Chapter map

1. 1. Original cluster
2. 2. Bootstrap match
3. 3. Intersection
4. 4. Union
5. 5. Jaccard
6. 6. Second resample
7. 7. Third resample
8. 8. Fourth resample
9. 9. Fifth resample
10. 10. Mean
11. 11. Average
12. 12. Recovery count
13. 13. Dissolution count
14. 14. Resampling design
15. 15. Cluster-wise view
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. Original cluster

This is where modelling choices begin to affect evidence. Set C={1,2,3,4}.

A concrete situation

Four indexed members define the reference set.

What the mathematics does not license

Values and identities are different concepts. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record IDs. 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. Bootstrap match

The central mathematical idea is concrete. Set D={1,2,3,5}.

A concrete situation

Choose the most similar resampled cluster.

What the mathematics does not license

State how duplicates and absent IDs are handled. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record set. 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. Intersection

Here the geometry becomes operational. Compute C intersect D.

A concrete situation

Members {1,2,3}; size 3.

What the mathematics does not license

Count shared identities once. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show set. 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. Union

This step deserves a careful pause. Compute C union D.

A concrete situation

Members {1,2,3,4,5}; size 5.

What the mathematics does not license

Union is not 4+4. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Show set. 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. Jaccard

The useful question is not only how to calculate, but what the calculation assumes. Divide 3 by 5.

A concrete situation

Similarity=.6.

What the mathematics does not license

A label match is irrelevant. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record 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. Second resample

A small example makes the mechanism visible. Use maximum J=.80.

A concrete situation

This cluster recovers strongly in that run.

What the mathematics does not license

Maximum requires comparing all candidates. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep match. 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. Third resample

This is where modelling choices begin to affect evidence. Use maximum J=.75.

A concrete situation

This meets an inclusive .75 rule.

What the mathematics does not license

State inclusivity. 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.

---

8. Fourth resample

The central mathematical idea is concrete. Use maximum J=.60.

A concrete situation

Partial recovery remains.

What the mathematics does not license

Do not discard inconvenient runs. The habit of checking assumptions is one of the practical benefits of learning mathematics.

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.

---

9. Fifth resample

Here the geometry becomes operational. Use maximum J=.90.

A concrete situation

This is the strongest recovery.

What the mathematics does not license

One strong run cannot define stability. The same reasoning helps students move from an exercise to a real dataset.

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.

---

10. Mean

This step deserves a careful pause. Add .80+.75+.60+.90.

A concrete situation

Total=3.05.

What the mathematics does not license

Preserve precision. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

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

---

11. Average

The useful question is not only how to calculate, but what the calculation assumes. Divide 3.05 by 4.

A concrete situation

Mean=.7625.

What the mathematics does not license

Also inspect distribution. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report mean. 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. Recovery count

A small example makes the mechanism visible. Count values at least .75.

A concrete situation

Three of four recover.

What the mathematics does not license

Threshold is a convention. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

State rule. 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. Dissolution count

This is where modelling choices begin to affect evidence. Count values below .50.

A concrete situation

Zero of four dissolve.

What the mathematics does not license

Other fields use other cut-offs. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

State rule. 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. Resampling design

The central mathematical idea is concrete. Compare bootstrap, subsetting and jitter.

A concrete situation

Perturbations answer different questions.

What the mathematics does not license

Duplicate points can form mini-clusters. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Name design. 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. Cluster-wise view

Here the geometry becomes operational. Repeat for every original cluster.

A concrete situation

Weak clusters remain visible.

What the mathematics does not license

Do not average them away. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report all. 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

This step deserves a careful pause. Name method, B, matching and thresholds.

A concrete situation

Teaching mean=.7625.

What the mathematics does not license

Do not equate stable with useful. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

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

---

A seven-day practice lab

1. Day 1: define the decision, observation unit, X variables and Y or reference condition.
2. Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
3. Day 3: build a transparent baseline and write what it assumes.
4. Day 4: fit the main method inside a training-only pipeline; record every parameter.
5. Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
6. Day 6: compare validation metrics with the practical workload created by errors.
7. Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

• Which part of the model is a mathematical definition, and which part is a choice?
• What small calculation would catch a sign, scale or dimension error?
• What evidence comes from training, validation and the untouched test?
• What benign condition could look like a fault or prediction failure?
• How would the answer change for another response, asset, school project or time period?
• What should a human review before acting on the score?

A method-specific audit before deployment

Cluster-wise bootstrap stability evaluates each original cluster separately. Repeated bootstrap samples are reclustered, every original cluster is compared with the most similar resampled cluster, and Jaccard similarity records intersection divided by union. This prevents a strong whole-partition average from hiding one dissolving group.

Let the original cluster be C={1,2,3,4} and its best bootstrap match be D={1,2,3,5}. Their intersection has three members and their union has five, so J(C,D)=3/5=.6. The label names do not matter; set membership does.

Suppose four resamples give maximum Jaccard similarities .80, .75, .60 and .90. Their mean is 3.05/4=.7625. Three of four meet an inclusive .75 recovery convention, while none falls below .50 under a dissolution convention. Those cut-offs are reporting choices, not mathematical constants.

Duplicates created by ordinary bootstrapping, unmatched observations, noise clusters, cluster-size imbalance and rigid algorithms can change the result. Report resampling design, matching rule, repetition count, full distribution and thresholds. A stable cluster can still be unhelpful, and an unstable cluster can reveal a genuine boundary that the sample estimates poorly.

---

Frequently asked questions

What is cluster-wise bootstrap stability?

A resampling analysis that matches each original cluster to its most similar resampled cluster.

Why use Jaccard similarity?

It measures shared members divided by all members appearing in either set.

Why take the best resampled match?

Cluster labels are arbitrary and one original group may correspond most closely to any resampled label.

Are .50 and .75 universal thresholds?

No. Dissolution and recovery cut-offs are conventions that must be declared and justified.

Can a rigid method look stable?

Yes. Stability can reflect algorithmic inflexibility as well as data structure.

What must be reported?

Resampling scheme, B, preprocessing, algorithm, identity handling, matching rule, per-cluster distributions, thresholds and sensitivity checks.

---

Primary source and fact-check note

Christian Hennig, “Cluster-wise Assessment of Cluster Stability,” Computational Statistics & Data Analysis 52(1), 258–271 (2007), DOI 10.1016/j.csda.2006.11.025. 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

• Prediction strength and weakest clusters
• Consensus clustering matrix
• Silhouette coefficient
• Dunn 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

Data audit

Freeze observations, identities, features, units, exclusions, missing values, duplicates, transformations and random seeds.

Geometry audit

Document distance, scaling, centres, cluster count, sampling fraction, perturbation design and every pair or membership matrix.

Formula audit

Show co-sampling denominators, intersections, unions, squared memberships, weighted distances, centre gaps and final ratios.

Convention audit

Name CDF or PAC intervals, best-match rules, recovery thresholds, fuzzifier, normalization and lower-is-better direction.

Uncertainty audit

Repeat resamples, bootstrap draws, random starts, candidate counts, feature scales and plausible preprocessing choices.

Failure audit

Test rare pairs, duplicate draws, tiny clusters, rigid algorithms, local optima, outliers and coincident centres.

Meaning audit

Compare numerical reproducibility or crispness with domain evidence without assuming groups are natural, fair or useful.

Final claim

State exactly what reproduced or separated and stop before causal, semantic or guaranteed-usefulness claims.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Original cluster

Begin with this clustering mechanism: Set C={1,2,3,4}. Recreate the concrete case: Four indexed members define the reference set. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Values and identities are different concepts. Complete the transfer step: Record IDs. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Bootstrap match

Begin with this clustering mechanism: Set D={1,2,3,5}. Recreate the concrete case: Choose the most similar resampled cluster. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: State how duplicates and absent IDs are handled. Complete the transfer step: Record set. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Intersection

Begin with this clustering mechanism: Compute C intersect D. Recreate the concrete case: Members {1,2,3}; size 3. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Count shared identities once. Complete the transfer step: Show set. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Union

Begin with this clustering mechanism: Compute C union D. Recreate the concrete case: Members {1,2,3,4,5}; size 5. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Union is not 4+4. Complete the transfer step: Show set. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 5: Jaccard

Begin with this clustering mechanism: Divide 3 by 5. Recreate the concrete case: Similarity=.6. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A label match is irrelevant. Complete the transfer step: Record value. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Second resample

Begin with this clustering mechanism: Use maximum J=.80. Recreate the concrete case: This cluster recovers strongly in that run. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Maximum requires comparing all candidates. Complete the transfer step: Keep match. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Third resample

Begin with this clustering mechanism: Use maximum J=.75. Recreate the concrete case: This meets an inclusive .75 rule. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: State inclusivity. Complete the transfer step: Keep value. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Fourth resample

Begin with this clustering mechanism: Use maximum J=.60. Recreate the concrete case: Partial recovery remains. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not discard inconvenient runs. Complete the transfer step: Keep value. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 9: Fifth resample

Begin with this clustering mechanism: Use maximum J=.90. Recreate the concrete case: This is the strongest recovery. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One strong run cannot define stability. Complete the transfer step: Keep value. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Mean

Begin with this clustering mechanism: Add .80+.75+.60+.90. Recreate the concrete case: Total=3.05. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Preserve precision. Complete the transfer step: Show sum. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Average

Begin with this clustering mechanism: Divide 3.05 by 4. Recreate the concrete case: Mean=.7625. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Also inspect distribution. Complete the transfer step: Report mean. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Recovery count

Begin with this clustering mechanism: Count values at least .75. Recreate the concrete case: Three of four recover. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Threshold is a convention. Complete the transfer step: State rule. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 13: Dissolution count

Begin with this clustering mechanism: Count values below .50. Recreate the concrete case: Zero of four dissolve. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Other fields use other cut-offs. Complete the transfer step: State rule. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Resampling design

Begin with this clustering mechanism: Compare bootstrap, subsetting and jitter. Recreate the concrete case: Perturbations answer different questions. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Duplicate points can form mini-clusters. Complete the transfer step: Name design. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Cluster-wise view

Begin with this clustering mechanism: Repeat for every original cluster. Recreate the concrete case: Weak clusters remain visible. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not average them away. Complete the transfer step: Report all. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. 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 method, B, matching and thresholds. Recreate the concrete case: Teaching mean=.7625. Write original members, resampled members, intersections, unions, per-run best Jaccard scores, means, quantiles and thresholds with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not equate stable with useful. Complete the transfer step: Bound conclusion. Change exactly one resampling fraction, bootstrap draw, candidate cluster count, pair denominator, matching rule, ambiguity interval, recovery threshold, membership degree, fuzzifier, feature scale, distance, centre, initialization or random seed. Explain which mathematical object changes, which score moves and which interpretation must weaken. Keep reproducibility, crispness, compactness, separation, uncertainty and practical usefulness on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a uniform cloud, one outlier, a tiny cluster, duplicate draws, rare co-sampling, a rigid algorithm, evenly spread memberships or coincident centres. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.