WHY MATHEMATICS?

Why Mathematics? | Cox–Small Test: Nonlinear Regression, Direction Search, Standardized Residual Geometry and Multivariate Normality

How multivariate normality implies linear conditional relationships, how quadratic curvature is measured, and why searching projection directions can reveal hidden joint structure.

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 the multivariate observations
Project the cloud along a unit direction
Measure quadratic regression curvature using directional moments
Maximize over directions and calibrate the search statistic

Chapter map

1. Rows
2. Centre
3. Whiten
4. Direction
5. Projection
6. Square projection
7. Weighted vector
8. Third moment
9. Fourth moment
10. Eta numerator
11. Eta denominator
12. Directional eta
13. Grid search
14. Maximum
15. Plots
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

The useful question is not only how to calculate, but what the calculation assumes. Define independent p-vectors.

A concrete situation

Each case has X and Y.

What the mathematics does not license

Repeated cases distort curvature. The same reasoning helps students move from an exercise to a real dataset.

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. Centre

A small example makes the mechanism visible. Subtract the vector mean.

A concrete situation

Place the cloud at zero.

What the mathematics does not license

Selective missingness bends shape. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Freeze inclusion. 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. Whiten

This is where modelling choices begin to affect evidence. Apply inverse covariance square root.

A concrete situation

Set covariance to identity.

What the mathematics does not license

Near singularity destabilizes directions. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Inspect rank. 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. Direction

The central mathematical idea is concrete. Choose a unit vector b.

A concrete situation

In 2D use cos theta,sin theta.

What the mathematics does not license

Non-unit b changes scale. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Normalize. 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. Projection

Here the geometry becomes operational. Calculate uj=b transpose yj.

A concrete situation

Each row becomes one score.

What the mathematics does not license

Keep original y for vector moments. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Archive 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.

6. Square projection

This step deserves a careful pause. Calculate uj squared.

A concrete situation

Curvature uses even magnitude.

What the mathematics does not license

Large projections gain influence. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show 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.

7. Weighted vector

The useful question is not only how to calculate, but what the calculation assumes. Average yj uj squared.

A concrete situation

This points toward nonlinear association.

What the mathematics does not license

A few rows can dominate. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Map contributions. 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. Third moment

A small example makes the mechanism visible. Average uj cubed.

A concrete situation

Directional skewness is removed appropriately.

What the mathematics does not license

Sign errors alter numerator. That distinction matters because a correct formula can still answer the wrong operational question.

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.

9. Fourth moment

This is where modelling choices begin to affect evidence. Average uj fourth.

A concrete situation

This enters the normalization.

What the mathematics does not license

Heavy tails shrink or inflate denominator. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Inspect. 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. Eta numerator

The central mathematical idea is concrete. Subtract the squared third-moment part from vector norm squared.

A concrete situation

Keep nonnegative geometry.

What the mathematics does not license

Floating error may produce tiny negatives. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use tolerance. 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. Eta denominator

Here the geometry becomes operational. Combine fourth moment, one and skewness correction.

A concrete situation

It standardizes available variation.

What the mathematics does not license

Near-zero denominator is unstable. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Flag it. 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. Directional eta

This step deserves a careful pause. Divide numerator by denominator.

A concrete situation

One theta yields one value.

What the mathematics does not license

It is not the final statistic. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Label 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.

13. Grid search

The useful question is not only how to calculate, but what the calculation assumes. Evaluate many theta values.

A concrete situation

Start every five degrees.

What the mathematics does not license

Coarse grids miss narrow maxima. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Refine. 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. Maximum

A small example makes the mechanism visible. Take the largest directional value.

A concrete situation

Record maximizing direction.

What the mathematics does not license

Optimization overfits noise under null. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Use matching critical values. 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. Plots

This is where modelling choices begin to affect evidence. Graph regression along maximizing pair.

A concrete situation

Quadratic curvature becomes visible.

What the mathematics does not license

A smooth curve is descriptive. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Add uncertainty. 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. State nonlinearity evidence.

A concrete situation

The whitened cloud shows conditional curvature.

What the mathematics does not license

General non-normality cause is not proven. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Compare omnibus test. 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 Cox–Small approach treats multivariate normality as a question about linear conditional relationships. In a multivariate normal distribution, regression of one linear combination on another is linear. The test searches directions for the strongest quadratic curvature after centring and whitening the data.

Let b be a unit direction and u_j=b transpose y_j for standardized residual vectors y_j. The method forms second-, third- and fourth-order directional moments, measures the part of the vector weighted quadratic term that lies beyond the direction b, and standardizes it by the remaining fourth-moment variation. The final statistic maximizes this quantity over all directions.

For a two-dimensional illustration, examine unit directions b=(cos theta,sin theta). A numerical search might evaluate theta every five degrees, then refine near the largest value. This explains the optimization but does not reproduce a continuum maximum or provide a p-value. Resolution and the optimization algorithm belong in the audit trail.

The test is designed to expose nonlinear regression structure rather than every possible departure from normality. A large value can arise from curvature, clusters, outliers or measurement limits. Use a scatterplot matrix, component-plus-residual views and influence checks, and compare with an omnibus method if the decision needs broad coverage.

Frequently asked questions

What feature does Cox–Small target?

It targets nonlinear regression relationships among linear combinations of variables.

Why is linear regression relevant?

Multivariate normal conditional means are linear.

What is optimized?

A standardized curvature measure is maximized over unit projection directions.

Why whiten first?

Whitening provides affine-invariant standardized geometry.

Is a grid search exact?

No. It approximates the direction maximum and needs resolution checks.

Is it an omnibus test against every alternative?

No. It is especially interpretable for nonlinear-dependence departures.

Primary source and fact-check note

D. R. Cox and N. J. H. Small, “Testing Multivariate Normality,” Biometrika 65(2), 263–272 (1978), DOI 10.1093/biomet/65.2.263. 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 p-vectors. Recreate the concrete case: Each case has X and Y. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Repeated cases distort curvature. 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: Centre

Begin with this mechanism: Subtract the vector mean. Recreate the concrete case: Place the cloud at zero. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Selective missingness bends shape. Complete the transfer step: Freeze inclusion. 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: Whiten

Begin with this mechanism: Apply inverse covariance square root. Recreate the concrete case: Set covariance to identity. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Near singularity destabilizes directions. Complete the transfer step: Inspect rank. 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: Direction

Begin with this mechanism: Choose a unit vector b. Recreate the concrete case: In 2D use cos theta,sin theta. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Non-unit b changes scale. Complete the transfer step: Normalize. 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: Projection

Begin with this mechanism: Calculate uj=b transpose yj. Recreate the concrete case: Each row becomes one score. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Keep original y for vector moments. Complete the transfer step: Archive 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 6: Square projection

Begin with this mechanism: Calculate uj squared. Recreate the concrete case: Curvature uses even magnitude. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Large projections gain influence. Complete the transfer step: Show 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.

Workshop 7: Weighted vector

Begin with this mechanism: Average yj uj squared. Recreate the concrete case: This points toward nonlinear association. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A few rows can dominate. Complete the transfer step: Map contributions. 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: Third moment

Begin with this mechanism: Average uj cubed. Recreate the concrete case: Directional skewness is removed appropriately. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Sign errors alter numerator. 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.

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: Fourth moment

Begin with this mechanism: Average uj fourth. Recreate the concrete case: This enters the normalization. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Heavy tails shrink or inflate denominator. Complete the transfer step: Inspect. 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: Eta numerator

Begin with this mechanism: Subtract the squared third-moment part from vector norm squared. Recreate the concrete case: Keep nonnegative geometry. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Floating error may produce tiny negatives. Complete the transfer step: Use tolerance. 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: Eta denominator

Begin with this mechanism: Combine fourth moment, one and skewness correction. Recreate the concrete case: It standardizes available variation. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Near-zero denominator is unstable. Complete the transfer step: Flag it. 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: Directional eta

Begin with this mechanism: Divide numerator by denominator. Recreate the concrete case: One theta yields one value. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: It is not the final statistic. Complete the transfer step: Label direction. 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: Grid search

Begin with this mechanism: Evaluate many theta values. Recreate the concrete case: Start every five degrees. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Coarse grids miss narrow maxima. Complete the transfer step: Refine. 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: Maximum

Begin with this mechanism: Take the largest directional value. Recreate the concrete case: Record maximizing direction. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Optimization overfits noise under null. Complete the transfer step: Use matching critical values. 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: Plots

Begin with this mechanism: Graph regression along maximizing pair. Recreate the concrete case: Quadratic curvature becomes visible. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A smooth curve is descriptive. Complete the transfer step: Add uncertainty. 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 nonlinearity evidence. Recreate the concrete case: The whitened cloud shows conditional curvature. Write rows, dimensions, whitened vectors, unit directions, projections, squared projections, vector moments, third moments, fourth moments, directional ratios, maxima and calibrated critical values with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: General non-normality cause is not proven. Complete the transfer step: Compare omnibus test. 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.