WHY MATHEMATICS?

Why Mathematics? | Quantile Regression Forests: Leaf Weights, Conditional Distributions, Weighted Percentiles and Tail Prediction

How tree leaf membership becomes an adaptive weighted empirical distribution, why mean and tail summaries differ, and where sparse support and calibration 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

Grow and document the forest neighbourhood mechanism
Convert shared leaves into normalised observation weights
Build the weighted conditional CDF and invert it
Validate each quantile, effective neighbours, coverage and support

Chapter map

1. Query
2. Forest
3. Leaf sharing
4. Weights
5. Responses
6. CDF at one
7. CDF at two
8. CDF at five
9. CDF at nine
10. Median
11. Upper quantiles
12. Mean
13. Effective neighbours
14. Validation
15. Extrapolation
16. Claim

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

1. Query

A small example makes the mechanism visible. Specify predictor vector x.

A concrete situation

One query receives forest weights.

What the mathematics does not license

Weights change for another x. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Keep local. 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. Forest

This is where modelling choices begin to affect evidence. Grow trees under declared sampling and split rules.

A concrete situation

Retain terminal leaf membership.

What the mathematics does not license

Tree defaults affect neighbourhoods. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record settings. 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. Leaf sharing

The central mathematical idea is concrete. Find training rows sharing leaves with the query.

A concrete situation

Aggregate across trees.

What the mathematics does not license

Sharing is not causal similarity. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Use careful language. 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. Weights

Here the geometry becomes operational. Normalise query-specific contributions.

A concrete situation

Use .10,.40,.30,.20.

What the mathematics does not license

Weights must sum to one. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Audit total. 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. Responses

This step deserves a careful pause. Sort paired responses 1,2,5,9.

A concrete situation

Preserve weights with rows.

What the mathematics does not license

Sorting weights separately is wrong. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep 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.

6. CDF at one

The useful question is not only how to calculate, but what the calculation assumes. Cumulate weight at y=1.

A concrete situation

F=0.10.

What the mathematics does not license

CDF is stepwise in finite data. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show steps. 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. CDF at two

A small example makes the mechanism visible. Add .40.

A concrete situation

F=0.50.

What the mathematics does not license

Quantile conventions matter at jumps. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

State inverse. 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. CDF at five

This is where modelling choices begin to affect evidence. Add .30.

A concrete situation

F=0.80.

What the mathematics does not license

Rounding can change a boundary. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep precision. 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. CDF at nine

The central mathematical idea is concrete. Add .20.

A concrete situation

F=1.00.

What the mathematics does not license

Every valid CDF ends at one. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Unit 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.

10. Median

Here the geometry becomes operational. Take smallest y with F at least .5.

A concrete situation

Median=2.

What the mathematics does not license

Interpolation conventions can differ. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Name 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.

11. Upper quantiles

This step deserves a careful pause. Read .8 and .9 levels.

A concrete situation

They are 5 and 9.

What the mathematics does not license

Small leaves make tails coarse. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Report resolution. 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. Mean

The useful question is not only how to calculate, but what the calculation assumes. Compute weighted response mean.

A concrete situation

Mean=4.2.

What the mathematics does not license

Mean and median answer different questions. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Do not rank blindly. 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. Effective neighbours

A small example makes the mechanism visible. Inspect sum of squared weights or leaf counts.

A concrete situation

Concentration indicates local information.

What the mathematics does not license

Nominal n can mislead. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Report n_eff. 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. Validation

This is where modelling choices begin to affect evidence. Use held-out check loss, coverage and width.

A concrete situation

Evaluate each tau.

What the mathematics does not license

Training forest fit is optimistic. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Use untouched cases. 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. Extrapolation

The central mathematical idea is concrete. Check whether query lies inside training support.

A concrete situation

Forests cannot create new response patterns reliably.

What the mathematics does not license

Leaf routing can hide extrapolation. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Flag support. 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 forest settings, weights and inverse-CDF convention.

A concrete situation

Teaching median=2.

What the mathematics does not license

Do not call weights posterior probabilities. 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 regression forests extend a regression forest from conditional means to conditional distributions. Each tree sends a new predictor vector into a terminal leaf. Training observations that repeatedly share leaves with the query receive weights. Instead of averaging their responses only, the method forms a weighted empirical distribution and reads any requested quantile from its cumulative distribution function.

Suppose four training responses 1,2,5,9 receive query-specific weights .10,.40,.30,.20. The weighted cumulative probabilities are .10,.50,.80,1.00. Using Q_tau=inf{y:F(y)>=tau}, the median is 2, the .80 quantile is 5 and the .90 quantile is 9. The weighted mean is 4.2. These summaries come from the same local distribution but answer different questions about centre and tail.

The forest weights are adaptive neighbourhood weights, not causal importance scores and not posterior probabilities that a training row is correct. Leaf size, tree count, candidate variables, subsampling and split rules influence the neighbourhood. The original consistency result relies on assumptions; finite datasets can still produce biased tails, coarse quantiles or unstable estimates in sparse predictor regions.

Use honest validation with held-out check loss, interval coverage and width. Inspect effective weight concentration, repeated response values, extrapolation, quantile crossing and subgroup calibration. Standard random forests may optimise mean-squared-error splits rather than tail-specific objectives, so strong mean prediction does not guarantee strong upper- or lower-quantile prediction.

Frequently asked questions

What does a quantile regression forest retain?

Training response values in leaves, rather than only their mean.

How are query weights formed?

By averaging shared-leaf contributions across trees.

How is a quantile obtained?

Invert the weighted empirical conditional CDF.

Are weights causal importances?

No. They are predictive neighbourhood weights.

Why inspect effective neighbours?

Concentrated weights make tail estimates coarse and unstable.

What must be reported?

Forest settings, weight rule, inverse-CDF convention, support and held-out calibration.

Primary source and fact-check note

Nicolai Meinshausen, “Quantile Regression Forests,” Journal of Machine Learning Research 7, 983–999 (2006). 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 sampling units, predictor support, time order, response orientation, missingness and dependence.

Objective audit

Write likelihood, prior, forest weights, check losses, penalties or lag equations explicitly.

Tuning audit

Record scale, priors, sampler settings, tree settings, lambda grids, tau grids and lag order.

Computation audit

Check posterior convergence, weight normalisation, solution paths, tied optima and solver tolerances.

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.

Coherence audit

Inspect quantile order, support, stability, effective neighbours, hit sequences and extrapolation.

Uncertainty audit

Use model-compatible posterior, sandwich or bootstrap methods and include tuning or repair.

Validation audit

Use untouched or rolling data for check loss, coverage, width and subgroup calibration.

Final claim

Translate the conditional tail result without causal, posterior or guarantee overclaims.

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: Query

Begin with this mechanism: Specify predictor vector x. Recreate the concrete case: One query receives forest weights. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Weights change for another x. Complete the transfer step: Keep local. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Forest

Begin with this mechanism: Grow trees under declared sampling and split rules. Recreate the concrete case: Retain terminal leaf membership. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Tree defaults affect neighbourhoods. Complete the transfer step: Record settings. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Leaf sharing

Begin with this mechanism: Find training rows sharing leaves with the query. Recreate the concrete case: Aggregate across trees. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Sharing is not causal similarity. Complete the transfer step: Use careful language. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Weights

Begin with this mechanism: Normalise query-specific contributions. Recreate the concrete case: Use .10,.40,.30,.20. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Weights must sum to one. Complete the transfer step: Audit total. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or 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: Responses

Begin with this mechanism: Sort paired responses 1,2,5,9. Recreate the concrete case: Preserve weights with rows. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Sorting weights separately is wrong. Complete the transfer step: Keep IDs. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: CDF at one

Begin with this mechanism: Cumulate weight at y=1. Recreate the concrete case: F=0.10. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: CDF is stepwise in finite data. Complete the transfer step: Show steps. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: CDF at two

Begin with this mechanism: Add .40. Recreate the concrete case: F=0.50. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Quantile conventions matter at jumps. Complete the transfer step: State inverse. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: CDF at five

Begin with this mechanism: Add .30. Recreate the concrete case: F=0.80. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Rounding can change a boundary. Complete the transfer step: Keep precision. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or 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: CDF at nine

Begin with this mechanism: Add .20. Recreate the concrete case: F=1.00. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Every valid CDF ends at one. Complete the transfer step: Unit test. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Median

Begin with this mechanism: Take smallest y with F at least .5. Recreate the concrete case: Median=2. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Interpolation conventions can differ. Complete the transfer step: Name rule. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Upper quantiles

Begin with this mechanism: Read .8 and .9 levels. Recreate the concrete case: They are 5 and 9. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Small leaves make tails coarse. Complete the transfer step: Report resolution. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Mean

Begin with this mechanism: Compute weighted response mean. Recreate the concrete case: Mean=4.2. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Mean and median answer different questions. Complete the transfer step: Do not rank blindly. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or 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: Effective neighbours

Begin with this mechanism: Inspect sum of squared weights or leaf counts. Recreate the concrete case: Concentration indicates local information. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Nominal n can mislead. Complete the transfer step: Report n_eff. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Validation

Begin with this mechanism: Use held-out check loss, coverage and width. Recreate the concrete case: Evaluate each tau. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Training forest fit is optimistic. Complete the transfer step: Use untouched cases. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Extrapolation

Begin with this mechanism: Check whether query lies inside training support. Recreate the concrete case: Forests cannot create new response patterns reliably. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Leaf routing can hide extrapolation. Complete the transfer step: Flag support. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or 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 forest settings, weights and inverse-CDF convention. Recreate the concrete case: Teaching median=2. Write predictors, responses, leaf memberships, observation weights, cumulative probabilities, conditional quantiles, means, effective neighbour counts and original measurement units with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call weights posterior probabilities. Complete the transfer step: Bound conclusion. Change exactly one observation, quantile level, ALD scale, prior, chain, tree, leaf size, forest weight, inverse-CDF convention, predictor scaling, penalty factor, lambda, selected variable, lag, structural break, solver vertex, forecast origin, bootstrap rule or validation sample. Explain which mathematical object changes, which estimator is selected, which conclusion survives and which claim must weaken. Keep design, objective, tuning, computation, coherence, 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 likelihood misspecification, sparse support, correlated predictors, time leakage, dependence, a real minority group, high leverage or measurement error. State what extra graph or measurement would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.