Chapter map

1. 1. Objects
2. 2. Reference labels
3. 3. Proposed labels
4. 4. Contingency table
5. 5. Cluster X winner
6. 6. Cluster Y winner
7. 7. Cluster Z winner
8. 8. Purity numerator
9. 9. Purity
10. 10. Class A capture
11. 11. Class B capture
12. 12. Inverse numerator
13. 13. Inverse purity
14. 14. Imbalance
15. 15. Label permutation
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. Objects

Here the geometry becomes operational. Use six aligned objects.

A concrete situation

Reference and proposal cover the same rows.

What the mathematics does not license

Duplicates distort counts. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Audit identifiers. 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. Reference labels

This step deserves a careful pause. Write A,A,A,B,B,B.

A concrete situation

Two reference classes have size three.

What the mathematics does not license

Reference labels are evidence, not automatic truth. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record source. 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. Proposed labels

The useful question is not only how to calculate, but what the calculation assumes. Write X,X,Y,Y,Z,Z.

A concrete situation

Three clusters have size two.

What the mathematics does not license

Names carry no order. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Freeze partition. 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. Contingency table

A small example makes the mechanism visible. Cross classes with clusters.

A concrete situation

Rows are [2,1,0] and [0,1,2].

What the mathematics does not license

Keep the entire table visible. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show cells. 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. Cluster X winner

This is where modelling choices begin to affect evidence. Find its largest class count.

A concrete situation

max(2,0)=2.

What the mathematics does not license

Only the winning cell enters purity. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Mark 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.

---

6. Cluster Y winner

The central mathematical idea is concrete. Find max(1,1).

A concrete situation

Winner count=1 with a tie.

What the mathematics does not license

A tie needs a declared rule only for assigning a name. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Keep count. 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. Cluster Z winner

Here the geometry becomes operational. Find max(0,2).

A concrete situation

Winner count=2.

What the mathematics does not license

Minority counts still matter to readers. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Mark 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.

---

8. Purity numerator

This step deserves a careful pause. Add column maxima.

A concrete situation

2+1+2=5.

What the mathematics does not license

The score discards losing-cell detail. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show sum. 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. Purity

The useful question is not only how to calculate, but what the calculation assumes. Divide five by six.

A concrete situation

Purity=.8333.

What the mathematics does not license

Singleton clusters can force purity to one. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Report k. 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. Class A capture

A small example makes the mechanism visible. Find its largest cluster cell.

A concrete situation

max(2,1,0)=2.

What the mathematics does not license

This reverses the direction. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Mark cell. 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. Class B capture

This is where modelling choices begin to affect evidence. Find max(0,1,2).

A concrete situation

Largest cell=2.

What the mathematics does not license

Fragmentation reduces the maximum. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Mark cell. 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. Inverse numerator

The central mathematical idea is concrete. Add row maxima.

A concrete situation

2+2=4.

What the mathematics does not license

Different from purity numerator. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Show sum. 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. Inverse purity

Here the geometry becomes operational. Divide four by six.

A concrete situation

Inverse purity=.6667.

What the mathematics does not license

One giant cluster can force inverse purity to one. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report 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.

---

14. Imbalance

This step deserves a careful pause. Make one class dominate the sample.

A concrete situation

Majority counts can dominate averages.

What the mathematics does not license

Report class sizes and macro companions. The same reasoning helps students move from an exercise to a real dataset.

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.

---

15. Label permutation

The useful question is not only how to calculate, but what the calculation assumes. Rename rows or columns.

A concrete situation

Both scores stay unchanged.

What the mathematics does not license

Membership must remain fixed. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Test invariance. 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

A small example makes the mechanism visible. Name table, winners, k and class sizes.

A concrete situation

Teaching scores .8333 and .6667.

What the mathematics does not license

Do not report purity alone. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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

Cluster purity asks a simple majority-label question. For each proposed cluster, find the most frequent reference class, add those winning counts and divide by the number of objects. Inverse purity reverses the direction: for each reference class, find the proposed cluster containing the largest share of that class, add those counts and divide by the total. The pair exposes both mixing and fragmentation.

Use reference labels A,A,A,B,B,B and proposed labels X,X,Y,Y,Z,Z. The contingency table is A:[2,1,0] and B:[0,1,2]. Column maxima are 2,1,2, so purity=(2+1+2)/6=5/6=.8333. Row maxima are 2 and 2, so inverse purity=(2+2)/6=4/6=.6667.

High purity says proposed clusters are internally dominated by single reference classes, but it can reach one by putting every object in its own cluster. High inverse purity says each reference class is mostly captured by one proposed cluster, but it can reach one by putting all objects in one cluster. Reporting both reveals why a single majority-count score can reward a useless extreme.

Purity ignores minority structure beyond the winning cell in each cluster, so two very different contingency tables can share the same score. It is label-permutation invariant but not adjusted for chance and can be influenced by class imbalance. Students should show the full table, class and cluster sizes, winning cells, number of clusters, both directional scores and a balanced companion measure before drawing conclusions.

---

Frequently asked questions

What is cluster purity?

The sum of the largest reference-class count inside each proposed cluster, divided by the number of objects.

What is inverse purity?

The sum, over reference classes, of the largest count captured by any one proposed cluster, divided by the total.

Why report both?

Purity detects mixing while inverse purity exposes fragmentation, and each alone rewards a degenerate extreme.

Can purity equal one for a bad solution?

Yes. Making every object its own cluster gives purity one.

Can inverse purity equal one for a bad solution?

Yes. Putting all objects in one cluster captures every class in one place.

What must be reported?

Full contingency table, class and cluster sizes, winning cells, ties, cluster count, both directional scores and balanced companions.

---

Primary source and fact-check note

Christopher D. Manning, Prabhakar Raghavan and Hinrich Schütze, Introduction to Information Retrieval, Chapter 16: Evaluation of Clustering (Cambridge University Press, 2008). 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

• V-measure and split-merge balance
• Fowlkes–Mallows pair precision and recall
• Adjusted Rand index
• How AI finds patterns without labels
• Mathematics World at Bukit Timah Tutor

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 evaluated objects, exclusions, missing-value rules, duplicates and the join between reference and proposed labels.

Contingency audit

Show every cell, row total, column total, class size, cluster size and any tie before compressing evidence into a score.

Formula audit

Show observed and expected information, conditional entropies, majority cells, item intersections and every denominator.

Convention audit

Name logarithm base, entropy normalizer, chance model, self-inclusion, averaging weights, label model and edge-case rules.

Degeneracy audit

Test singleton clusters, one giant cluster, zero entropy, empty cells, class imbalance and denominator limits.

Stability audit

Repeat across seeds, bootstrap samples, perturbations, label uncertainty and plausible numbers of clusters.

Meaning audit

Compare numeric agreement with domain evidence without assuming that reference classes are complete, causal or fair.

Final claim

State exact partition agreement or distance evidence and stop before claims of truth, causation or guaranteed usefulness.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Objects

Begin with this clustering mechanism: Use six aligned objects. Recreate the concrete case: Reference and proposal cover the same rows. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Duplicates distort counts. Complete the transfer step: Audit identifiers. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Reference labels

Begin with this clustering mechanism: Write A,A,A,B,B,B. Recreate the concrete case: Two reference classes have size three. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Reference labels are evidence, not automatic truth. Complete the transfer step: Record source. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Proposed labels

Begin with this clustering mechanism: Write X,X,Y,Y,Z,Z. Recreate the concrete case: Three clusters have size two. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Names carry no order. Complete the transfer step: Freeze partition. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Contingency table

Begin with this clustering mechanism: Cross classes with clusters. Recreate the concrete case: Rows are [2,1,0] and [0,1,2]. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Keep the entire table visible. Complete the transfer step: Show cells. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 5: Cluster X winner

Begin with this clustering mechanism: Find its largest class count. Recreate the concrete case: max(2,0)=2. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Only the winning cell enters purity. Complete the transfer step: Mark winner. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Cluster Y winner

Begin with this clustering mechanism: Find max(1,1). Recreate the concrete case: Winner count=1 with a tie. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A tie needs a declared rule only for assigning a name. Complete the transfer step: Keep count. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Cluster Z winner

Begin with this clustering mechanism: Find max(0,2). Recreate the concrete case: Winner count=2. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Minority counts still matter to readers. Complete the transfer step: Mark winner. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Purity numerator

Begin with this clustering mechanism: Add column maxima. Recreate the concrete case: 2+1+2=5. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The score discards losing-cell detail. Complete the transfer step: Show sum. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 9: Purity

Begin with this clustering mechanism: Divide five by six. Recreate the concrete case: Purity=.8333. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Singleton clusters can force purity to one. Complete the transfer step: Report k. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Class A capture

Begin with this clustering mechanism: Find its largest cluster cell. Recreate the concrete case: max(2,1,0)=2. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This reverses the direction. Complete the transfer step: Mark cell. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Class B capture

Begin with this clustering mechanism: Find max(0,1,2). Recreate the concrete case: Largest cell=2. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Fragmentation reduces the maximum. Complete the transfer step: Mark cell. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Inverse numerator

Begin with this clustering mechanism: Add row maxima. Recreate the concrete case: 2+2=4. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different from purity numerator. Complete the transfer step: Show sum. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 13: Inverse purity

Begin with this clustering mechanism: Divide four by six. Recreate the concrete case: Inverse purity=.6667. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One giant cluster can force inverse purity to one. Complete the transfer step: Report both. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Imbalance

Begin with this clustering mechanism: Make one class dominate the sample. Recreate the concrete case: Majority counts can dominate averages. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Report class sizes and macro companions. Complete the transfer step: Run sensitivity. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Label permutation

Begin with this clustering mechanism: Rename rows or columns. Recreate the concrete case: Both scores stay unchanged. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Membership must remain fixed. Complete the transfer step: Test invariance. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. 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 table, winners, k and class sizes. Recreate the concrete case: Teaching scores .8333 and .6667. Write object IDs, reference classes, proposed clusters, full table, column winners, row winners, class sizes, cluster sizes and both scores with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not report purity alone. Complete the transfer step: Bound conclusion. Change exactly one object, cluster, class, contingency cell, marginal, cluster count, logarithm base, expected-information model, normalizer, winning cell, tie, per-object intersection, averaging weight or label uncertainty. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep raw agreement, chance correction, directional error, metric distance, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving singleton clusters, one giant cluster, severe imbalance, renamed labels, overlapping membership or an uncertain reference. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.