Chapter map

1. 1. Objects
2. 2. Contingency table
3. 3. Reference entropy
4. 4. Proposed entropy
5. 5. Mutual information
6. 6. Lost reference information
7. 7. Gained proposal information
8. 8. Variation distance
9. 9. Equivalent formula
10. 10. Symmetry
11. 11. Identity
12. 12. Triangle inequality
13. 13. Log base
14. 14. Dataset size
15. 15. Chance
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

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

A concrete situation

Both partitions describe the same set.

What the mathematics does not license

Missing rows change probabilities. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Audit join. 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. Contingency table

A small example makes the mechanism visible. Use counts 2,1,0 and 0,1,2.

A concrete situation

Four cells are non-zero.

What the mathematics does not license

Keep zeros explicit. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Show matrix. 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. Reference entropy

This is where modelling choices begin to affect evidence. Use probabilities one half and one half.

A concrete situation

H(C)=.693147 nats.

What the mathematics does not license

Log base sets the distance unit. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

State base. 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. Proposed entropy

The central mathematical idea is concrete. Use three probabilities one third.

A concrete situation

H(K)=1.098612 nats.

What the mathematics does not license

More groups can raise entropy. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show 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.

---

5. Mutual information

Here the geometry becomes operational. Sum four non-zero cell contributions.

A concrete situation

I(C;K)=.462098 nats.

What the mathematics does not license

MI is shared information. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

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

---

6. Lost reference information

This step deserves a careful pause. Compute H(C)-I.

A concrete situation

H(C|K)=.231049.

What the mathematics does not license

This direction reflects cluster mixing. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Name direction. 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. Gained proposal information

The useful question is not only how to calculate, but what the calculation assumes. Compute H(K)-I.

A concrete situation

H(K|C)=.636514.

What the mathematics does not license

This direction reflects class fragmentation. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Name direction. 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. Variation distance

A small example makes the mechanism visible. Add both conditional entropies.

A concrete situation

VI=.867563 nats.

What the mathematics does not license

Smaller means closer partitions. 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.

---

9. Equivalent formula

This is where modelling choices begin to affect evidence. Compute H(C)+H(K)-2I.

A concrete situation

The same result is .867563.

What the mathematics does not license

Algebra should reproduce the sum. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Verify 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.

---

10. Symmetry

The central mathematical idea is concrete. Swap reference and proposal.

A concrete situation

Total VI is unchanged.

What the mathematics does not license

Directional components swap roles. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Explain 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.

---

11. Identity

Here the geometry becomes operational. Compare a partition with itself.

A concrete situation

VI=0.

What the mathematics does not license

Label names do not affect membership. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Test relabeling. 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. Triangle inequality

This step deserves a careful pause. Insert a third partition.

A concrete situation

VI(C,K)≤VI(C,T)+VI(T,K).

What the mathematics does not license

This metric property supports geometry on partitions. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

State use. 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. Log base

The useful question is not only how to calculate, but what the calculation assumes. Replace natural logs with base two.

A concrete situation

Units become bits.

What the mathematics does not license

Numbers scale by a constant. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Do not mix units. 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. Dataset size

A small example makes the mechanism visible. Compare unrelated datasets.

A concrete situation

Maximum possible entropy can differ.

What the mathematics does not license

Raw VI lacks universal cross-dataset scale. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Normalize if justified. 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. Chance

This is where modelling choices begin to affect evidence. Compare random partitions.

A concrete situation

Unadjusted VI has no zero chance baseline.

What the mathematics does not license

Cluster count can affect expected distance. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Add adjusted companion. 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 objects, table, log base and components.

A concrete situation

Teaching VI=.867563 nats.

What the mathematics does not license

Do not call low distance usefulness. 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

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

Variation of information treats disagreement between two partitions as a distance. It adds the uncertainty in the reference labels that remains after the proposed clustering is known to the uncertainty in the proposed labels that remains after the reference is known. In symbols, VI=H(C|K)+H(K|C)=H(C)+H(K)-2I(C;K). Smaller values mean closer partitions.

Use reference labels A,A,A,B,B,B and proposed labels X,X,Y,Y,Z,Z. The reference entropy is .693147 nats, the proposed entropy is 1.098612 nats and their mutual information is .462098 nats. Therefore H(C|K)=.231049 and H(K|C)=.636514.

The variation of information is .231049+.636514=.867563 nats. The two directional components tell different stories: the smaller first term reflects relatively limited reference-class mixing inside proposed clusters, while the larger second term reflects reference classes being split across several proposed clusters. The total is symmetric even though the explanatory components are directional.

VI is a true metric on partitions: it is non-negative, symmetric, zero only for identical partitions up to label names and obeys the triangle inequality. Its numerical range depends on sample size, logarithm base and entropy scale, so raw values from different datasets should not be compared casually. It is not inherently chance adjusted, and lower distance does not prove the reference labels or the resulting use are valid.

---

Frequently asked questions

What is variation of information?

A metric distance equal to H(C|K)+H(K|C), or H(C)+H(K)-2I(C;K).

Is higher or lower better?

Lower means the two partitions are closer; zero means identical membership up to label renaming.

What do the two components show?

One captures reference information lost within proposed clusters and the other captures proposal information added through splits.

Why is triangle inequality useful?

It makes VI a true metric and supports coherent comparisons through intermediate partitions.

Is VI chance adjusted?

The standard measure is not; cluster count and sample size can affect random-partition distances.

What must be reported?

Objects, contingency table, log base, both entropies, MI, two conditional entropies, VI, cluster sizes and comparison context.

---

Primary source and fact-check note

Marina Meilă, “Comparing Clusterings—An Information Based Distance,” Journal of Multivariate Analysis 98(5), 873–895 (2007), DOI 10.1016/j.jmva.2006.11.013. 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

• Normalized mutual information
• V-measure and directional entropy
• Adjusted mutual information and chance correction
• Adjusted Rand index
• eduKateSingapore learning routes

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Extended laboratory: 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: Both partitions describe the same set. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Missing rows change probabilities. Complete the transfer step: Audit join. 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: Contingency table

Begin with this clustering mechanism: Use counts 2,1,0 and 0,1,2. Recreate the concrete case: Four cells are non-zero. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Keep zeros explicit. Complete the transfer step: Show matrix. 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: Reference entropy

Begin with this clustering mechanism: Use probabilities one half and one half. Recreate the concrete case: H(C)=.693147 nats. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Log base sets the distance unit. Complete the transfer step: State base. 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: Proposed entropy

Begin with this clustering mechanism: Use three probabilities one third. Recreate the concrete case: H(K)=1.098612 nats. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: More groups can raise entropy. Complete the transfer step: Show marginals. 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: Mutual information

Begin with this clustering mechanism: Sum four non-zero cell contributions. Recreate the concrete case: I(C;K)=.462098 nats. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: MI is shared information. Complete the transfer step: Keep value. 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: Lost reference information

Begin with this clustering mechanism: Compute H(C)-I. Recreate the concrete case: H(C|K)=.231049. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This direction reflects cluster mixing. Complete the transfer step: Name direction. 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: Gained proposal information

Begin with this clustering mechanism: Compute H(K)-I. Recreate the concrete case: H(K|C)=.636514. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This direction reflects class fragmentation. Complete the transfer step: Name direction. 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: Variation distance

Begin with this clustering mechanism: Add both conditional entropies. Recreate the concrete case: VI=.867563 nats. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Smaller means closer partitions. Complete the transfer step: Cross-check. 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: Equivalent formula

Begin with this clustering mechanism: Compute H(C)+H(K)-2I. Recreate the concrete case: The same result is .867563. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Algebra should reproduce the sum. Complete the transfer step: Verify identity. 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: Symmetry

Begin with this clustering mechanism: Swap reference and proposal. Recreate the concrete case: Total VI is unchanged. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Directional components swap roles. Complete the transfer step: Explain 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 11: Identity

Begin with this clustering mechanism: Compare a partition with itself. Recreate the concrete case: VI=0. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Label names do not affect membership. Complete the transfer step: Test relabeling. 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: Triangle inequality

Begin with this clustering mechanism: Insert a third partition. Recreate the concrete case: VI(C,K)≤VI(C,T)+VI(T,K). Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This metric property supports geometry on partitions. Complete the transfer step: State use. 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: Log base

Begin with this clustering mechanism: Replace natural logs with base two. Recreate the concrete case: Units become bits. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Numbers scale by a constant. Complete the transfer step: Do not mix units. 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: Dataset size

Begin with this clustering mechanism: Compare unrelated datasets. Recreate the concrete case: Maximum possible entropy can differ. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Raw VI lacks universal cross-dataset scale. Complete the transfer step: Normalize if justified. 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: Chance

Begin with this clustering mechanism: Compare random partitions. Recreate the concrete case: Unadjusted VI has no zero chance baseline. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Cluster count can affect expected distance. Complete the transfer step: Add adjusted companion. 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 objects, table, log base and components. Recreate the concrete case: Teaching VI=.867563 nats. Write object IDs, table counts, marginals, entropies, mutual information, both conditional entropies, log base and VI with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call low distance usefulness. 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.