WHY MATHEMATICS?

Why Mathematics? | BHEP Test: Multivariate Characteristic Functions, Gaussian Kernels, Affine Invariance and Consistent Normality Evidence

How vector data can be whitened, compared with a normal characteristic function at every frequency, and summarized through pairwise Gaussian kernels without losing joint geometry.

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

Centre and whiten each multivariate observation
Compare empirical and normal characteristic functions
Convert the weighted integral into Gaussian kernel terms
Calibrate the statistic and diagnose the joint departure

Chapter map

1. Rows
2. Dimension
3. Mean vector
4. Covariance
5. Whitening
6. Frequency vector
7. Empirical CF
8. Normal CF
9. Squared gap
10. Gaussian weight
11. Kernel form
12. Radius terms
13. Affine invariance
14. Calibration
15. Diagnostics
16. Claim

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

1. Rows

This is where modelling choices begin to affect evidence. Define independent vector observations.

A concrete situation

One student contributes one row.

What the mathematics does not license

Repeated rows violate independence. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Audit 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. Dimension

The central mathematical idea is concrete. Record p and n.

A concrete situation

The illustration has p=2.

What the mathematics does not license

Small n relative to p destabilizes covariance. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

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

3. Mean vector

Here the geometry becomes operational. Subtract coordinate means.

A concrete situation

Move the cloud to zero.

What the mathematics does not license

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

Try it, check it, transfer it

Freeze cohort. 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. Covariance

This step deserves a careful pause. Estimate the full matrix.

A concrete situation

Cross-products capture tilt.

What the mathematics does not license

Near singularity magnifies noise. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Inspect eigenvalues. 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. Whitening

The useful question is not only how to calculate, but what the calculation assumes. Apply S inverse one-half.

A concrete situation

Residual covariance becomes identity.

What the mathematics does not license

Different valid roots preserve distances. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Verify. 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. Frequency vector

A small example makes the mechanism visible. Let t range over R p.

A concrete situation

Each t probes a projection.

What the mathematics does not license

One chosen t is incomplete. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use integral. 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. Empirical CF

This is where modelling choices begin to affect evidence. Average exp(i t transpose yj).

A concrete situation

Every row contributes phase.

What the mathematics does not license

Complex parts both matter. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Audit. 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. Normal CF

The central mathematical idea is concrete. Use exp(-norm(t) squared/2).

A concrete situation

The benchmark is fixed after whitening.

What the mathematics does not license

It is not a p-value. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Label function. 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. Squared gap

Here the geometry becomes operational. Take modulus squared.

A concrete situation

The result is nonnegative.

What the mathematics does not license

Pointwise gaps can cancel only before squaring. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Square first. 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. Gaussian weight

This step deserves a careful pause. Apply the declared beta.

A concrete situation

Remote frequencies are downweighted.

What the mathematics does not license

Changing beta after inspection is tuning. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Predeclare. 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. Kernel form

The useful question is not only how to calculate, but what the calculation assumes. Evaluate closed-form pair terms.

A concrete situation

Distances 1,4,1 yield different similarities.

What the mathematics does not license

The toy cloud is singular. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Treat as warning. 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. Radius terms

A small example makes the mechanism visible. Add sample-to-normal components.

A concrete situation

Large radii alter contributions.

What the mathematics does not license

Radius alone misses direction. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep all terms. 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. Affine invariance

This is where modelling choices begin to affect evidence. Rotate and rescale original axes.

A concrete situation

Evidence should agree.

What the mathematics does not license

Nonlinear transforms change the hypothesis. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Document transforms. 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. Calibration

The central mathematical idea is concrete. Use matching simulated or asymptotic critical values.

A concrete situation

Preserve n,p,beta.

What the mathematics does not license

Mismatched tables invalidate inference. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record implementation. 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. Diagnostics

Here the geometry becomes operational. Map contributions to rows and pairs.

A concrete situation

Clusters may drive the result.

What the mathematics does not license

Deletion after inspection biases evidence. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Investigate provenance. 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. State joint normal-model evidence.

A concrete situation

The whitened cloud conflicts with normality.

What the mathematics does not license

No cause is proven. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Use design knowledge. 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 Baringhaus–Henze–Epps–Pulley test extends empirical-characteristic-function normality checking to vector data. It centres and whitens the observations, compares their empirical characteristic function with the standard multivariate normal function, and integrates the squared gap under a Gaussian weight.

For standardized two-dimensional points z1=(-1,0), z2=(0,0), z3=(1,0), the distinct squared pairwise distances are 1,4 and 1. A Gaussian kernel turns the shorter pairs into larger similarities than the distant pair. This cloud is singular in two dimensions, so it is deliberately a warning example: the ordinary full-rank test cannot proceed without enough non-collinear observations.

The statistic has a closed form made from pairwise kernel terms, single-point radius terms and a normal-model constant. Standardization makes it affine invariant when covariance is nonsingular. A tuning parameter controls the Gaussian weight and therefore the balance between broad and fine distribution features.

Consistency means that, under suitable conditions, power approaches one against every fixed non-normal alternative as sample size grows. It does not mean every small dataset reveals its departure. Report n, dimension, covariance conditioning, tuning rule, calibration and plots that locate tails, clusters or nonlinear structure.

Frequently asked questions

What does BHEP stand for?

It refers to the Baringhaus–Henze extension of the Epps–Pulley characteristic-function idea.

What does the test assess?

It assesses a non-degenerate multivariate normal model for independent vector observations.

Why whiten the data?

Whitening removes fitted location, scale and linear coordinate choice.

What does the tuning parameter control?

It controls the Gaussian frequency weight and the test’s sensitivity profile.

What does consistency mean?

Power approaches one against fixed alternatives as sample size grows under the method’s conditions.

Can a singular covariance be used directly?

No. The ordinary full-rank construction needs an invertible sample covariance matrix.

Primary source and fact-check note

Ludwig Baringhaus and Norbert Henze, “A Consistent Test for Multivariate Normality Based on the Empirical Characteristic Function,” Metrika 35, 339–348 (1988), DOI 10.1007/BF02613322. 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: topic-specific validation and responsible use

Design audit

Freeze independent units, target variables and inclusion rules.

Standardization audit

Reproduce centres, scales, covariance roots and projections.

Statistic audit

Show every distance, kernel, moment, direction or weight contribution.

Calibration audit

Match sample size, dimension, tuning, fitting and optimization.

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.

Graph audit

Preserve scatterplots, distance plots, direction curves and influence tables.

Magnitude audit

Explain the practical consequence in original units.

Sensitivity audit

Use only prespecified tuning and influence checks.

Final claim

Keep inference inside the model and sampling design.

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: Rows

Begin with this mechanism: Define independent vector observations. Recreate the concrete case: One student contributes one row. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Repeated rows violate independence. Complete the transfer step: Audit IDs. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 2: Dimension

Begin with this mechanism: Record p and n. Recreate the concrete case: The illustration has p=2. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Small n relative to p destabilizes covariance. Complete the transfer step: Report both. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 3: Mean vector

Begin with this mechanism: Subtract coordinate means. Recreate the concrete case: Move the cloud to zero. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Missing rows move all coordinates. Complete the transfer step: Freeze cohort. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 4: Covariance

Begin with this mechanism: Estimate the full matrix. Recreate the concrete case: Cross-products capture tilt. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Near singularity magnifies noise. Complete the transfer step: Inspect eigenvalues. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support 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: Whitening

Begin with this mechanism: Apply S inverse one-half. Recreate the concrete case: Residual covariance becomes identity. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different valid roots preserve distances. Complete the transfer step: Verify. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 6: Frequency vector

Begin with this mechanism: Let t range over R p. Recreate the concrete case: Each t probes a projection. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One chosen t is incomplete. Complete the transfer step: Use integral. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 7: Empirical CF

Begin with this mechanism: Average exp(i t transpose yj). Recreate the concrete case: Every row contributes phase. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Complex parts both matter. Complete the transfer step: Audit. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 8: Normal CF

Begin with this mechanism: Use exp(-norm(t) squared/2). Recreate the concrete case: The benchmark is fixed after whitening. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: It is not a p-value. Complete the transfer step: Label function. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support 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: Squared gap

Begin with this mechanism: Take modulus squared. Recreate the concrete case: The result is nonnegative. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Pointwise gaps can cancel only before squaring. Complete the transfer step: Square first. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 10: Gaussian weight

Begin with this mechanism: Apply the declared beta. Recreate the concrete case: Remote frequencies are downweighted. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Changing beta after inspection is tuning. Complete the transfer step: Predeclare. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 11: Kernel form

Begin with this mechanism: Evaluate closed-form pair terms. Recreate the concrete case: Distances 1,4,1 yield different similarities. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The toy cloud is singular. Complete the transfer step: Treat as warning. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 12: Radius terms

Begin with this mechanism: Add sample-to-normal components. Recreate the concrete case: Large radii alter contributions. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Radius alone misses direction. Complete the transfer step: Keep all terms. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support 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: Affine invariance

Begin with this mechanism: Rotate and rescale original axes. Recreate the concrete case: Evidence should agree. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Nonlinear transforms change the hypothesis. Complete the transfer step: Document transforms. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 14: Calibration

Begin with this mechanism: Use matching simulated or asymptotic critical values. Recreate the concrete case: Preserve n,p,beta. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Mismatched tables invalidate inference. Complete the transfer step: Record implementation. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 15: Diagnostics

Begin with this mechanism: Map contributions to rows and pairs. Recreate the concrete case: Clusters may drive the result. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Deletion after inspection biases evidence. Complete the transfer step: Investigate provenance. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: State joint normal-model evidence. Recreate the concrete case: The whitened cloud conflicts with normality. Write rows, dimensions, means, covariance matrices, whitened residuals, frequencies, characteristic functions, Gaussian kernels, radii, tuning parameters and calibrated statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: No cause is proven. Complete the transfer step: Use design knowledge. Change exactly one observation, row, tie, covariance entry, whitening rule, distance, frequency weight, direction-grid resolution, median, MAD, tuning constant, kernel parameter, null approximation or bootstrap seed. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep design, transformation, statistic, calibration, magnitude, interpretation and validation on separate notebook lines. Explain the result twice: first to a student using the visible geometry or weight table, then to a parent using the real decision and original units. Add a counterexample in which the same statistic arises from heavy tails, clusters, curvature, rounding, censoring or dependence. State what extra graph, design record or held-out observation would distinguish the explanations. Finish by writing one sentence the data support, one sentence they do not support and one useful next measurement.