to bottom.

23. When two checks disagree, do not average them—trace the divergence

Independent checks are valuable partly because they can disagree. A discrepancy is information.

Suppose algebra gives a root near four while a graph suggests the intersection is near six. Do not average to five. At least one representation, input, scale or calculation is wrong.

Trace each route back to the last point of agreement. Were the same coefficients used? Was the graph window correct? Was an equation copied with the same sign? Did one method approximate early?

Suppose an area-by-addition method gives 84 while area-by-subtraction gives 80. Compare the partitions. Is there overlap? Was a cut-out subtracted twice? Did both routes use the same total dimensions?

Suppose a probability complement gives 7/8 while direct counting gives 3/4. Define the event in words for both routes. One may be calculating “at least one head”; the other may be calculating “exactly one head”.

Disagreement narrows the problem. Two routes that share some steps but diverge later point toward the divergence region.

Clara keeps both routes visible rather than erasing one immediately. She marks the first place where the mathematical claims differ. That location becomes the diagnostic target.

Do not assume the longer or more sophisticated method is correct. A short estimate can expose an error in an advanced symbolic derivation. A calculator can expose arithmetic in a hand method. Either route can fail.

When both routes depend on the same original model, also consider whether the model itself is wrong. Agreement between two calculations cannot validate a misread question.

The correct response to inconsistent evidence is investigation, not compromise.

Verification becomes powerful when disagreement is treated as a map to the error rather than as an inconvenience to be smoothed away.

24. Under time pressure, allocate checking by risk, value and error probability

An examination does not provide unlimited time for verification. The best strategy is not “check everything equally”. It is to allocate checking where it has the highest expected value.

Prioritise questions with many marks, many steps, unusual answers, difficult domain conditions, values close to thresholds, or calculations feeding later parts.

A one-mark integer addition may need only a quick glance. A six-mark modelling question may justify target, units, bounds and one independent check.

Check recently repaired error patterns. If sign-changing inequalities have been unstable in practice, a five-second inequality-direction check may be more valuable than redoing a secure multiplication elsewhere.

Check answers that feel surprising—but use evidence, not anxiety. A negative gradient in a decreasing graph may be entirely correct. A negative physical length is not.

Check boundary decisions. Minimum whole-number counts, rounding thresholds, excluded endpoints and final answer forms can lose marks after otherwise correct working.

Use the cheapest strong check. For an equation, substitution may take seconds. For a probability, a complement may be quicker than recomputing every branch. For a geometry result, a unit and rough-scale check may be enough.

Do not spend three minutes constructing a second complete solution to a one-mark question while leaving later high-value work untouched.

Mira divides checking into three levels. Scan: sign, unit, magnitude, target. Verify: substitution, inverse or local structural check. Redundant route: second method or representation for high-value or suspicious results.

The exact timing strategy depends on the assessment format, so use current official rules and the appropriate examination owner. The transferable principle is resource allocation.

Good examination checking is selective confidence-building, not maximum repetition.

The learner should finish the paper knowing that the most dangerous answers received the strongest available evidence.

25. Tutors and parents should teach the learner how to verify before supplying external confirmation

A learner often asks, “Is this right?” The adult can answer immediately—or turn the moment into verification practice.

Ask, “How could you find out?” This returns authority to the mathematics.

If the question is an equation, suggest substitution rather than announcing correctness. If it is an area, ask about units and scale. If it is a percentage, ask what the base is and whether the direction of change makes sense.

Use the smallest prompt that restores checking. “What could you substitute?” is better than giving the entire verification route when the learner can supply it.

Parents do not need to know the answer. They can ask whether the child has an independent check, whether the units match, or whether the result is in a plausible range.

Tutors should sometimes leave correct work unconfirmed for a moment so the learner can gather their own evidence. Permanent external confirmation can prevent internal verification habits from developing.

Record support honestly. “Verified after tutor said to substitute” is progress, but it is not the same as spontaneous checking.

Move through levels: adult announces error → adult asks for a check → learner selects the check → learner checks spontaneously before asking.

Ben’s tutor stops answering “yes” immediately. Ben increasingly arrives with: “I substituted it into both equations and it works, but I am not sure whether I found all the solutions.” That is a better mathematical question.

Verification also protects the tutoring relationship from becoming answer-dependent. The goal is not a student who asks the tutor to certify every page. It is a student who knows which mathematical evidence is available.

Adult support succeeds when the learner’s question changes from “Is this right?” to “What check would give me enough confidence here?”

26. Verification checkpoint: twenty-four answers to validate, not merely recalculate

The following original tasks test checking decisions. They are not a standardised test and have no validated cut score. For each task, solve if needed, choose an appropriate check, and explain what that check establishes. The worked explanations show both the answer and the verification route.

Task 1 — Linear equation substitution

Solve 5x − 8 = 27 and verify the result.

Worked explanation: 5x = 35, so x = 7. Substitute into the original: 35 − 8 = 27. The check confirms that the candidate satisfies the original equation. It does not by itself teach the solving method, but it gives strong evidence that this candidate is valid.

Task 2 — Factorisation by expansion

A learner writes x² − x − 20 = (x − 5)(x + 4). Verify the factorisation.

Worked explanation: Expand: x² + 4x − 5x − 20 = x² − x − 20. The original expression is recovered. Expansion is a natural inverse structural check for factorisation.

Task 3 — Reverse percentage

A sale price of 136 units follows a 15% discount. Find and verify the original price.

Worked explanation: 136 = 0.85p, so p = 160. Verify forward: 15% of 160 is 24; 160 − 24 = 136. This check reconstructs the transaction and confirms the correct percentage base.

Task 4 — Ratio split

Split 84 in the ratio 2:5 and verify both conditions.

Worked explanation: Seven parts make 84, so one part is 12. The parts are 24 and 60. Check total: 24 + 60 = 84. Check ratio: 24:60 simplifies to 2:5. Both defining conditions are satisfied.

Task 5 — Probability bound

A calculation gives probability 17/12. What does verification tell you immediately?

Worked explanation: The answer is impossible under ordinary probability because probability must lie between zero and one. The bound check rejects the result before any full recomputation. It does not reveal the exact error location.

Task 6 — Weighted mean

A group of 8 has mean 14 and a group of 12 has mean 20. Find the combined mean and give one check.

Worked explanation: Totals are 112 and 240, so combined mean is 352/20 = 17.6. Check that 17.6 lies between fourteen and twenty and closer to twenty because the larger group has the higher mean. The range-and-weight check supports the result.

Task 7 — Triangle side bound

Two sides of a triangle are 7 and 11. A learner obtains the third side as 20. Verify.

Worked explanation: Triangle inequality requires the third side to be less than 18 and greater than 4. Twenty is impossible. This hard geometric bound rejects the answer without needing to inspect the original calculation first.

Task 8 — Area unit check

A rectangle measuring 9 cm by 4 cm is reported to have area 36 cm. What is wrong?

Worked explanation: The numerical value is correct, but the unit type is wrong. Area should be 36 cm². Unit verification catches a completion error that arithmetic checking would not.

Task 9 — Minimum whole-number count

Fifty people require vehicles holding at most twelve each. A learner writes 50/12 = 4.166… and answers four vehicles. Verify.

Worked explanation: Four vehicles hold only forty-eight people, so the answer fails the capacity condition. Five vehicles are required. The strong check is not ordinary rounding; it is testing sufficiency at the boundary.

Task 10 — Radical equation candidates

Solve √(x + 2) = x and verify all candidates.

Worked explanation: Squaring gives x + 2 = x², so x² − x − 2 = 0, hence x = 2 or −1. Substitute into the original. x = 2 works: √4 = 2. x = −1 fails: √1 = 1, not −1. Final solution is x = 2. Verification removes the extraneous candidate.

Task 11 — Direct proportion check

Is y = 6x + 2 directly proportional to x?

Worked explanation: No. A direct-proportion relation has the form y = kx under the standard school model and passes through the origin. At x = 0, this relation gives y = 2. The zero-input boundary case exposes the fixed offset.

Task 12 — Calculator grouping

The intended value is (24 − 6)/(3 + 3). A calculator displays 19 after the user enters 24 − 6 ÷ 3 + 3. Verify the input.

Worked explanation: The intended expression is 18/6 = 3. The entered expression is different because grouping was lost. Verification must begin by checking expression fidelity before trusting the display. Repeating the same wrong input would not be an independent check.

Task 13 — Graph and equation agreement

The equation y = −3x + 6 is graphed as a line rising from left to right. What does verification tell you?

Worked explanation: The algebraic gradient is negative three, so the graph should fall as x increases. The visual representation disagrees with the equation. Either the graph was drawn incorrectly or the equation was copied incorrectly. Direction provides a low-cost cross-representation check.

Task 14 — Similar-figure area ratio

Two similar figures have corresponding side lengths in ratio 2:5. A learner says the area ratio is 2:5. Verify.

Worked explanation: Area scales with the square of the linear factor, so the area ratio is 4:25. Dimensional reasoning rejects the linear ratio for a two-dimensional quantity.

Task 15 — Simultaneous equations

A learner proposes x = 3, y = 4 for x + y = 7 and 2x − y = 1. Verify.

Worked explanation: First equation: 3 + 4 = 7, true. Second: 6 − 4 = 2, not 1. The pair is not a simultaneous solution. Checking both originals is essential.

Task 16 — Exact versus approximate

A learner writes √2 = 1.41. What is the verification issue?

Worked explanation: 1.41 is an approximation, not exact equality. A correct statement is √2 ≈ 1.41 to two decimal places. The check concerns epistemic status and notation, not the closeness of the decimal alone.

Task 17 — Probability complement

Three fair coins are tossed. Verify P(at least one head) = 7/8 using a complement.

Worked explanation: P(no heads) = P(TTT) = 1/8. Therefore P(at least one head) = 1 − 1/8 = 7/8. The complement route is independent of listing all seven successful outcomes and gives strong agreement.

Task 18 — Data sufficiency

A set of five numbers has mean 12. A learner claims its median is also 12. Verify the claim.

Worked explanation: The claim is not determined by the mean. For example, 10,11,12,13,14 has median twelve, but 0,1,5,20,34 also has total sixty and mean twelve while its median is five. A counterexample shows the claim is unjustified.

Task 19 — Inequality check

A learner solves −2x ≥ 8 as x ≥ −4. Verify.

Worked explanation: Dividing by negative two reverses the inequality, so x ≤ −4. Test x = −5 in the original: 10 ≥ 8, true. Test x = 0: false. Test values expose the wrong direction.

Task 20 — Boundary in a piecewise model

A parking model charges 5 units for t ≤ 2 hours and 5 + 3(t − 2) for t > 2. Verify the cost at t = 2 and immediately above t = 2.

Worked explanation: At t = 2, the first rule gives 5. The second rule approaches 5 as t approaches two from above. The pieces meet without a jump under this model. Boundary checking confirms consistency at the transition.

Task 21 — Composite area by two routes

A 10-by-8 rectangle has a 3-by-2 corner cut out. Find the remaining area and verify using two decompositions.

Worked explanation: Subtraction route: 80 − 6 = 74. Addition route: decompose the remaining L-shape into, for example, a 7-by-8 rectangle and a 3-by-6 rectangle: 56 + 18 = 74. Independent decompositions agree.

Task 22 — Model boundary

A tank begins with 30 litres and fills at 4 litres per minute until its 70-litre capacity is reached. A learner uses V = 30 + 4t and reports V = 90 at t = 15. Verify.

Worked explanation: The tank reaches capacity when 30 + 4t = 70, so t = 10 minutes. The original linear retained-volume model cannot continue unchanged beyond that boundary without information about overflow or stopping. Ninety litres is not a valid retained volume in a 70-litre tank.

Task 23 — A universal claim

A learner claims, “If ab is positive, then a and b are both positive.” Verify or refute.

Worked explanation: Refute with a counterexample: a = −2 and b = −3 give ab = 6 positive, yet both factors are negative. A repaired statement is that a and b have the same sign when their nonzero product is positive.

Task 24 — Independent estimate

A calculator gives 49.8 × 19.7 = 98.106. Verify.

Worked explanation: Estimate 50 × 20 ≈ 1,000. The displayed result is an order of magnitude too small. The correct product is 981.06. Estimation catches the misplaced decimal without repeating the same calculator entry.

After the checkpoint, sort any failures by verification job: target/type, substitution, inverse relation, unit, bound, completeness, representation, precision, model boundary or independent route. Do not create one generic “check more” goal. Choose the checking method that would have caught the actual failure.

27. Frequently asked questions about checking maths answers

Should I check every answer?

Use at least a quick target and plausibility scan on most meaningful work, but the depth of checking should match risk. High-mark, multi-step, unusual or domain-sensitive answers deserve stronger verification than routine one-step arithmetic.

Is redoing the same calculation a check?

It can catch a copying or arithmetic slip, but it is not strongly independent. A different route—substitution, estimation, units, graph, complement or inverse operation—often provides better evidence.

What is the fastest useful check?

Often target, sign, magnitude, units or domain. These cheap filters can reject impossible results before a full recomputation is attempted.

When should I use a second complete method?

When the question is high-value, the answer is surprising, a second route is genuinely different, and time permits. Do not solve every short problem twice.

Does substitution prove I found all the roots?

No. Substitution verifies candidates. Completeness comes from the solving method, domain analysis or a proof that no other candidates exist.

Can a graph verify algebra?

Yes for shape, sign, approximate intersections and plausibility. A graph window may hide behaviour, so it is not automatically proof of completeness or exactness.

Why do two methods sometimes agree even when the answer is wrong?

They may share the same wrong input, assumption or model. Independent checking is strongest when the routes have different failure modes and separately traceable inputs.

What if my check says the answer is wrong but I cannot find the error?

Keep the failed check as evidence and move into error analysis. Locate the first point where the original route and the verifying condition can no longer both be true.

How should I check a calculator answer?

Check the expression entered, brackets, mode, magnitude, exact-versus-decimal form, and whether the output satisfies the original mathematical conditions.

How do I check a word problem?

Return to every defining condition: totals, rates, counts, units, bounds and requested target. Replay the result through the story when possible.

Can checking waste time?

Yes, if every trivial line is recomputed or if low-value questions receive expensive second methods. Verification should be proportionate to risk and consequence.

How do I become independent at checking?

Learn a small repertoire—estimate, substitute, reverse, check units, test bounds, switch representation—and practise choosing the check yourself before asking for external confirmation.

28. A compact verification routine for everyday secondary mathematics

When time is ordinary, use this routine after a substantial question:

REQUEST: Did I answer what was asked?

TYPE: Is it the correct mathematical object and unit?

PLAUSIBILITY: Do sign, magnitude and direction make sense?

DOMAIN: Is the value allowed?

ORIGINAL CONDITION: Can I substitute, reverse or replay the answer?

COMPLETENESS: Have all required roots, cases, boundaries or units been included?

INDEPENDENT EVIDENCE: If the result is important or suspicious, is there another representation or method worth using?

The routine is intentionally ordered from cheap to expensive. A failure near the top means you often do not need to spend time on the lower levels yet.

For a one-line arithmetic question, the routine may collapse to magnitude and target. For a long model, every layer may matter.

As verification becomes fluent, the questions become internal. The learner does not need to recite the list forever.

Clara’s progression is typical of the goal. At first, her tutor asks each question. Later she writes “units?” or “sub?” in the margin. Eventually she checks without prompts.

The routine becomes successful when it reduces dependence while preserving efficiency.

29. An independence test for checking itself

A learner can know that checking is important while still depending on someone else to choose the check. Verification independence needs its own test.

Can the learner notice when a result is suspicious without being told?

Can the learner choose a check appropriate to the mathematical object?

Can the learner explain what the check would establish?

Can the learner interpret a failed check as evidence rather than simply erase the answer?

Can the learner decide when another check is unnecessary?

Can the learner preserve time by choosing a cheap strong check instead of a full second solution?

Can the learner identify when verification has confirmed a candidate but not completeness?

These are practical indicators, not a standardised scale.

Ben may be independent at equation substitution but still need prompts to check a geometry theorem condition. Mira may be strong at estimation but weak at recognising shared assumptions between two methods. Independence can be local.

A useful teaching progression is: check supplied → check prompted → check selected → check spontaneous → check risk-weighted.

The final stage matters. A learner who checks everything exhaustively has not necessarily become efficient. Mature verification includes knowing when enough evidence has been gathered for the task.

The aim is not permanent suspicion. It is calibrated confidence.

30. The complete secondary verification system: turn answers into justified confidence

Return to Clara, Ben and Mira. Clara’s neat page was not enough by itself. Ben’s unit check exposed a mismatch before he touched the arithmetic again. Mira’s failed substitution told her the answer had not survived the original equation.

Those three moments capture the system:

PREDICT → SOLVE → CHECK HIGH-LEVERAGE NODES → TEST TARGET, TYPE, SIGN, MAGNITUDE, UNITS AND DOMAIN → RETURN TO THE ORIGINAL CONDITION → CHECK COMPLETENESS → USE AN INDEPENDENT ROUTE WHEN THE RISK JUSTIFIES IT → TRACE DISAGREEMENT IF CHECKS CONFLICT.

Verification is not the same as accuracy, although the two support each other. Accuracy protects the mathematical state while the solution is being produced. Verification gathers evidence that the produced result deserves confidence.

Verification is also not the same as error repair. When a check fails and the reason keeps recurring, How to Stop Repeating Math Mistakes owns the diagnosis-repair-retest loop.

For the focused skill of changing routes, use Check an Answer Through an Independent Route. For the broad theory of mathematical verification across proof, modelling, computation and higher mathematics, use How Mathematical Verification Works | From Answer to Confidence. For Singapore-specific fault-tolerance architecture, use Redundancy, Independent Checks and Mathematical Fault Tolerance.

For course and stage routes, return to the Secondary Mathematics Learning Hub. Examination-specific pacing and paper control remain with Mathematics Examination Craft.

Scope note: the fictional learner scenes, verification routines and checkpoint tasks in this article are original explanatory material. They are not a validated assessment instrument and do not guarantee examination outcomes. They should be adapted to the learner’s taught content, course requirements and available time.

A mathematical answer becomes trustworthy not because it looks finished, but because it survives the right attempts to prove it wrong.

Appendix A — Verification matrix: what to check, what it catches, and what it cannot prove

A mature checking system does not use the same verification move for every answer. Different checks detect different failure classes. The matrix below is written as a set of practical routes rather than as a rigid checklist. Choose the row that matches the mathematical claim you are trying to validate.

Equation solution → substitute into the original equation

Use when: you have found one or more candidate values for an equation or system. What it catches: arithmetic slips, extraneous candidates, copied signs and transformations that changed the solution set. What it does not prove by itself: that every possible solution has been found. A candidate can pass substitution while another valid candidate remains missing.

Example: x = 4 is proposed for 3x + 2 = 14. Substitution gives 12 + 2 = 14, so the candidate works. If the original problem was quadratic, however, verifying one candidate would not establish completeness.

Factorisation → expand the factors

Use when: an expression has been factorised. What it catches: wrong signs, missing terms and incorrect factor pairs. What it does not prove: that the chosen factorisation is the most useful form for the next task.

Example: (x − 6)(x + 2) expands to x² − 4x − 12. If the original expression was x² − 4x − 12, the factorisation is structurally verified.

Percentage change → replay the transformation forward

Use when: an original value has been recovered from a final value, or several percentage operations were chained. What it catches: wrong bases, wrong retained multipliers and wrong operation order. What it does not prove: that unrelated fees or conditions were modelled correctly unless they are included in the replay.

Example: a final value of 153 after a 15% decrease implies original p = 153/0.85 = 180. Replaying gives 180 − 27 = 153.

Ratio split → check both total and ratio

Use when: a whole has been divided according to a ratio. What it catches: wrong one-part value, switched components and arithmetic slips. What it does not prove: that the original interpretation of the given quantity was correct if the problem involved a difference rather than a total.

Example: splitting 72 in ratio 5:7 gives 30 and 42. They sum to 72 and 30:42 simplifies to 5:7.

Measurement result → check units and dimensional type

Use when: length, area, volume, speed, density or another measured quantity is involved. What it catches: inverted rates, missing square or cubic units, incompatible additions and some wrong formulas. What it does not prove: numerical correctness when a wrong formula happens to have the right dimensions.

Example: a reported area of 54 cm fails the dimensional check; the unit should be cm². But 48 cm² and 54 cm² both have correct dimensions, so units alone cannot distinguish them.

Probability → check range, total mass and complement

Use when: probabilities are calculated through cases, trees or counting. What it catches: outputs outside [0,1], omitted exhaustive branches and some event-definition errors. What it does not prove: that a plausible probability inside [0,1] represents the correct event.

Example: if exhaustive terminal branches sum to 0.91, the tree is incomplete or arithmetically wrong. If they sum to one, the tree may still be modelling the wrong event.

Graph result → check direction, intercepts, domain and visible intersections

Use when: algebra and graph describe the same relation. What it catches: wrong gradient signs, impossible intercepts, obvious missing crossings and some model-domain errors. What it does not prove: global completeness outside the viewing window or exact coordinates read only approximately from pixels.

Example: y = −4x + 9 should fall as x increases. A rising plotted line disagrees immediately with the algebra.

General formula → test simple and boundary cases

Use when: a formula is claimed to work for a range of inputs. What it catches: missing constants, wrong indexing, incorrect endpoint behaviour and formulas that fail special cases. What it does not prove: universal validity. Passing many cases is evidence, not proof.

Example: a proposed arithmetic-sequence nth-term formula should reproduce the first term at n = 1 and the common difference between consecutive generated terms.

Universal claim → search for a counterexample

Use when: the statement says “all”, “every”, “always” or otherwise claims universality. What it catches: false universal statements. What it does not prove: truth when no counterexample is found quickly. Proof may still be required.

Example: “Every prime number is odd” is refuted by the prime number 2.

Optimization answer → check domain, boundary and objective value

Use when: a maximum or minimum has been found. What it catches: stationary points outside the domain, local results mistaken for global ones and missed endpoints. What it does not prove: model validity outside the mathematical objective.

Example: for A = 100 − (x − 10)² on 0 < x < 20, the form itself shows A ≤ 100. The maximum is achieved at x = 10, which lies in the domain.

Integer count → test neighbouring allowed values

Use when: a continuous calculation must become a whole-number decision. What it catches: ordinary rounding used where floor or ceiling logic is required. What it does not prove: that the model producing the real-valued boundary was correct.

Example: if 5.2 vans are needed by capacity, test five and six. Five is insufficient; six is the minimum whole count.

Model output → check physical or contextual bounds

Use when: a formula represents a real situation. What it catches: negative lengths, values beyond capacity, impossible times, unsupported extrapolation and domain violations. What it does not prove: that the model is empirically adequate across every situation.

Example: a 70-litre tank model cannot report 92 litres retained unless the problem supplies a new rule about overflow or changed capacity.

Calculator output → estimate independently and inspect the entered expression

Use when: a machine produced the numerical result. What it catches: missing brackets, decimal slips, wrong angle mode and some copied inputs. What it does not prove: that the mathematical model entered was the correct model for the question.

Example: 29.8 × 41.2 should be roughly 30 × 40 = 1,200. A display of 122.776 is immediately suspicious; the exact product is 1,227.76.

Two methods agree → inspect shared assumptions before declaring victory

Use when: independent-looking routes produce the same answer. What it catches: method-specific failures when the routes are genuinely different. What it does not prove: correctness if both methods share the same wrong input or model.

Example: elimination and substitution can agree on the solution of the same incorrectly copied system. Agreement confirms internal consistency of the copied system, not fidelity to the original page.

Appendix B — Twelve composite verification cases where one check is not enough

The following cases are longer because real checking often requires more than one layer. Each case demonstrates how cheap checks can be combined without turning verification into a second full solution.

Case 1 — The correct root, incomplete answer set

A learner solves x² − 9 = 0 and reports x = 3. Substitution confirms that 3 is valid: 9 − 9 = 0. The answer still fails completeness. Factorisation gives (x − 3)(x + 3) = 0, so x = 3 or x = −3.

Lesson: candidate verification and completeness verification are separate. A passing substitution check can coexist with an incomplete final answer.

Case 2 — Correct arithmetic, wrong percentage model

A jacket is reduced by 20% and then a 10-unit fixed delivery fee is added. A learner models the final price as 0.8(p + 10). Every later arithmetic line is correct.

Replay the story. The fee is added after the discount, so it should not itself be discounted. The correct model is 0.8p + 10. A boundary thought experiment also helps: if p = 0 under the simplified model, the stated fixed fee should still contribute ten, while 0.8(p + 10) would contribute only eight.

Lesson: numerical checking cannot rescue a wrong model. Whole-problem replay must sometimes come before arithmetic verification.

Case 3 — Two methods agree because both use the same copied number

A geometry problem states a side length of 18, but the learner copies it as 13. They calculate an area by trigonometry and independently by coordinates. Both routes agree because both use thirteen.

The shared-input audit catches the issue. Independent methods are not independent of a common transcription error.

Lesson: provenance matters. Verification routes should be checked not only for different methods but also for shared data sources.

Case 4 — Plausible probability, wrong event

A question asks for the probability of at least one six in two fair dice rolls. A learner calculates 10/36 by counting exactly one six and ignoring the double-six outcome. The result 5/18 lies comfortably between zero and one.

The range check passes, so a stronger event check is needed. Complement gives 1 − (5/6)² = 11/36. Direct counting confirms eleven outcomes: six with first die six, six with second die six, minus the double-six overlap counted twice.

Lesson: passing a weak check does not mean the answer is verified. Match check strength to the failure class.

Case 5 — Rounding changes a threshold decision

A computed length is √52, and a design is allowed only if the length is at most 7.21. A learner rounds √52 to 7.2 and accepts the design.

Precision checking gives √52 ≈ 7.211102…, which is greater than 7.21. The early rounded value changed the yes/no conclusion.

Lesson: when a value is near a decision threshold, retain exactness or sufficient precision until the comparison is complete.

Case 6 — Graph agrees locally but misses another root

A graphing window displays an x-intercept near 2 for the polynomial (x − 2)(x + 12). The learner reports one real root.

The visible graph supports the root at two, but the current window may not include −12. Factorisation supplies the complete algebraic root structure.

Lesson: a graph can verify a visible candidate without verifying completeness. The scope of the representation matters.

Case 7 — Unit check passes, formula is still wrong

A learner proposes area of a rectangle as A = 2lw. The units are length × length, so the result has area units. Dimensional checking passes. The formula is still wrong because rectangle area is lw, not 2lw.

A special case catches it: a 3-by-4 rectangle is twelve square units by direct counting or decomposition, not twenty-four.

Lesson: units are a strong rejection filter but not a complete correctness certificate.

Case 8 — Stationary point is not enough

A learner differentiates an objective and finds a stationary point inside the domain, then declares it the maximum without checking endpoints or shape.

A verification route must establish what kind of point it is and whether any allowed boundary produces a larger value. Depending on the function and course, this may use completed-square structure, derivative sign changes, second derivative, graph shape, or direct endpoint comparison.

Lesson: finding a candidate extremum and verifying a global optimum are different jobs.

Case 9 — Exact algebra, impossible physical answer

A quadratic model for a rectangular width produces roots 6 and −14. Both satisfy the polynomial exactly.

The whole-problem domain rejects −14 as a physical width. The positive candidate must still be checked in the original dimensional relationships.

Lesson: algebraic validity is not the final authority when variables represent constrained quantities.

Case 10 — Complement and direct probability disagree

A learner obtains P(at least one head in three fair coin tosses) = 5/8 by listing outcomes. Complement gives 1 − 1/8 = 7/8.

Do not average the two results. Inspect the listed outcomes. There are eight equally likely ordered outcomes and seven contain at least one head. The direct listing omitted two successful cases.

Lesson: disagreement between independent routes is a diagnostic signal. Trace the route that cannot satisfy the shared event definition.

Case 11 — A correct candidate does not prove a universal claim

A learner tests the identity (a + b)² = a² + 2ab + b² at a = 2, b = 3 and gets 25 on both sides. This numerical check supports the identity at one case.

For a universal identity, algebraic expansion supplies the general verification: (a + b)(a + b) = a² + ab + ab + b² = a² + 2ab + b².

Lesson: the verification burden grows with the scope of the claim. One example can falsify a universal statement but cannot establish it universally.

Case 12 — Verification itself can be wrong

A learner solves 6x − 5 = 25 as x = 5. They try to check by substitution but accidentally enter 6 × 5 + 5 and obtain 35, concluding the original solution is wrong.

The check has miscopied the original equation. Correct substitution gives 30 − 5 = 25, so x = 5 is valid.

Lesson: checks are mathematical work too. They need faithful inputs, clear conditions and interpretation. When a check contradicts a reliable route, verify the verification before discarding the original result.

Appendix C — A five-level verification ladder for tutors and independent learners

Level 1 — Plausibility. Check target, sign, units, magnitude and obvious bounds. Cost: seconds. Best for quickly rejecting impossible results.

Level 2 — Reconstruction. Substitute, reverse an operation, replay a percentage transformation or reconstruct a total. Cost: usually under a minute. Best for candidate answers whose original relation is easy to test.

Level 3 — Structural cross-check. Expand a factorisation, use a complement, compare with a graph, test a boundary or switch representation. Cost: moderate. Best when one method may hide a structural mistake.

Level 4 — Independent solution route. Solve by a genuinely different method or decomposition. Cost: higher. Best for high-value, suspicious or reusable results.

Level 5 — Proof or model validation. Use rigorous proof for a universal mathematical claim, or empirical/model validation when the question concerns fit to reality. Cost and method depend on the claim. This level is not required for every school exercise.

The ladder prevents two opposite errors: underchecking important work and overchecking trivial work. A learner should climb only as high as the task and risk require.

Over time, checking decisions themselves become part of mathematical judgment. The strongest learner is not the one who checks the most. It is the one who knows which evidence is enough, which uncertainty remains, and when another route would genuinely add confidence.

Find the First Invalid Step: Worked Solutions With Planted Errors

A wrong final answer can contain several correct lines after the original mistake. Correcting the last line therefore teaches less than locating the first line where the mathematics stopped being valid.

This case bank trains a precise routine: identify the first invalid step, name the mathematical reason, make the smallest repair that restores validity, complete the solution, and then retest the same idea on a fresh problem. The cases are original BTT practice material. They are deliberately plausible because useful error analysis should resemble mistakes a capable learner might actually make.

The first-invalid-step routine

1. Read the original question before reading the solution.
2. State any domain, unit or context restriction.
3. Compare each line with the line immediately before it.
4. Stop at the first line that is not justified.
5. Name the error mechanism.
6. Repair only what is necessary at that point.
7. Continue from the repaired line.
8. Use an independent check where possible.
9. Attempt a changed problem later without being told the error type.

Case 1 — the sign changes without an operation

Problem: solve 4x-7=21.

Planted solution:

4x-7=21
4x=21-7
4x=14
x=3.5

First invalid step: 4x=21-7. Adding 7 to both sides should give 4x=28. The familiar phrase “move -7 across” has been applied with the wrong sign.

Smallest repair: replace the second line with 4x=21+7=28. Then x=7.

Independent check: 4(7)-7=28-7=21.

Case 2 — cancellation across addition

Problem: simplify (x+6)/x for x≠0.

Planted solution: cancel x with the x in the numerator to obtain 6.

First invalid step: cancellation has been applied to a term inside a sum. Factors can cancel; added terms cannot be cancelled individually across a fraction bar.

Smallest repair: rewrite (x+6)/x=x/x+6/x=1+6/x.

Check: choose x=2. Original gives 8/2=4; repaired form gives 1+3=4; the incorrect answer 6 fails.

Case 3 — squaring creates an untested candidate

Problem: solve √(x+2)=x.

Planted solution:

x+2=x²
x²-x-2=0
(x-2)(x+1)=0
x=2 or x=-1.

First invalid conclusion: the algebraic factorisation is correct; the error is accepting both roots as final without returning to the original equation. Squaring produced candidates.

Smallest repair: check both in the original. x=2 gives √4=2. x=-1 gives √1=1≠-1. Final solution: x=2.

Case 4 — a denominator restriction disappears

Problem: solve (x²-9)/(x-3)=6.

Planted solution:

x+3=6
x=3.

First invalid omission: before cancellation, the original denominator requires x≠3. Simplifying to x+3 is valid only on that restricted domain.

Smallest repair: write x≠3 before simplification. The transformed equation produces the excluded candidate 3, so the original equation has no solution.

Case 5 — percentage change uses the wrong base

Problem: a price rises from $80 to $92. Find the percentage increase.

Planted solution: increase is $12, so percentage increase is 12/92×100%=13.0%.

First invalid step: the denominator should be the original value, not the new value.

Smallest repair: 12/80×100%=15%.

Check: 15% of 80 is 12, and 80+12=92.

Case 6 — Pythagoras uses the wrong side as the hypotenuse

Problem: a right triangle has hypotenuse 13 cm and one leg 5 cm. Find the other leg.

Planted solution: x²+13²=5².

First invalid step: the hypotenuse must be the side opposite the right angle and the largest side, so 13 belongs alone on the c² side.

Smallest repair: x²+5²=13², so x²=144 and the positive length is x=12.

Case 7 — a probability model double-counts outcomes

Problem: two fair coins are tossed. Find the probability of exactly one head.

Planted solution: possible outcomes are HH, HT, TT, so one of three outcomes has exactly one head; probability 1/3.

First invalid step: the sample space is incomplete. TH is distinct from HT when the two toss positions are tracked.

Smallest repair: sample space HH, HT, TH, TT. Two of four equally likely outcomes have exactly one head, so probability 1/2.

Case 8 — the mean is recomputed with the wrong count

Problem: four scores have mean 18. A fifth score of 22 is added. Find the new mean.

Planted solution: original total is 18×4=72. New total is 94. New mean is 94/4=23.5.

First invalid step: the total is now spread across five scores, not four.

Smallest repair: 94/5=18.8.

Case 9 — inequality direction is not reversed

Problem: solve -3x>12.

Planted solution: divide by -3 to get x>-4.

First invalid step: dividing an inequality by a negative quantity reverses the inequality sign.

Smallest repair: x<-4.

Check: choose x=-5; then -3(-5)=15>12. Choose x=0; it fails, as expected.

Case 10 — a formula is used outside its assumptions

Problem: find the area of a triangle with sides 8 cm and 11 cm and included angle 30°.

Planted solution: A=1/2×8×11=44 cm².

First invalid step: 1/2 bh requires a perpendicular height, and 11 cm has not been given as perpendicular to the 8 cm side.

Smallest repair: use A=1/2 ab sin C: 1/2×8×11×sin30°=22 cm².

Lesson: formula recall without condition recall is incomplete knowledge.

Case 11 — the quadratic formula changes one sign

Problem: solve 2x²+3x-2=0.

Planted solution: x=[3±√(9+16)]/4.

First invalid step: the numerator should begin with -b, so it is -3, not 3.

Smallest repair: x=[-3±5]/4, giving x=1/2 or x=-2.

Check: substitute both into the original or factor as (2x-1)(x+2)=0.

Case 12 — a graph reading becomes an exact claim

Problem: a graph suggests a root near x=1.41 for x²-2=0.

Planted conclusion: therefore the exact positive root is 1.41.

First invalid step: the graph supplies an approximation. It does not establish exactness.

Smallest repair: state x≈1.41 to two decimal places, or solve exactly to get x=√2 if exact form is required.

Case 13 — a unit conversion is applied twice

Problem: convert 2.4 m to centimetres.

Planted solution: 2.4×100=240, then because centimetres are smaller, multiply by 100 again to get 24,000 cm.

First invalid step: the first multiplication already completed the conversion. A metre contains 100 centimetres.

Smallest repair: stop at 240 cm.

Check: a 2.4 m object should be a few hundred centimetres, not hundreds of metres expressed in centimetres.

Case 14 — a negative length survives algebra

Problem: a rectangle has width x and length x+3; area is 40. Solve for the dimensions.

Planted solution: x(x+3)=40, so x²+3x-40=0=(x+8)(x-5). Thus x=-8 or 5. The rectangle has either width -8 or 5.

First invalid conclusion: both algebraic roots solve the equation, but a physical width cannot be negative in the stated model.

Smallest repair: apply the context domain x>0. Width is 5 and length 8.

Case 15 — the correct operation is performed on the wrong quantity

Problem: a quantity of 300 is increased by 20% and then decreased by 20%.

Planted solution: +20% and -20% cancel, so final value is 300.

First invalid assumption: the two percentage changes use different bases. After the increase, the new value is 360; the 20% decrease is 72, giving 288.

Smallest repair: use multipliers 300×1.2×0.8=288.

Why the smallest repair matters

If a learner makes the error in line 2 and the tutor rewrites the entire solution from scratch, the learner sees a correct answer but may not learn which decision failed. The smallest repair preserves the correct work around the error and focuses attention on the exact mechanism that needs updating.

This is especially valuable when long solutions contain only one conceptual fault. Rebuilding everything can make the task look harder than it is.

The distractor-reasoning test

For each planted error, ask why the wrong step felt plausible. Good distractors usually borrow a real rule and apply it in the wrong structure:

• “move and change sign” without reconstructing the balancing operation;
• cancellation, but across addition instead of common factors;
• squaring, but without candidate checking;
• percentage, but with the wrong reference value;
• Pythagoras, but with the wrong side identified as hypotenuse;
• probability counting, but with an incomplete sample space;
• formula substitution, but without checking assumptions.

Understanding the attraction of the wrong move is more useful than calling it careless.

Fresh retest set

Attempt these later without being told which error family is being tested.

1. A solution to 5x+4=29 writes 5x=29+4. Find and repair the first invalid step.
2. A learner simplifies (x+4)/(x+2) to 2 by “cancelling x”. Explain why this fails.
3. Solve √(x+6)=x and identify any extraneous candidate.
4. A price falls from $250 to $200. A learner divides the decrease by 200. Repair the percentage calculation.
5. For a right triangle with legs 9 and 12, a learner writes c=9+12. Repair and check.
6. Two dice are rolled. A learner treats sums 2 through 12 as equally likely. Explain the sample-space error.
7. Solve -4x≤20 and justify the inequality direction.
8. A graph shows a root at approximately 2.24. State what further work is needed before claiming the exact root is √5.

Release test

The learner has moved beyond correction copying when they can:

• find the earliest invalid step rather than circling the final answer;
• name the mathematical mechanism that failed;
• preserve earlier correct work;
• make the smallest valid repair;
• choose an independent check;
• explain why the wrong move was tempting;
• solve a fresh version after a delay without being told the error type.

Continue through BTT

Use How to Learn From Maths Feedback when the next job is turning a marked error into a durable repair. Use How to Stop Repeating Math Mistakes when the same failure returns across tasks. Use How to Solve Math Problems when the difficulty lies in constructing the route rather than verifying it.

The aim is not to become good at spotting planted mistakes. It is to build a checking habit precise enough to catch your own first invalid step before the rest of the page grows from it.