WHY MATHEMATICS?

Why Mathematics? | Quantile-on-Quantile Regression: Double Quantile Grids, Local Linear Slopes, Kernel Weights and Dependence Surfaces

How the theta-quantile of an outcome can respond differently near the tau-quantile of a predictor, and why local support and dependence-aware uncertainty matter.

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

Choose separate quantile grids for outcome Y and predictor X
Localise observations around each sample quantile of X
Fit kernel-weighted local linear quantile regressions
Validate bandwidth, dependence, multiplicity and effective support

Chapter map

1. Outcome index
2. Predictor index
3. Local coordinate
4. Bandwidth
5. Kernel
6. Local line
7. Predictions
8. Median loss
9. Perturbation
10. Weighted contribution
11. Surface
12. Effective sample
13. Dependence
14. Benchmark
15. Multiplicity
16. Claim

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

1. Outcome index

A small example makes the mechanism visible. Choose theta for Y.

A concrete situation

Use theta=.5.

What the mathematics does not license

Another theta changes check-loss asymmetry. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record grid. 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. Predictor index

This is where modelling choices begin to affect evidence. Choose tau for X.

A concrete situation

Use q_X(tau)=2.

What the mathematics does not license

Sample quantiles have uncertainty. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Report convention. 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. Local coordinate

The central mathematical idea is concrete. Centre X at q.

A concrete situation

Use x-q=-1,0,1.

What the mathematics does not license

Uncentred fits change intercept meaning. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show 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.

4. Bandwidth

Here the geometry becomes operational. Set h=2.

A concrete situation

Scaled distances are -.5,0,.5.

What the mathematics does not license

Bandwidth drives effective neighbours. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Tune honestly. 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. Kernel

This step deserves a careful pause. Use triangular K(u)=max(1-|u|,0).

A concrete situation

Weights are .5,1,.5.

What the mathematics does not license

Kernel edge rules matter. Keeping this boundary clear makes the result easier to audit and transfer.

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.

6. Local line

The useful question is not only how to calculate, but what the calculation assumes. Fit a+b(x-q).

A concrete situation

Try a=5,b=3.

What the mathematics does not license

Slope is local to tau. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

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

7. Predictions

A small example makes the mechanism visible. Evaluate the line.

A concrete situation

Get 2,5,8.

What the mathematics does not license

Exact fit is constructed. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Label illustration. 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. Median loss

This is where modelling choices begin to affect evidence. Use theta=.5 check loss.

A concrete situation

Total weighted loss is zero.

What the mathematics does not license

Zero training loss can overfit. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Validate elsewhere. 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. Perturbation

The central mathematical idea is concrete. Change last y from 8 to 9.

A concrete situation

Residual becomes 1.

What the mathematics does not license

One point can shape sparse tails. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Recompute. 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. Weighted contribution

Here the geometry becomes operational. Multiply kernel and check weights.

A concrete situation

Contribution is .25.

What the mathematics does not license

Some implementations normalise weights. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Publish objective. 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. Surface

This step deserves a careful pause. Repeat over theta and tau.

A concrete situation

Form a coefficient matrix.

What the mathematics does not license

Dense grids multiply noise. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Smooth cautiously. 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. Effective sample

The useful question is not only how to calculate, but what the calculation assumes. Sum and square local weights.

A concrete situation

Inspect concentration.

What the mathematics does not license

Nominal n overstates local support. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Report n effective. 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. Dependence

A small example makes the mechanism visible. Preserve time or cluster structure.

A concrete situation

Use block resampling for a series.

What the mathematics does not license

IID errors can be false. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Mirror 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.

14. Benchmark

This is where modelling choices begin to affect evidence. Average or compare with ordinary QR.

A concrete situation

Check consistency of summaries.

What the mathematics does not license

Agreement does not prove correctness. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Use diagnostic. 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. Multiplicity

The central mathematical idea is concrete. Account for many surface cells.

A concrete situation

Use simultaneous bands or cautious maps.

What the mathematics does not license

Cellwise stars overstate evidence. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

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

16. Claim

Here the geometry becomes operational. State variables, grids, kernel and bandwidth.

A concrete situation

Teaching perturbation contributes .25.

What the mathematics does not license

Do not turn association into causation. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

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

A seven-day practice lab

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

Questions for a student, parent or mentor

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

A method-specific audit before deployment

Quantile-on-quantile regression asks a two-index question: how does the theta-quantile of Y respond near the tau-quantile of X? Sim and Zhou combined quantile regression with local linear approximation so that the relationship can vary across both the outcome distribution and the predictor distribution. The result is a surface rather than one slope, useful when low, ordinary and high values of X may relate differently to low, ordinary and high outcomes.

Let the tau-quantile of X be q=2, use bandwidth h=2 and a triangular kernel. For x=1,2,3, the local weights are .5,1,.5. With y=2,5,8, the local line a+b(x-q) at a=5 and b=3 predicts 2,5,8 exactly, so its weighted median check loss is zero. If the last y becomes 9, the residual there is 1 and its contribution at theta=.5 is .5 kernel weight times .5 check-loss weight times 1, or .25. This calculation exposes the two weighting systems: proximity in X and asymmetry in Y.

A colourful coefficient surface can be fragile. Small bandwidths create few effective neighbours, large bandwidths blur the predictor quantile being studied, and a dense theta-by-tau grid creates a multiple-comparison problem. Serial dependence, persistence and common trends can also manufacture smooth-looking patterns. The original application studied oil-price shocks and US stock returns; its empirical findings should not be copied into a different market or treated as a universal law.

Report the definitions and transformations of X and Y, theta and tau grids, sample quantiles, kernel, bandwidth, effective local sample size, local-polynomial order, dependence-aware uncertainty and sensitivity. Compare the average over tau with ordinary quantile regression where appropriate. Use the surface to describe conditional association unless design and identification justify a causal statement.

Frequently asked questions

What are theta and tau?

Theta indexes the outcome quantile; tau indexes the predictor quantile around which the fit is localised.

Why use a kernel?

It gives more influence to observations near the selected X quantile.

Why local linear fitting?

It estimates a slope while reducing some local-constant boundary bias.

What does the surface show?

How a conditional association varies across both distributions.

Is every coloured cell independent evidence?

No. Cells share data and a dense grid raises multiplicity concerns.

What must be reported?

Variables, grids, kernel, bandwidth, effective sample, dependence, uncertainty and sensitivity.

Primary source and fact-check note

Nicholas Sim and Hongtao Zhou, “Oil Prices, US Stock Return, and the Dependence Between Their Quantiles,” Journal of Banking & Finance 55, 1–8 (2015), DOI 10.1016/j.jbankfin.2015.01.013. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

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 outcomes, reference ranks, covariates, coordinates, monitoring locations, timestamps, sampling intervals, outcome and predictor quantile grids before estimation.

Objective audit

Write the transport assignment, convex-gradient constraint, kernel-weighted check loss, spatial conditional-distribution model or trigonometric quantile objective explicitly.

Tuning audit

Record rank grid, basis, kernel, bandwidth, spatial range, knots, priors, frequency grid, quantile grid, normalisation and stopping tolerance.

Computation audit

Check primal-dual gaps, cyclic monotonicity, effective local sample sizes, covariance conditioning, chain convergence, approximation error and frequency resolution.

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.

Support audit

Inspect sparse reference corners, predictor tails, unsampled regions, map boundaries, missing time blocks, short series and off-support predictions.

Uncertainty audit

Propagate rank discretisation, sample-quantile, bandwidth, spatial-prior, hyperparameter, dependence, frequency-search and solver uncertainty through the complete workflow.

Validation audit

Use untouched joint outcomes, blocked time windows or held-out locations for multivariate scores, check loss, quantile coverage, spatial calibration and future-cycle stability.

Final claim

Translate the conditional result without turning a vector rank, dependence surface, smooth percentile map or spectral peak into a causal or universal guarantee.

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: Outcome index

Begin with this mechanism: Choose theta for Y. Recreate the concrete case: Use theta=.5. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Another theta changes check-loss asymmetry. Complete the transfer step: Record grid. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Predictor index

Begin with this mechanism: Choose tau for X. Recreate the concrete case: Use q_X(tau)=2. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Sample quantiles have uncertainty. Complete the transfer step: Report convention. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Local coordinate

Begin with this mechanism: Centre X at q. Recreate the concrete case: Use x-q=-1,0,1. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Uncentred fits change intercept meaning. Complete the transfer step: Show values. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Bandwidth

Begin with this mechanism: Set h=2. Recreate the concrete case: Scaled distances are -.5,0,.5. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Bandwidth drives effective neighbours. Complete the transfer step: Tune honestly. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 5: Kernel

Begin with this mechanism: Use triangular K(u)=max(1-|u|,0). Recreate the concrete case: Weights are .5,1,.5. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Kernel edge rules matter. Complete the transfer step: State rule. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Local line

Begin with this mechanism: Fit a+b(x-q). Recreate the concrete case: Try a=5,b=3. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Slope is local to tau. Complete the transfer step: Keep units. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Predictions

Begin with this mechanism: Evaluate the line. Recreate the concrete case: Get 2,5,8. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Exact fit is constructed. Complete the transfer step: Label illustration. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Median loss

Begin with this mechanism: Use theta=.5 check loss. Recreate the concrete case: Total weighted loss is zero. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Zero training loss can overfit. Complete the transfer step: Validate elsewhere. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 9: Perturbation

Begin with this mechanism: Change last y from 8 to 9. Recreate the concrete case: Residual becomes 1. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One point can shape sparse tails. Complete the transfer step: Recompute. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Weighted contribution

Begin with this mechanism: Multiply kernel and check weights. Recreate the concrete case: Contribution is .25. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Some implementations normalise weights. Complete the transfer step: Publish objective. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Surface

Begin with this mechanism: Repeat over theta and tau. Recreate the concrete case: Form a coefficient matrix. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Dense grids multiply noise. Complete the transfer step: Smooth cautiously. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Effective sample

Begin with this mechanism: Sum and square local weights. Recreate the concrete case: Inspect concentration. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Nominal n overstates local support. Complete the transfer step: Report n effective. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 13: Dependence

Begin with this mechanism: Preserve time or cluster structure. Recreate the concrete case: Use block resampling for a series. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: IID errors can be false. Complete the transfer step: Mirror design. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Benchmark

Begin with this mechanism: Average or compare with ordinary QR. Recreate the concrete case: Check consistency of summaries. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Agreement does not prove correctness. Complete the transfer step: Use diagnostic. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Multiplicity

Begin with this mechanism: Account for many surface cells. Recreate the concrete case: Use simultaneous bands or cautious maps. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Cellwise stars overstate evidence. Complete the transfer step: Control search. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: State variables, grids, kernel and bandwidth. Recreate the concrete case: Teaching perturbation contributes .25. Write outcomes, predictors, theta levels, tau levels, sample quantiles, centred distances, bandwidths, kernel weights, local slopes, residuals and check losses with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not turn association into causation. Complete the transfer step: Bound conclusion. Change exactly one outcome coordinate, reference rank, transport assignment, covariate basis, theta level, tau level, sample quantile, kernel weight, bandwidth, monitoring site, spatial distance, correlation range, prior, sampling interval, frequency, phase, sine coefficient, cosine coefficient or validation block. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, support, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the complete table and once to a parent with the real decision and original units. Add a counterexample involving unit-dependent transport, non-convex maps, sparse local tails, serial dependence, spatial barriers, preferential monitoring, irregular sampling, structural breaks or multiple frequency searches. State what extra measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.