Chapter map

1. 1. Total scatter
2. 2. Two clusters
3. 3. Three clusters
4. 4. Four clusters
5. 5. Five clusters
6. 6. Raw pattern
7. 7. Gain into three
8. 8. Gain into four
9. 9. Gain into five
10. 10. Second difference at three
11. 11. Second difference at four
12. 12. Selection
13. 13. Zero determinant
14. 14. Scale transformation
15. 15. Range
16. 16. Claim

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

---

1. Total scatter

The central mathematical idea is concrete. Set det(T)=100.

A concrete situation

T stays fixed across k.

What the mathematics does not license

Use the same observations and centring. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record T. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

2. Two clusters

Here the geometry becomes operational. Set det(W2)=25.

A concrete situation

R2=100/25=4.

What the mathematics does not license

W must be compatible with T. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show step. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

3. Three clusters

This step deserves a careful pause. Set det(W3)=10.

A concrete situation

R3=10.

What the mathematics does not license

Generalised within variation falls. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Show step. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

4. Four clusters

The useful question is not only how to calculate, but what the calculation assumes. Set det(W4)=8.

A concrete situation

R4=12.5.

What the mathematics does not license

Raw ratio rises. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Show step. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

5. Five clusters

A small example makes the mechanism visible. Set det(W5)=7.5.

A concrete situation

R5=13.3333.

What the mathematics does not license

Round at the end. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show step. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

6. Raw pattern

This is where modelling choices begin to affect evidence. List 4,10,12.5,13.3333.

A concrete situation

Improvement decelerates.

What the mathematics does not license

Largest raw score would favour the edge. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Plot sequence. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

7. Gain into three

The central mathematical idea is concrete. Compute 10-4.

A concrete situation

Gain=6.

What the mathematics does not license

This is a first difference. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record value. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

8. Gain into four

Here the geometry becomes operational. Compute 12.5-10.

A concrete situation

Gain=2.5.

What the mathematics does not license

Marginal gain shrinks. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Record value. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

9. Gain into five

This step deserves a careful pause. Compute 13.3333-12.5.

A concrete situation

Gain=.8333.

What the mathematics does not license

Improvement is flattening. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record value. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

10. Second difference at three

The useful question is not only how to calculate, but what the calculation assumes. Compute 4-2×10+12.5.

A concrete situation

D2=-3.5.

What the mathematics does not license

The label belongs to the centre k. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Show step. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

11. Second difference at four

A small example makes the mechanism visible. Compute 10-2×12.5+13.3333.

A concrete situation

D2=-1.6667.

What the mathematics does not license

Use consistent precision. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show step. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

12. Selection

This is where modelling choices begin to affect evidence. Choose the minimum second difference.

A concrete situation

k=3 is the sharper bend.

What the mathematics does not license

State the software convention. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record result. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

13. Zero determinant

The central mathematical idea is concrete. Let det(W)=0.

A concrete situation

Ratio is infinite or undefined.

What the mathematics does not license

Do not treat infinity as automatic evidence. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Flag failure. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

14. Scale transformation

Here the geometry becomes operational. Transform every feature nonsingularly.

A concrete situation

Determinant factors cancel for a fixed partition.

What the mathematics does not license

Refitting can still change labels. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Check both. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

15. Range

This step deserves a careful pause. Extend the candidate set.

A concrete situation

Curvature can change.

What the mathematics does not license

Boundary choices are evidence. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Run variants. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

16. Claim

The useful question is not only how to calculate, but what the calculation assumes. Name T,W,determinant method,range and second-difference rule.

A concrete situation

Teaching elbow supports k=3.

What the mathematics does not license

Do not call clusters real. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Bound conclusion. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

A seven-day practice lab

1. Day 1: define the decision, observation unit, X variables and Y or reference condition.
2. Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
3. Day 3: build a transparent baseline and write what it assumes.
4. Day 4: fit the main method inside a training-only pipeline; record every parameter.
5. Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
6. Day 6: compare validation metrics with the practical workload created by errors.
7. Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

• Which part of the model is a mathematical definition, and which part is a choice?
• What small calculation would catch a sign, scale or dimension error?
• What evidence comes from training, validation and the untouched test?
• What benign condition could look like a fault or prediction failure?
• How would the answer change for another response, asset, school project or time period?
• What should a human review before acting on the score?

A method-specific audit before deployment

The Rubin determinant ratio is determinant T divided by determinant Wk. T measures total multivariate scatter and Wk measures scatter remaining within the proposed clusters. Because T is fixed for a dataset and Wk usually falls as k increases, the raw ratio often rises monotonically. A cluster-count rule therefore studies curvature, not simply the largest raw value.

Let determinant T=100. With determinant W2=25, Rubin R2=4. With W3=10, R3=10. With W4=8, R4=12.5. With W5=7.5, R5=13.3333. The raw sequence improves every time, but the gains are 6, 2.5 and about 0.8333.

The centred second difference at k=3 is 4 minus 2 times 10 plus 12.5, or minus 3.5. At k=4 it is 10 minus 2 times 12.5 plus 13.3333, or about minus 1.6667. Under the stated minimum-second-difference convention, k=3 is the sharper bend.

The ratio is scale invariant only under compatible nonsingular linear rescaling applied to both T and W, while the fitted partition itself can still change with preprocessing. Singular determinants make the ratio infinite or undefined. Pseudo-determinants and regularisation change the meaning and must be disclosed. Use stability and domain evidence before treating the elbow as a useful grouping.

---

Frequently asked questions

What is the Rubin index?

The determinant of total scatter divided by the determinant of pooled within-cluster scatter.

Which direction is better?

The raw ratio usually rises with k, so the stated selection rule uses the minimum centred second difference to find a sharp bend.

Why not choose the largest ratio?

That often selects the largest allowed k because within scatter generally shrinks as clusters are added.

What does the determinant represent?

It is a multivariate generalised-variance measure related to the volume of a scatter ellipsoid.

What if W is singular?

The ordinary ratio is infinite or undefined; pseudo-determinants and ridges define modified criteria.

What must be reported?

T and W definitions, determinant algorithm, variables, scaling, candidate range, raw ratios, differences, selection rule and stability checks.

---

Primary source and fact-check note

H. P. Friedman and J. Rubin, “On Some Invariant Criteria for Grouping Data,” Journal of the American Statistical Association 62(320), 1159–1178 (1967), DOI 10.1080/01621459.1967.10500923. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

• Friedman–Rubin trace criterion
• Scott–Symons log-determinant ratio
• Marriott determinant penalty
• Davies–Bouldin scatter and separation
• How Mathematics Works: Statistics

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Extended laboratory: matrix evidence and responsible cluster selection

Data audit

Freeze observations, variables, units, exclusions, missing-data rules, weights, standardisation, sample size and seeds.

Scatter audit

Define total, within and between sums-of-squares and cross-products matrices with centring and denominators.

Linear-algebra audit

Record determinants, log determinants, eigenvalues, condition numbers, inverse method and any numerical tolerance.

Sequence audit

Show raw scores, first differences, centred second differences, candidate labels and selection direction.

Convention audit

Separate historical optimisation rules from current software conventions and name the implementation and version.

Uncertainty audit

Repeat starts, samples, variable sets, transformations, candidate ranges and plausible regularisation settings.

Meaning audit

Compare internal geometry with stability and domain evidence without assuming clusters are natural categories.

Final claim

State the exact matrix evidence and stop before causal, semantic, career or guaranteed-performance claims.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Total scatter

Begin with this mechanism: Set det(T)=100. Recreate the concrete case: T stays fixed across k. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Use the same observations and centring. Complete the transfer step: Record T. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Two clusters

Begin with this mechanism: Set det(W2)=25. Recreate the concrete case: R2=100/25=4. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: W must be compatible with T. Complete the transfer step: Show step. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Three clusters

Begin with this mechanism: Set det(W3)=10. Recreate the concrete case: R3=10. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Generalised within variation falls. Complete the transfer step: Show step. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Four clusters

Begin with this mechanism: Set det(W4)=8. Recreate the concrete case: R4=12.5. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Raw ratio rises. Complete the transfer step: Show step. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 5: Five clusters

Begin with this mechanism: Set det(W5)=7.5. Recreate the concrete case: R5=13.3333. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Round at the end. Complete the transfer step: Show step. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Raw pattern

Begin with this mechanism: List 4,10,12.5,13.3333. Recreate the concrete case: Improvement decelerates. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Largest raw score would favour the edge. Complete the transfer step: Plot sequence. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Gain into three

Begin with this mechanism: Compute 10-4. Recreate the concrete case: Gain=6. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is a first difference. Complete the transfer step: Record value. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Gain into four

Begin with this mechanism: Compute 12.5-10. Recreate the concrete case: Gain=2.5. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Marginal gain shrinks. Complete the transfer step: Record value. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 9: Gain into five

Begin with this mechanism: Compute 13.3333-12.5. Recreate the concrete case: Gain=.8333. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Improvement is flattening. Complete the transfer step: Record value. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Second difference at three

Begin with this mechanism: Compute 4-2×10+12.5. Recreate the concrete case: D2=-3.5. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The label belongs to the centre k. Complete the transfer step: Show step. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Second difference at four

Begin with this mechanism: Compute 10-2×12.5+13.3333. Recreate the concrete case: D2=-1.6667. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Use consistent precision. Complete the transfer step: Show step. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Selection

Begin with this mechanism: Choose the minimum second difference. Recreate the concrete case: k=3 is the sharper bend. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: State the software convention. Complete the transfer step: Record result. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 13: Zero determinant

Begin with this mechanism: Let det(W)=0. Recreate the concrete case: Ratio is infinite or undefined. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not treat infinity as automatic evidence. Complete the transfer step: Flag failure. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Scale transformation

Begin with this mechanism: Transform every feature nonsingularly. Recreate the concrete case: Determinant factors cancel for a fixed partition. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Refitting can still change labels. Complete the transfer step: Check both. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Range

Begin with this mechanism: Extend the candidate set. Recreate the concrete case: Curvature can change. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Boundary choices are evidence. Complete the transfer step: Run variants. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: Name T,W,determinant method,range and second-difference rule. Recreate the concrete case: Teaching elbow supports k=3. Write T, Wk, determinants, raw ratios, first differences, centred second differences and chosen k with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call clusters real. Complete the transfer step: Bound conclusion. Change exactly one sample size, variable, scale, cluster count, scatter entry, determinant, eigenvalue, matrix condition, regularisation constant, candidate range, first difference, second difference, initialisation or seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep fit, penalty, curvature, conditioning, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a constant variable, collinearity, singular W, tiny eigenvalue, singleton cluster or boundary candidate. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.