WHY MATHEMATICS?

Why Mathematics? | Fowlkes–Mallows Index for Clustering: Pair Precision, Pair Recall, Geometric Means and Hierarchical Cuts

How unordered-pair recovery compares two clusterings, with a complete six-object example, invariance, hierarchy and honest validation.

Why mathematics is important becomes easier to see when a real decision depends on it. This guide follows the mechanism from definitions and dimensions to a worked calculation, validation, practical learning and responsible use. It keeps factual evidence separate from interpretation and treats every illustrative scenario as an example rather than a promise of results.

Choose a reading route

Build pair true positives, false positives and false negatives
Calculate pair precision and pair recall
Use their geometric mean as Fowlkes–Mallows
Understand label invariance, imbalance and cut selection

Chapter map

1. Objects
2. Reference labels
3. Proposal labels
4. All pairs
5. True positives
6. False positives
7. False negatives
8. True negatives
9. Pair precision
10. Pair recall
11. Geometric mean
12. Equivalent form
13. Label permutation
14. Imbalance
15. Hierarchy
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

The useful question is not only how to calculate, but what the calculation assumes. Use six aligned object IDs.

A concrete situation

Both label vectors cover the same objects.

What the mathematics does not license

Join errors corrupt every pair. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Audit alignment. 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

A small example makes the mechanism visible. Write A,A,A,B,B,B.

A concrete situation

Reference cluster sizes are three and three.

What the mathematics does not license

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

Try it, check it, transfer it

Freeze 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. Proposal labels

This is where modelling choices begin to affect evidence. Write X,X,Y,Y,Z,Z.

A concrete situation

Proposal sizes are two, two and two.

What the mathematics does not license

Renaming clusters changes nothing. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Freeze 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.

4. All pairs

The central mathematical idea is concrete. Compute C(6,2).

A concrete situation

There are fifteen unordered pairs.

What the mathematics does not license

FM does not use all categories equally. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

List once. 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.

Did You Know? In this chapter, the most important number is useful only together with its construction rule. Changing scaling, sample membership or validation design can change the numerical answer without changing the arithmetic.

5. True positives

Here the geometry becomes operational. Find pairs together in both.

A concrete situation

TP=2.

What the mathematics does not license

These are recovered same-cluster relations. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Show pairs. 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. False positives

This step deserves a careful pause. Find pairs together only in proposal.

A concrete situation

FP=1.

What the mathematics does not license

This is an unwanted merge under the reference. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show 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.

7. False negatives

The useful question is not only how to calculate, but what the calculation assumes. Find pairs together only in reference.

A concrete situation

FN=4.

What the mathematics does not license

These are unwanted splits. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show pairs. 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. True negatives

A small example makes the mechanism visible. Use the remaining eight pairs.

A concrete situation

TN=8.

What the mathematics does not license

TN does not enter FM. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Keep context. 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. Pair precision

This is where modelling choices begin to affect evidence. Compute 2/(2+1).

A concrete situation

Precision=2/3=.6667.

What the mathematics does not license

Precision alone ignores missed reference pairs. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Keep component. 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. Pair recall

The central mathematical idea is concrete. Compute 2/(2+4).

A concrete situation

Recall=1/3=.3333.

What the mathematics does not license

Recall alone ignores extra merges. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep component. 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. Geometric mean

Here the geometry becomes operational. Compute sqrt[(2/3)(1/3)].

A concrete situation

FM=sqrt(2/9)=.4714.

What the mathematics does not license

Round after the square root. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Cross-check. 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. Equivalent form

This step deserves a careful pause. Compute 2/sqrt(3×6).

A concrete situation

The same value is .4714.

What the mathematics does not license

Denominators reveal degenerate cases. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show identity. 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. Label permutation

The useful question is not only how to calculate, but what the calculation assumes. Rename A,B or X,Y,Z bijectively.

A concrete situation

FM is unchanged.

What the mathematics does not license

Object membership must remain fixed. The same reasoning helps students move from an exercise to a real dataset.

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.

14. Imbalance

A small example makes the mechanism visible. Make one reference cluster very large.

A concrete situation

Same-cluster pair totals grow quadratically.

What the mathematics does not license

Pair weighting may emphasise large groups. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Report sizes. 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. Hierarchy

This is where modelling choices begin to affect evidence. Evaluate several dendrogram cut levels.

A concrete situation

FM becomes a curve over cuts.

What the mathematics does not license

Choosing and testing on the same reference can overfit. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

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

16. Claim

The central mathematical idea is concrete. Name reference, cut, pair counts and convention.

A concrete situation

Teaching FM=.4714.

What the mathematics does not license

Do not call it chance corrected. The same reasoning helps students move from an exercise to a real dataset.

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

Day 1: define the decision, observation unit, X variables and Y or reference condition.
Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
Day 3: build a transparent baseline and write what it assumes.
Day 4: fit the main method inside a training-only pipeline; record every parameter.
Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
Day 6: compare validation metrics with the practical workload created by errors.
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 Fowlkes–Mallows index compares two clusterings through unordered pairs. It treats a pair placed together in both partitions as a true positive, together only in the proposed partition as a false positive, and together only in the reference as a false negative. The score is the geometric mean of pair precision and pair recall, so it rewards a proposal that avoids both unwanted merges and unwanted splits.

Use six aligned objects. The reference labels are A,A,A,B,B,B and the proposal is X,X,Y,Y,Z,Z. There are two pairs together in both partitions, one pair together only in the proposal and four pairs together only in the reference. Pair precision is 2/(2+1)=2/3, while pair recall is 2/(2+4)=1/3.

The Fowlkes–Mallows index is sqrt[(2/3)(1/3)]=sqrt(2/9)=.4714. Equivalently, it is TP/sqrt[(TP+FP)(TP+FN)]=2/sqrt(3×6). True-negative pairs do not enter the formula. That choice prevents the many pairs separated by both clusterings from dominating the score when there are many clusters, but it also means the index answers a narrower pair-recovery question.

Renaming clusters leaves the result unchanged because pair membership does not depend on label symbols. The unadjusted score has no fixed-marginal chance correction, class-like cluster imbalance affects pair totals, and degenerate denominators need a declared convention. In hierarchical clustering, the score can be traced across cut levels; selecting the best cut and reporting it on the same data requires honest separation from external validation.

Frequently asked questions

What is the Fowlkes–Mallows index?

The geometric mean of pair precision and pair recall for two partitions of the same objects.

What pairs enter the formula?

Pairs together in both, together only in the proposal and together only in the reference; true-negative pairs do not enter.

Why use a geometric mean?

It rewards balance and falls when either pair precision or pair recall is weak.

Does relabelling clusters change it?

No. A bijective renaming keeps every together-or-apart relation unchanged.

Is it corrected for chance?

The common score is unadjusted; adjusted Rand or adjusted mutual information answer different chance-baseline questions.

What must be reported?

Objects, both partitions, cluster sizes, TP, FP, FN, precision, recall, score, hierarchy cut, denominator convention and uncertainty.

Primary source and fact-check note

Edward B. Fowlkes and Colin L. Mallows, “A Method for Comparing Two Hierarchical Clusterings,” Journal of the American Statistical Association 78(383), 553–569 (1983), DOI 10.1080/01621459.1983.10478008. 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

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.

Representation audit

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

Partition audit

Record algorithm, parameters, random seed, cluster count, noise handling, labels and cluster-size marginals.

Formula audit

Show dispersion sums, degrees of freedom, extreme distances, pair counts, contingency cells, entropies and every denominator.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Convention audit

Name centre, scatter, metric, Dunn numerator and denominator, hierarchy cut, logarithm base, beta and edge-case rules.

Stability audit

Repeat across seeds, bootstrap samples, perturbations, feature sets, outlier rules and plausible cluster counts.

Meaning audit

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

Final claim

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

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Objects

Begin with this clustering mechanism: Use six aligned object IDs. Recreate the concrete case: Both label vectors cover the same objects. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Join errors corrupt every pair. Complete the transfer step: Audit alignment. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. 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: Reference cluster sizes are three and three. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Names carry no numeric order. Complete the transfer step: Freeze labels. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Proposal labels

Begin with this clustering mechanism: Write X,X,Y,Y,Z,Z. Recreate the concrete case: Proposal sizes are two, two and two. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Renaming clusters changes nothing. Complete the transfer step: Freeze labels. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: All pairs

Begin with this clustering mechanism: Compute C(6,2). Recreate the concrete case: There are fifteen unordered pairs. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: FM does not use all categories equally. Complete the transfer step: List once. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 5: True positives

Begin with this clustering mechanism: Find pairs together in both. Recreate the concrete case: TP=2. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: These are recovered same-cluster relations. Complete the transfer step: Show pairs. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: False positives

Begin with this clustering mechanism: Find pairs together only in proposal. Recreate the concrete case: FP=1. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is an unwanted merge under the reference. Complete the transfer step: Show pair. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: False negatives

Begin with this clustering mechanism: Find pairs together only in reference. Recreate the concrete case: FN=4. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: These are unwanted splits. Complete the transfer step: Show pairs. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: True negatives

Begin with this clustering mechanism: Use the remaining eight pairs. Recreate the concrete case: TN=8. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: TN does not enter FM. Complete the transfer step: Keep context. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 9: Pair precision

Begin with this clustering mechanism: Compute 2/(2+1). Recreate the concrete case: Precision=2/3=.6667. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Precision alone ignores missed reference pairs. Complete the transfer step: Keep component. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Pair recall

Begin with this clustering mechanism: Compute 2/(2+4). Recreate the concrete case: Recall=1/3=.3333. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Recall alone ignores extra merges. Complete the transfer step: Keep component. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Geometric mean

Begin with this clustering mechanism: Compute sqrt[(2/3)(1/3)]. Recreate the concrete case: FM=sqrt(2/9)=.4714. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Round after the square root. Complete the transfer step: Cross-check. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Equivalent form

Begin with this clustering mechanism: Compute 2/sqrt(3×6). Recreate the concrete case: The same value is .4714. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Denominators reveal degenerate cases. Complete the transfer step: Show identity. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 13: Label permutation

Begin with this clustering mechanism: Rename A,B or X,Y,Z bijectively. Recreate the concrete case: FM is unchanged. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Object membership must remain fixed. Complete the transfer step: Test invariance. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. 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 reference cluster very large. Recreate the concrete case: Same-cluster pair totals grow quadratically. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Pair weighting may emphasise large groups. Complete the transfer step: Report sizes. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Hierarchy

Begin with this clustering mechanism: Evaluate several dendrogram cut levels. Recreate the concrete case: FM becomes a curve over cuts. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Choosing and testing on the same reference can overfit. Complete the transfer step: Separate evaluation. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. 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 reference, cut, pair counts and convention. Recreate the concrete case: Teaching FM=.4714. Write object IDs, both label vectors, cluster sizes, all pairs, TP, FP, FN, TN, pair precision, pair recall and geometric mean with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call it chance corrected. Complete the transfer step: Bound conclusion. Change exactly one object assignment, cluster count, feature scale, outlier, squared distance, degree of freedom, extreme pair, diameter, pair count, contingency cell, logarithm base, beta, hierarchy cut or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep external agreement, internal geometry, stability and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving an outlier-shifted mean, a bridge point, many tiny clusters, identical clusters with renamed labels, an imbalanced reference or curved density structure. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.