WHY MATHEMATICS?

Why Mathematics? | Adjusted Rand Index for Clustering: Pair Agreements, Chance Correction, Label Permutations and Cluster Recovery

How two partitions become pair counts and a chance-adjusted score, with a complete six-object example, limits and practical checks.

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 the pair contingency behind the Rand index
Calculate raw and adjusted agreement by hand
Understand label invariance and negative values
Keep cluster marginals, stability and purpose visible

Chapter map

1. Objects
2. Reference labels
3. Proposed labels
4. All pairs
5. Together in both
6. Together only in reference
7. Together only in proposal
8. Apart in both
9. Raw Rand index
10. Reference combinations
11. Proposal combinations
12. Expected overlap
13. Maximum term
14. Adjusted score
15. Negative values
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

A small example makes the mechanism visible. Use six fixed objects.

A concrete situation

Keep identities aligned across labelings.

What the mathematics does not license

ARI requires the same object set. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Audit joins. 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 is where modelling choices begin to affect evidence. Write A,A,A,B,B,B.

A concrete situation

Cluster sizes are three and three.

What the mathematics does not license

Names themselves carry no order. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Freeze membership. 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 central mathematical idea is concrete. Write X,X,Y,Y,Z,Z.

A concrete situation

Cluster sizes are two, two and two.

What the mathematics does not license

Renaming X,Y,Z changes nothing. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Freeze membership. 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

Here the geometry becomes operational. Compute C(6,2).

A concrete situation

There are fifteen unordered pairs.

What the mathematics does not license

Do not count ordered duplicates. Evidence becomes stronger when the limitation is tested instead of hidden.

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. Together in both

This step deserves a careful pause. Read contingency cells of size two.

A concrete situation

C(2,2)+C(2,2)=2.

What the mathematics does not license

Singleton intersections add zero. Keeping this boundary clear makes the result easier to audit and transfer.

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.

6. Together only in reference

The useful question is not only how to calculate, but what the calculation assumes. Compare the six reference-together pairs.

A concrete situation

Four are split by the proposal.

What the mathematics does not license

This is a pairwise false split. A trustworthy workflow records the choice before seeing the final test result.

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. Together only in proposal

A small example makes the mechanism visible. Compare the three proposal-together pairs.

A concrete situation

One crosses the reference boundary.

What the mathematics does not license

This is a pairwise false merge. Evidence becomes stronger when the limitation is tested instead of hidden.

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.

8. Apart in both

This is where modelling choices begin to affect evidence. Use 15-2-4-1.

A concrete situation

Eight pairs agree as apart.

What the mathematics does not license

Large many-cluster partitions can inflate this raw agreement. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Show complement. 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. Raw Rand index

The central mathematical idea is concrete. Add two and eight, then divide by fifteen.

A concrete situation

RI=.6667.

What the mathematics does not license

Raw agreement is not chance corrected. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Name metric. 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. Reference combinations

Here the geometry becomes operational. Compute C(3,2)+C(3,2).

A concrete situation

Row-combination sum=6.

What the mathematics does not license

Cluster-size marginals drive expectation. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record marginals. 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. Proposal combinations

This step deserves a careful pause. Compute three times C(2,2).

A concrete situation

Column-combination sum=3.

What the mathematics does not license

Different cluster counts change this term. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record marginals. 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. Expected overlap

The useful question is not only how to calculate, but what the calculation assumes. Compute 6(3)/15.

A concrete situation

Expected same-same count=1.2.

What the mathematics does not license

This expectation uses fixed marginals. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

State model. 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. Maximum term

A small example makes the mechanism visible. Average 6 and 3.

A concrete situation

Maximum normalization term=4.5.

What the mathematics does not license

Do not replace with total pairs. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Show formula. 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. Adjusted score

This is where modelling choices begin to affect evidence. Compute (2-1.2)/(4.5-1.2).

A concrete situation

ARI=.2424.

What the mathematics does not license

Near zero is model-relative chance agreement. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Report precision. 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. Negative values

The central mathematical idea is concrete. Consider cross-cut partitions.

A concrete situation

ARI may fall below zero.

What the mathematics does not license

Negative is not an impossible error. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Interpret baseline. 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 object set, marginals and reference.

A concrete situation

Teaching ARI=.2424.

What the mathematics does not license

Do not call it cluster usefulness. 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

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 adjusted Rand index compares two partitions by asking what every unordered pair of objects experiences. A pair may be together in both partitions, apart in both, together only in the reference or together only in the proposed clustering. The raw Rand index counts the first two as agreements; the adjusted version subtracts the agreement expected under a declared random-partition model.

Use six objects. The reference labels are A,A,A,B,B,B and the proposed labels are X,X,Y,Y,Z,Z. Of the fifteen unordered pairs, two are together in both partitions, eight are apart in both, four are together only in the reference and one is together only in the proposal. The raw Rand index is therefore (2+8)/15=.6667.

For the Hubert–Arabie adjustment, the reference cluster sizes contribute C(3,2)+C(3,2)=6 same-cluster pairs, while the proposal sizes contribute C(2,2)+C(2,2)+C(2,2)=3. The expected same-same count is 6(3)/15=1.2 and the maximum term is (6+3)/2=4.5. ARI=(2-1.2)/(4.5-1.2)=.2424.

ARI is unchanged if cluster names are permuted because it uses memberships, not the symbols printed on labels. A score of one means identical partitions up to renaming, a score near zero is close to the chance baseline under the adjustment model, and a negative score indicates pair agreement below that baseline. None of these values proves that the variables, distance, number of clusters or real-world use are appropriate.

Frequently asked questions

What is the adjusted Rand index?

A pair-based similarity between two partitions, corrected for the agreement expected under a fixed-marginal chance model.

Why does renaming clusters not matter?

The calculation uses which objects share a cluster, not the symbols used as labels.

What does ARI=1 mean?

The two partitions are identical up to a permutation of cluster names.

Can ARI be negative?

Yes. Especially discordant partitions can agree less than the adjustment baseline.

Is zero proof of no structure?

No. It means agreement near the stated chance baseline, not absence of every pattern.

What must be reported?

Objects, both partitions, cluster marginals, pair counts, adjustment convention, uncertainty, stability and intended use.

Primary source and fact-check note

Lawrence Hubert and Phipps Arabie, “Comparing Partitions,” Journal of Classification 2, 193–218 (1985), DOI 10.1007/BF01908075. 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 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.

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 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.

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 fixed objects. Recreate the concrete case: Keep identities aligned across labelings. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: ARI requires the same object set. Complete the transfer step: Audit joins. 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: Reference labels

Begin with this clustering mechanism: Write A,A,A,B,B,B. Recreate the concrete case: Cluster sizes are three and three. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Names themselves carry no order. Complete the transfer step: Freeze membership. 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: Proposed labels

Begin with this clustering mechanism: Write X,X,Y,Y,Z,Z. Recreate the concrete case: Cluster sizes are two, two and two. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Renaming X,Y,Z changes nothing. Complete the transfer step: Freeze membership. 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: 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, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not count ordered duplicates. Complete the transfer step: List once. 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.

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: Together in both

Begin with this clustering mechanism: Read contingency cells of size two. Recreate the concrete case: C(2,2)+C(2,2)=2. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Singleton intersections add zero. Complete the transfer step: Show cells. 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: Together only in reference

Begin with this clustering mechanism: Compare the six reference-together pairs. Recreate the concrete case: Four are split by the proposal. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is a pairwise false split. Complete the transfer step: Keep count. 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: Together only in proposal

Begin with this clustering mechanism: Compare the three proposal-together pairs. Recreate the concrete case: One crosses the reference boundary. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is a pairwise false merge. Complete the transfer step: Keep count. 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: Apart in both

Begin with this clustering mechanism: Use 15-2-4-1. Recreate the concrete case: Eight pairs agree as apart. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Large many-cluster partitions can inflate this raw agreement. Complete the transfer step: Show complement. 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.

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: Raw Rand index

Begin with this clustering mechanism: Add two and eight, then divide by fifteen. Recreate the concrete case: RI=.6667. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Raw agreement is not chance corrected. Complete the transfer step: Name metric. 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: Reference combinations

Begin with this clustering mechanism: Compute C(3,2)+C(3,2). Recreate the concrete case: Row-combination sum=6. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Cluster-size marginals drive expectation. Complete the transfer step: Record marginals. 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: Proposal combinations

Begin with this clustering mechanism: Compute three times C(2,2). Recreate the concrete case: Column-combination sum=3. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different cluster counts change this term. Complete the transfer step: Record marginals. 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: Expected overlap

Begin with this clustering mechanism: Compute 6(3)/15. Recreate the concrete case: Expected same-same count=1.2. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This expectation uses fixed marginals. Complete the transfer step: State model. 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.

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: Maximum term

Begin with this clustering mechanism: Average 6 and 3. Recreate the concrete case: Maximum normalization term=4.5. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not replace with total pairs. Complete the transfer step: Show formula. 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: Adjusted score

Begin with this clustering mechanism: Compute (2-1.2)/(4.5-1.2). Recreate the concrete case: ARI=.2424. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Near zero is model-relative chance agreement. Complete the transfer step: Report precision. 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: Negative values

Begin with this clustering mechanism: Consider cross-cut partitions. Recreate the concrete case: ARI may fall below zero. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Negative is not an impossible error. Complete the transfer step: Interpret baseline. 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 object set, marginals and reference. Recreate the concrete case: Teaching ARI=.2424. Write object IDs, both label vectors, contingency cells, all pairs, pair agreements, marginal combinations, expected term and adjusted score with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call it cluster usefulness. 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.