Chapter map

1. 1. Representation
2. 2. Partition
3. 3. First centre
4. 4. Second centre
5. 5. First scatter
6. 6. Second scatter
7. 7. Centre separation
8. 8. Pair similarity
9. 9. Worst rival
10. 10. Final mean
11. 11. Translation
12. 12. Uniform scaling
13. 13. Unequal scaling
14. 14. Curved clusters
15. 15. Outliers
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. Representation

A small example makes the mechanism visible. Place points at 0,2,6,8.

A concrete situation

Use one Euclidean feature.

What the mathematics does not license

Unequal feature scales change geometry. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Freeze transform. 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. Partition

This is where modelling choices begin to affect evidence. Group {0,2} and {6,8}.

A concrete situation

There are two clusters.

What the mathematics does not license

DB evaluates the supplied partition. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record labels. 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. First centre

The central mathematical idea is concrete. Average 0 and 2.

A concrete situation

Centroid c1=1.

What the mathematics does not license

Another centre definition changes the measure. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Name centre. 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. Second centre

Here the geometry becomes operational. Average 6 and 8.

A concrete situation

Centroid c2=7.

What the mathematics does not license

Keep the same rule for every cluster. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Name centre. 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. First scatter

This step deserves a careful pause. Average distances |0-1| and |2-1|.

A concrete situation

S1=1.

What the mathematics does not license

Scatter formula must be declared. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Show terms. 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. Second scatter

The useful question is not only how to calculate, but what the calculation assumes. Average distances |6-7| and |8-7|.

A concrete situation

S2=1.

What the mathematics does not license

Outliers can increase scatter. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show terms. 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. Centre separation

A small example makes the mechanism visible. Compute |7-1|.

A concrete situation

M12=6.

What the mathematics does not license

Coincident centres create a zero denominator. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Check geometry. 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. Pair similarity

This is where modelling choices begin to affect evidence. Compute (S1+S2)/M12.

A concrete situation

R12=2/6=.3333.

What the mathematics does not license

Lower means less scatter relative to separation. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep ratio. 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. Worst rival

The central mathematical idea is concrete. For each cluster, take maximum Rij.

A concrete situation

With two clusters, each maximum is .3333.

What the mathematics does not license

More clusters create different worst rivals. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Name pair. 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. Final mean

Here the geometry becomes operational. Average the two maxima.

A concrete situation

DB=.3333.

What the mathematics does not license

Lower is better only within comparable setups. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Report range. 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. Translation

This step deserves a careful pause. Add ten to every point.

A concrete situation

DB is unchanged.

What the mathematics does not license

Distances ignore common translation. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Verify. 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. Uniform scaling

The useful question is not only how to calculate, but what the calculation assumes. Multiply every point by five.

A concrete situation

Scatter and separation both multiply by five.

What the mathematics does not license

The ratio stays .3333. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show cancellation. 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. Unequal scaling

A small example makes the mechanism visible. Stretch only one feature in multidimensional data.

A concrete situation

Centres and scatters change unevenly.

What the mathematics does not license

Standardisation is a modelling choice. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Document it. 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. Curved clusters

This is where modelling choices begin to affect evidence. Apply the centroid measure to moons or rings.

A concrete situation

A poor score may reflect shape mismatch.

What the mathematics does not license

Internal indices encode geometry preferences. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Inspect plot. 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. Outliers

The central mathematical idea is concrete. Move one point far from its centre.

A concrete situation

Scatter and possibly centroid change.

What the mathematics does not license

Robust companion analysis may be needed. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Run sensitivity. 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

Here the geometry becomes operational. Name centres, scatter, distance and comparison set.

A concrete situation

Teaching DB=.3333.

What the mathematics does not license

Do not call it external validity. Evidence becomes stronger when the limitation is tested instead of hidden.

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 Davies–Bouldin index evaluates an internal partition through cluster scatter and separation. For each cluster, compute an average distance from its members to a representative centre. For every other cluster, add the two scatters and divide by the distance between the centres. Each cluster keeps its largest similarity ratio—the most troublesome rival—and the final index averages those worst-pair values.

Place two one-dimensional clusters at {0,2} and {6,8}. Their centroids are 1 and 7. Mean absolute Euclidean distance from each cluster’s points to its centroid is one, while the centroid distance is six. The only pair ratio is (1+1)/6=.3333, so the two-cluster Davies–Bouldin index is .3333.

Lower values indicate compact clusters whose most similar rivals are farther away under the chosen geometry. The score has no universal pass mark. It can favour spherical centroid-based structure, respond strongly to outliers through scatter or centres, and miss density-connected or curved groups that a method such as DBSCAN is designed to find.

Translations leave Euclidean distances unchanged and multiplying every coordinate by the same positive constant leaves the ratio unchanged because numerator and denominator scale together. Unequal feature scaling can still change the geometry. Students should report features, scaling, metric, centre definition, scatter formula, cluster sizes, outlier treatment and comparison range alongside the index.

---

Frequently asked questions

What is the Davies–Bouldin index?

The mean, across clusters, of each cluster’s largest scatter-to-separation similarity with another cluster.

Is higher or lower better?

Lower is better within a comparable feature, distance, centre and scatter setup.

Why is there no universal pass mark?

The scale and behaviour depend on data geometry, cluster definition and preprocessing.

Is it invariant to scaling?

Uniform positive scaling cancels in the ratio, while unequal feature scaling can change the result.

What structures can it misread?

Curved, density-connected, non-centroid or heavily outlier-contaminated clusters.

What must be reported?

Features, scaling, distance, centres, scatter formula, cluster sizes, worst rivals, final mean, outlier checks and comparison range.

---

Primary source and fact-check note

David L. Davies and Donald W. Bouldin, “A Cluster Separation Measure,” IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-1(2), 224–227 (1979), DOI 10.1109/TPAMI.1979.4766909. 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

• Silhouette coefficient and sample-level structure
• Feature scaling and distance geometry
• DBSCAN density clustering
• How AI finds patterns without labels
• 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: clustering evidence and responsible use

Object audit

Freeze the evaluated objects, exclusions, missing-value rules, duplicate handling and the row-to-row join between labelings.

Representation audit

Document features, units, transformations, distance or dissimilarity and any train-only scaling before clustering.

Partition audit

Record the algorithm, parameters, random seed, cluster count, noise handling, label vector and cluster-size marginals.

Formula audit

Show pair counts, contingency cells, entropies, mutual information, cohesion, rival distance, scatter, centres and every denominator.

Convention audit

Name the chance model, NMI normalizer, logarithm base, singleton rule, metric, centre and scatter definition.

Stability audit

Repeat across seeds, bootstrap samples, perturbations, feature sets and plausible numbers of clusters.

Meaning audit

Compare numeric clusters with domain evidence without claiming that mathematical separation creates natural kinds.

Final claim

State the exact partition agreement or internal geometry evidence and stop before claims of truth, fairness, causation or guaranteed usefulness.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Representation

Begin with this clustering mechanism: Place points at 0,2,6,8. Recreate the concrete case: Use one Euclidean feature. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Unequal feature scales change geometry. Complete the transfer step: Freeze transform. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Partition

Begin with this clustering mechanism: Group {0,2} and {6,8}. Recreate the concrete case: There are two clusters. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: DB evaluates the supplied partition. Complete the transfer step: Record labels. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: First centre

Begin with this clustering mechanism: Average 0 and 2. Recreate the concrete case: Centroid c1=1. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Another centre definition changes the measure. Complete the transfer step: Name centre. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Second centre

Begin with this clustering mechanism: Average 6 and 8. Recreate the concrete case: Centroid c2=7. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Keep the same rule for every cluster. Complete the transfer step: Name centre. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 5: First scatter

Begin with this clustering mechanism: Average distances |0-1| and |2-1|. Recreate the concrete case: S1=1. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Scatter formula must be declared. Complete the transfer step: Show terms. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Second scatter

Begin with this clustering mechanism: Average distances |6-7| and |8-7|. Recreate the concrete case: S2=1. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Outliers can increase scatter. Complete the transfer step: Show terms. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Centre separation

Begin with this clustering mechanism: Compute |7-1|. Recreate the concrete case: M12=6. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Coincident centres create a zero denominator. Complete the transfer step: Check geometry. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Pair similarity

Begin with this clustering mechanism: Compute (S1+S2)/M12. Recreate the concrete case: R12=2/6=.3333. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Lower means less scatter relative to separation. Complete the transfer step: Keep ratio. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 9: Worst rival

Begin with this clustering mechanism: For each cluster, take maximum Rij. Recreate the concrete case: With two clusters, each maximum is .3333. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: More clusters create different worst rivals. Complete the transfer step: Name pair. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Final mean

Begin with this clustering mechanism: Average the two maxima. Recreate the concrete case: DB=.3333. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Lower is better only within comparable setups. Complete the transfer step: Report range. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Translation

Begin with this clustering mechanism: Add ten to every point. Recreate the concrete case: DB is unchanged. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Distances ignore common translation. Complete the transfer step: Verify. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Uniform scaling

Begin with this clustering mechanism: Multiply every point by five. Recreate the concrete case: Scatter and separation both multiply by five. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The ratio stays .3333. Complete the transfer step: Show cancellation. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 13: Unequal scaling

Begin with this clustering mechanism: Stretch only one feature in multidimensional data. Recreate the concrete case: Centres and scatters change unevenly. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Standardisation is a modelling choice. Complete the transfer step: Document it. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Curved clusters

Begin with this clustering mechanism: Apply the centroid measure to moons or rings. Recreate the concrete case: A poor score may reflect shape mismatch. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Internal indices encode geometry preferences. Complete the transfer step: Inspect plot. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Outliers

Begin with this clustering mechanism: Move one point far from its centre. Recreate the concrete case: Scatter and possibly centroid change. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Robust companion analysis may be needed. Complete the transfer step: Run sensitivity. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. 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 clustering mechanism: Name centres, scatter, distance and comparison set. Recreate the concrete case: Teaching DB=.3333. Write features, units, labels, cluster centres, within-cluster scatters, centre distances, pair ratios, worst rivals and the final index with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call it external validity. Complete the transfer step: Bound conclusion. Change exactly one object assignment, cluster size, contingency cell, entropy normalizer, logarithm base, pairwise distance, feature scale, rival cluster, centroid, scatter rule, outlier or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, sampling stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving relabelled but identical clusters, many tiny clusters, one negative silhouette sample hidden by a strong mean, curved density clusters or an outlier that moves a centroid. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.