Chapter map

1. 1. Time order
2. 2. Target
3. 3. Lag model
4. 4. Candidate A
5. 5. Residuals A
6. 6. Loss A
7. 7. Candidate B
8. 8. Residuals B
9. 9. Loss B
10. 10. Non-uniqueness
11. 11. Hit sequence
12. 12. Stability
13. 13. Breaks
14. 14. Uncertainty
15. 15. Forecasting
16. 16. Claim

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

---

1. Time order

A small example makes the mechanism visible. Keep observations chronological.

A concrete situation

Lags 1,2,4,3 map to next 2,4,3,6.

What the mathematics does not license

Shuffling destroys dynamics. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Preserve index. 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. Target

This is where modelling choices begin to affect evidence. Choose tau=.5.

A concrete situation

Model conditional median.

What the mathematics does not license

Other tails can behave differently. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Fit 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.

---

3. Lag model

The central mathematical idea is concrete. Write alpha+phi Y_previous.

A concrete situation

Use one lag.

What the mathematics does not license

Omitted lags can leave dependence. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Check order. 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. Candidate A

Here the geometry becomes operational. Set alpha=0,phi=2.

A concrete situation

Predictions 2,4,8,6.

What the mathematics does not license

Large phi may be unstable dynamically. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

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

---

5. Residuals A

This step deserves a careful pause. Subtract predictions.

A concrete situation

Get 0,0,-5,0.

What the mathematics does not license

One large negative residual remains visible. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep sign. 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. Loss A

The useful question is not only how to calculate, but what the calculation assumes. Sum median check loss.

A concrete situation

Total=2.5.

What the mathematics does not license

Training loss is not forecast skill. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Use rolling 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.

---

7. Candidate B

A small example makes the mechanism visible. Set alpha=1,phi=1.

A concrete situation

Predictions 2,3,5,4.

What the mathematics does not license

Different dynamics can tie. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Compare path. 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. Residuals B

This is where modelling choices begin to affect evidence. Get 0,1,-2,2.

A concrete situation

Spread differs from candidate A.

What the mathematics does not license

Same loss hides residual structure. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Plot hits. 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. Loss B

The central mathematical idea is concrete. Sum median check loss.

A concrete situation

Total=2.5.

What the mathematics does not license

Tie requires reporting. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Do not pick silently. 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. Non-uniqueness

Here the geometry becomes operational. Recognise flat objective regions.

A concrete situation

Both candidates are optima among the comparison and LP confirms minimum 2.5.

What the mathematics does not license

Software returns one vertex. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Search range. 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. Hit sequence

This step deserves a careful pause. Record I(Y_t below fitted quantile).

A concrete situation

Median hits should balance conditionally.

What the mathematics does not license

Overall 50 percent is insufficient. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Test serial pattern. 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. Stability

The useful question is not only how to calculate, but what the calculation assumes. Assess alpha and phi dynamics by tau.

A concrete situation

Large tail phi may imply persistence.

What the mathematics does not license

OLS stationarity rules do not transfer mechanically. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Use QAR theory. 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. Breaks

A small example makes the mechanism visible. Check parameter change over time.

A concrete situation

A structural break can mimic tail dynamics.

What the mathematics does not license

One full-sample fit averages regimes. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Use windows. 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. Uncertainty

This is where modelling choices begin to affect evidence. Use dependence-aware covariance or bootstrap.

A concrete situation

Resample blocks or innovations as justified.

What the mathematics does not license

IID bootstrap is mismatched. Keeping this boundary clear makes the result easier to audit and transfer.

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.

---

15. Forecasting

The central mathematical idea is concrete. Evaluate rolling check loss and calibration.

A concrete situation

Never train on future data.

What the mathematics does not license

Random folds leak time. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Respect origins. 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 tau, lags, ties and validation.

A concrete situation

Teaching minimum loss=2.5.

What the mathematics does not license

Do not claim causality. 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

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

Questions for a student, parent or mentor

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

A method-specific audit before deployment

Quantile autoregression lets lag effects differ across the conditional distribution of a time series. Instead of assuming one autoregressive coefficient describes every outcome, the model writes a tau-specific conditional quantile such as Q_tau(Y_t|Y_{t-1})=alpha_tau+phi_tau Y_{t-1}. A predictor may have a mild effect near the median and a stronger effect in an upper tail, revealing asymmetric persistence or risk.

Take lag values 1,2,4,3 and next values 2,4,3,6 at tau=.5. Candidate alpha=0,phi=2 gives residuals 0,0,-5,0 and median check-loss total 2.5. Candidate alpha=1,phi=1 gives residuals 0,1,-2,2 and the same total 2.5. The tiny teaching sample therefore has tied optima, showing why piecewise-linear quantile objectives can produce non-unique autoregressive coefficients.

Time order cannot be shuffled. Serial dependence affects inference, and a fitted quantile autoregression needs dynamic checks: lag adequacy, residual hit sequences, stability, structural breaks and out-of-sample forecasts. Coefficients across tau can cross or imply implausible dynamics. Unit-root-like behaviour in a tail requires careful theory rather than ordinary least-squares interpretations.

Report lag order, tau grid, deterministic terms, stationarity assumptions, solver ties, covariance or bootstrap method, forecast origin, rolling evaluation and quantile calibration. A conditional quantile path describes an association with the observed past under a model; it does not say the lagged value causes the future or guarantee a particular extreme outcome.

---

Frequently asked questions

What is quantile autoregression?

A time-series model whose lag coefficients describe a chosen conditional quantile.

Why fit several tau values?

Persistence can differ between the centre and tails.

Why are random folds unsuitable?

They leak future information into training.

What is a hit sequence?

Indicators showing whether observations fall below fitted conditional quantiles.

Can coefficients be non-unique?

Yes. Piecewise-linear loss can create tied optima.

What must be reported?

Tau, lags, deterministic terms, ties, stability, uncertainty and rolling calibration.

---

Primary source and fact-check note

Roger Koenker and Zhijie Xiao, “Quantile Autoregression,” Journal of the American Statistical Association 101(475), 980–990 (2006), DOI 10.1198/016214506000000672. 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

• Quantile regression forests
• Quantile crossing
• Composite quantile regression
• How Quantiles Work

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.

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.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Time order

Begin with this mechanism: Keep observations chronological. Recreate the concrete case: Lags 1,2,4,3 map to next 2,4,3,6. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Shuffling destroys dynamics. Complete the transfer step: Preserve index. 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: Target

Begin with this mechanism: Choose tau=.5. Recreate the concrete case: Model conditional median. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Other tails can behave differently. Complete the transfer step: Fit grid. 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: Lag model

Begin with this mechanism: Write alpha+phi Y_previous. Recreate the concrete case: Use one lag. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Omitted lags can leave dependence. Complete the transfer step: Check order. 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: Candidate A

Begin with this mechanism: Set alpha=0,phi=2. Recreate the concrete case: Predictions 2,4,8,6. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Large phi may be unstable dynamically. Complete the transfer step: Inspect theory. 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 5: Residuals A

Begin with this mechanism: Subtract predictions. Recreate the concrete case: Get 0,0,-5,0. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: One large negative residual remains visible. Complete the transfer step: Keep sign. 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: Loss A

Begin with this mechanism: Sum median check loss. Recreate the concrete case: Total=2.5. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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 loss is not forecast skill. Complete the transfer step: Use rolling 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 7: Candidate B

Begin with this mechanism: Set alpha=1,phi=1. Recreate the concrete case: Predictions 2,3,5,4. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Different dynamics can tie. Complete the transfer step: Compare path. 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: Residuals B

Begin with this mechanism: Get 0,1,-2,2. Recreate the concrete case: Spread differs from candidate A. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Same loss hides residual structure. Complete the transfer step: Plot hits. 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 9: Loss B

Begin with this mechanism: Sum median check loss. Recreate the concrete case: Total=2.5. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Tie requires reporting. Complete the transfer step: Do not pick silently. 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: Non-uniqueness

Begin with this mechanism: Recognise flat objective regions. Recreate the concrete case: Both candidates are optima among the comparison and LP confirms minimum 2.5. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Software returns one vertex. Complete the transfer step: Search range. 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: Hit sequence

Begin with this mechanism: Record I(Y_t below fitted quantile). Recreate the concrete case: Median hits should balance conditionally. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Overall 50 percent is insufficient. Complete the transfer step: Test serial pattern. 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: Stability

Begin with this mechanism: Assess alpha and phi dynamics by tau. Recreate the concrete case: Large tail phi may imply persistence. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: OLS stationarity rules do not transfer mechanically. Complete the transfer step: Use QAR theory. 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 13: Breaks

Begin with this mechanism: Check parameter change over time. Recreate the concrete case: A structural break can mimic tail dynamics. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: One full-sample fit averages regimes. Complete the transfer step: Use windows. 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: Uncertainty

Begin with this mechanism: Use dependence-aware covariance or bootstrap. Recreate the concrete case: Resample blocks or innovations as justified. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: IID bootstrap is mismatched. Complete the transfer step: Mirror design. 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: Forecasting

Begin with this mechanism: Evaluate rolling check loss and calibration. Recreate the concrete case: Never train on future data. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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: Random folds leak time. Complete the transfer step: Respect origins. 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 tau, lags, ties and validation. Recreate the concrete case: Teaching minimum loss=2.5. Write time indexes, lagged values, responses, quantile levels, coefficients, predictions, residuals, check losses, hit sequences, forecast origins 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 claim causality. 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.