Chapter map

1. 1. Dimension
2. 2. Distortion one
3. 3. Distortion two
4. 4. Distortion three
5. 5. Distortion four
6. 6. Transform one
7. 7. Transform two
8. 8. Transform three
9. 9. Transform four
10. 10. Jump two
11. 11. Jump three
12. 12. Jump four
13. 13. Selection
14. 14. Dimension sensitivity
15. 15. Restarts
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. Dimension

This is where modelling choices begin to affect evidence. Set p=2.

A concrete situation

Transformation exponent is -p/2=-1.

What the mathematics does not license

Effective dimension may differ from column count. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State p. 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. Distortion one

The central mathematical idea is concrete. Use d1=10.

A concrete situation

One-centre distortion is largest.

What the mathematics does not license

Define distance and averaging. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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.

---

3. Distortion two

Here the geometry becomes operational. Use d2=4.

A concrete situation

Two centres sharply reduce distortion.

What the mathematics does not license

Same preprocessing is required. The same reasoning helps students move from an exercise to a real dataset.

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.

---

4. Distortion three

This step deserves a careful pause. Use d3=3.

A concrete situation

Improvement continues.

What the mathematics does not license

Raw distortion always tends downward. That distinction matters because a correct formula can still answer the wrong operational question.

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.

---

5. Distortion four

The useful question is not only how to calculate, but what the calculation assumes. Use d4=2.8.

A concrete situation

Improvement is now small.

What the mathematics does not license

Candidate range must be declared. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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.

---

6. Transform one

A small example makes the mechanism visible. Compute 10^-1.

A concrete situation

Value=.1.

What the mathematics does not license

Use the dimension-dependent power. The same reasoning helps students move from an exercise to a real dataset.

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.

---

7. Transform two

This is where modelling choices begin to affect evidence. Compute 4^-1.

A concrete situation

Value=.25.

What the mathematics does not license

Do not use negative distortion. That distinction matters because a correct formula can still answer the wrong operational question.

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.

---

8. Transform three

The central mathematical idea is concrete. Compute 3^-1.

A concrete situation

Value=.333333.

What the mathematics does not license

Retain precision. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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.

---

9. Transform four

Here the geometry becomes operational. Compute 2.8^-1.

A concrete situation

Value=.357143.

What the mathematics does not license

Retain precision. The same reasoning helps students move from an exercise to a real dataset.

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.

---

10. Jump two

This step deserves a careful pause. Subtract .1 from .25.

A concrete situation

J2=.15.

What the mathematics does not license

Index jump at its destination k. That distinction matters because a correct formula can still answer the wrong operational question.

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.

---

11. Jump three

The useful question is not only how to calculate, but what the calculation assumes. Subtract .25 from .333333.

A concrete situation

J3=.083333.

What the mathematics does not license

Keep sign. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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.

---

12. Jump four

A small example makes the mechanism visible. Subtract .333333 from .357143.

A concrete situation

J4=.023810.

What the mathematics does not license

Round last. The same reasoning helps students move from an exercise to a real dataset.

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.

---

13. Selection

This is where modelling choices begin to affect evidence. Find largest jump.

A concrete situation

Teaching choice k=2.

What the mathematics does not license

One large jump needs stability checks. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State winner. 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. Dimension sensitivity

The central mathematical idea is concrete. Try another p.

A concrete situation

Powers and jumps change.

What the mathematics does not license

Dimension is not a cosmetic input. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Compare. 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. Restarts

Here the geometry becomes operational. Refit clustering repeatedly.

A concrete situation

Distortion distribution becomes visible.

What the mathematics does not license

Local minima can invent jumps. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report best and spread. 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

This step deserves a careful pause. Name p,distance,distortion,algorithm and range.

A concrete situation

Teaching jump .15 at two.

What the mathematics does not license

Do not promise useful labels. That distinction matters because a correct formula can still answer the wrong operational question.

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 Sugar–James jump method transforms average distortion so that a genuine reduction in unexplained structure becomes a visible jump. If d(k) is distortion and p is the effective dimension, a common transformation is d(k)^(-p/2). The jump J(k) is the transformed value at k minus the value at k-1, and the largest jump identifies the candidate count.

Use p=2 with distortions d1=10, d2=4, d3=3 and d4=2.8. The exponent is -1, so the transformed sequence is .1, .25, .333333 and .357143. The jump from one to two clusters is .25-.1=.15.

The next jumps are .333333-.25=.083333 and .357143-.333333=.023810. The largest is .15 at k=2, so the teaching example selects two clusters. The method looks for a sharp transformed improvement rather than a subjective bend in the raw distortion curve.

Dimension is central: changing p changes the power and can change the winning jump. Distortion definition, covariance structure, scaling, cluster algorithm, candidate range and random starts also matter. A large jump supports a parsimonious representation under those choices; it does not certify semantic labels or stability.

---

Frequently asked questions

What is the jump method?

A cluster-count rule that finds the largest increase in inverse-power transformed distortion.

Why transform distortion?

Rate-distortion theory motivates a scale where genuine structure can appear as a sharp jump.

What exponent is used?

Commonly -p/2, where p is the declared effective dimension.

How is k chosen?

Select the candidate at the largest successive transformed jump.

Can the dimension choice change the answer?

Yes. It changes every transformed value and jump.

What must be reported?

Distortion definition, distance, p, algorithm, scaling, starts, candidate range, transformed sequence, jumps and sensitivity checks.

---

Primary source and fact-check note

Catherine A. Sugar and Gareth M. James, “Finding the Number of Clusters in a Dataset: An Information-Theoretic Approach,” Journal of the American Statistical Association 98(463), 750–763 (2003), DOI 10.1198/016214503000000666. 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

• Krzanowski–Lai dimension correction
• Gap statistic and reference distributions
• Prediction strength
• How AI finds patterns without labels
• eduKateSingapore learning routes

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: cluster-count evidence and responsible use

Data audit

Freeze observations, features, units, exclusions, missing values, transformations, sample size and seeds.

Fit audit

Use comparable algorithms, distances, covariance models, convergence rules and random-start budgets across candidates.

Formula audit

Show W sequences, dimension powers, distortion transforms, successive ratios, likelihoods, parameter counts and penalties.

Convention audit

Name effective dimension, indexing, threshold inequality, BIC sign orientation and higher-or-lower selection direction.

Uncertainty audit

Repeat starts, samples, candidate ranges, feature scales, covariance forms and plausible preprocessing choices.

Failure audit

Test near-zero denominators, negative corrected differences, local minima, singular mixtures, outliers and high dimensions.

Meaning audit

Compare numerical stopping evidence with stability and domain evidence without assuming clusters are natural or useful.

Final claim

State exact dispersion, distortion or model-comparison evidence and stop before causal, semantic or guaranteed-usefulness claims.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Dimension

Begin with this cluster-count mechanism: Set p=2. Recreate the concrete case: Transformation exponent is -p/2=-1. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Effective dimension may differ from column count. Complete the transfer step: State p. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Distortion one

Begin with this cluster-count mechanism: Use d1=10. Recreate the concrete case: One-centre distortion is largest. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Define distance and averaging. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Distortion two

Begin with this cluster-count mechanism: Use d2=4. Recreate the concrete case: Two centres sharply reduce distortion. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Same preprocessing is required. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Distortion three

Begin with this cluster-count mechanism: Use d3=3. Recreate the concrete case: Improvement continues. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Raw distortion always tends downward. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 5: Distortion four

Begin with this cluster-count mechanism: Use d4=2.8. Recreate the concrete case: Improvement is now small. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Candidate range must be declared. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Transform one

Begin with this cluster-count mechanism: Compute 10^-1. Recreate the concrete case: Value=.1. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Use the dimension-dependent power. Complete the transfer step: Show step. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Transform two

Begin with this cluster-count mechanism: Compute 4^-1. Recreate the concrete case: Value=.25. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not use negative distortion. Complete the transfer step: Show step. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Transform three

Begin with this cluster-count mechanism: Compute 3^-1. Recreate the concrete case: Value=.333333. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Retain precision. Complete the transfer step: Show step. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 9: Transform four

Begin with this cluster-count mechanism: Compute 2.8^-1. Recreate the concrete case: Value=.357143. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Retain precision. Complete the transfer step: Show step. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Jump two

Begin with this cluster-count mechanism: Subtract .1 from .25. Recreate the concrete case: J2=.15. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Index jump at its destination k. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Jump three

Begin with this cluster-count mechanism: Subtract .25 from .333333. Recreate the concrete case: J3=.083333. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Keep sign. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Jump four

Begin with this cluster-count mechanism: Subtract .333333 from .357143. Recreate the concrete case: J4=.023810. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Round last. Complete the transfer step: Record value. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 13: Selection

Begin with this cluster-count mechanism: Find largest jump. Recreate the concrete case: Teaching choice k=2. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One large jump needs stability checks. Complete the transfer step: State winner. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Dimension sensitivity

Begin with this cluster-count mechanism: Try another p. Recreate the concrete case: Powers and jumps change. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Dimension is not a cosmetic input. Complete the transfer step: Compare. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Restarts

Begin with this cluster-count mechanism: Refit clustering repeatedly. Recreate the concrete case: Distortion distribution becomes visible. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Local minima can invent jumps. Complete the transfer step: Report best and spread. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. 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 cluster-count mechanism: Name p,distance,distortion,algorithm and range. Recreate the concrete case: Teaching jump .15 at two. Write p, distortions, transformed values, successive jumps, algorithm, distance, candidate range and selected maximum with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not promise useful labels. Complete the transfer step: Bound conclusion. Change exactly one feature dimension, W value, distortion, power, sample size, threshold, candidate range, likelihood, parameter count, covariance model, feature scale, distance, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep compactness, model fit, complexity, stability, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving one outlier, a near-zero denominator, a negative corrected difference, a flat distortion curve, a poor local optimum, a singular mixture or non-Gaussian clusters. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.