Singapore School Mathematics Operating Manual · Chapter 37
A mathematical solution should not depend on one fragile chain if a cheap independent check can catch failure.
Engineers build redundancy into systems because one component can fail. Mathematics uses the same principle in a quieter form: substitute a root back into the original equation, estimate before calculating exactly, compare units, check a graph against algebra, solve by two decompositions, verify a probability total, test a boundary case, or use an invariant to see whether the answer could possibly be right.
This chapter calls that discipline mathematical fault tolerance. The goal is not to repeat every calculation twice. It is to place independent checks where they have the greatest value.
1. Redundancy means more information or more routes than the minimum required
If two non-parallel equations already determine x and y, a third independent relationship is redundant for solving but useful for checking.
Redundancy can therefore increase reliability even when it does not increase uniqueness.
2. A repeated calculation is not always an independent check
If the same mistaken formula is entered twice, agreement proves little.
A stronger check uses a different relationship: solve a quadratic by factorisation, then substitute roots; find an area by addition of parts, then compare with subtraction from a larger shape.
3. Independent checks should have different failure modes
Unit analysis catches dimension errors. Estimation catches scale errors. Substitution catches algebraic candidate errors. Graphs catch sign and intersection-count errors. Bounds catch impossible magnitudes.
One method rarely catches every class of mistake.
4. Cheap checks belong early
If a probability is 1.4, reject it before spending time on a full recomputation.
If an area is reported in cm rather than cm², the type check already failed.
Early low-cost filters save effort.
5. High-leverage intermediate results deserve redundancy
If one radius feeds circumference, area and volume calculations, check it before passing it forward.
A wrong high-leverage node can corrupt an entire dependency branch.
6. Substitution is a strong equation check
If x=5 is claimed to solve 3x−2=13, substitute: 15−2=13.
The check returns to the original equation rather than trusting the transformation chain.
7. Factorisation and expansion check one another
If x²−5x+6 is factorised as (x−2)(x−3), expanding the factors recovers the original polynomial.
This is a reversible structural check.
8. Differentiation and integration can sometimes check one another
If an antiderivative F is proposed for f, differentiate F.
If F′=f, the derivative check verifies the local antiderivative relationship.
The constant of integration remains a separate issue.
9. Graph and algebra should agree qualitatively
If algebra says a quadratic has no real roots but the graph visibly crosses the x-axis twice, investigate.
One representation is exposing inconsistency in the other.
10. Scale estimation is fault tolerance against decimal errors
Before calculating 49×201, estimate 50×200≈10,000.
An exact result near 100,000 is immediately suspicious.
11. Bounds are stronger than impressions
If x lies between 2 and 3, then x² lies between 4 and 9.
A computed x² of 14 is impossible.
Bounds create a hard consistency envelope.
12. Unit checking is a structural checksum
Distance/time should produce speed units.
Length×length should produce area units.
When the output type is wrong, arithmetic detail cannot rescue the result.
13. Probability totals provide conservation-style checks
Mutually exclusive exhaustive outcomes should sum to probability 1.
If a completed tree totals 0.83, a branch may be missing or miscalculated.
14. Complement methods create alternative routes
Instead of adding many successful probability branches, calculate 1−P(failure) when failure is simpler.
Agreement between direct and complement routes is strong evidence.
15. Geometry has built-in redundancy
Triangle angles sum to 180°.
Parallel-line angle relationships, congruence conditions and side constraints can cross-check one another.
A geometry answer should fit the whole figure, not just one local equation.
16. Coordinate geometry supports distance-gradient cross-checks
A claimed perpendicular relationship should satisfy gradient product −1 for non-vertical finite gradients, while vector dot product gives a more general orthogonality check.
Different formalisms can verify the same geometry.
17. A second method is most valuable when it is truly different
Two calculator entries based on the same wrong expression are not independent.
Algebra plus geometry, exact work plus estimation, or forward solving plus backward substitution are more independent.
18. Redundant data can detect model mismatch
If two observations determine a line but a third measured point lies far away, either the data contain error or the linear model is inadequate.
Redundancy can therefore test the model, not only the arithmetic.
19. Checks can be local or global
Local: verify one intermediate value.
Global: check the final answer satisfies every original condition.
Both are needed in long problems.
20. Fault tolerance is not the same as distrust
Strong mathematicians do not check because every line is expected to fail.
They check because cheap independent evidence increases confidence and reduces the cost of hidden error.
21. Too much redundancy can waste time
Recomputing every trivial arithmetic line three ways is inefficient.
Use redundancy where error probability, downstream impact or decision importance is high.
22. Examination checking should be risk-weighted
Prioritise high-mark multi-step questions, answers near thresholds, unusual signs, complicated substitutions and values used later.
Simple one-step arithmetic may need only a quick reasonableness scan.
23. Different checks cover different error classes
Substitution: equation validity.
Units: quantity type.
Bounds: feasible magnitude.
Alternative method: method-specific error.
Graph: global shape.
Estimation: order of magnitude.
24. Redundancy can be built into workflow
Predict → calculate → verify.
Or solve → substitute → interpret.
Or derive → estimate → compare.
These small loops create routine fault tolerance.
25. A discrepancy is useful information
If two checks disagree, do not average them.
Trace the two provenance chains and find where they diverge.
Disagreement narrows the search for error.
26. Independent checks need independent provenance
If both methods use the same miscopied input, they can agree falsely.
The Provenance and Result Lineage chapter helps identify shared dependencies.
27. Redundancy supports resilient learning
A student who knows only one narrow method has no fallback when the representation changes.
Knowing a second representation or check improves transfer even if the primary method remains fastest.
28. A practical fault-tolerance audit
Ask:
Which result is most dangerous if wrong? What independent check is cheapest? Does the check rely on a different relationship? What error class can it detect? Is there redundant information available? If two checks disagree, can I trace their separate lineages?
29. Independent practice
1. Solve x²−5x+6=0 and give one independent check.
2. A probability calculation gives 1.12. What immediate check fails?
3. Why is recalculating with the same formula not fully independent?
4. Give one scale estimate for 51×198.
5. Why should a high-leverage intermediate result be checked early?
6. Give one example of redundant data that can verify a model.
7. Name two checks with different failure modes.
8. Why should disagreement between checks not be averaged?
30. Worked answers
1. (x−2)(x−3)=0, so x=2 or 3. Substitute each into the original equation.
2. Probability must lie between 0 and 1.
3. The same conceptual error can be repeated both times.
4. 50×200≈10,000.
5. Many downstream results may inherit its error.
6. Two points determine a proposed line; a third observed point can test whether the linear model remains consistent.
7. Unit check and substitution; or estimation and graph check.
8. One route may be wrong; averaging would hide the inconsistency rather than diagnose it.
31. Continue through Batch 10
Use Decision Margins and Slack to know where redundancy matters most, Residuals and Model Mismatch when the check disagrees with the model, and Evidence Sufficiency and Verification Depth to decide when enough checking has been done.
Return to the BTT Mathematics Hub for Batch 10.