WHY MATHEMATICS?

Why Mathematics? | Weighted Interval Score: Nested Prediction Bands, Median Error, Quantile Losses and CRPS Approximation

How several prediction intervals and a median become one transparent forecast score, why the quantile grid matters, and how students can audit every component.

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

Assemble a median and nested central prediction intervals
Calculate each interval score and its declared weight
Decompose the total into sharpness and directional penalties
Compare models on matched targets, horizons and quantile grids

Chapter map

1. Median
2. Outcome
3. Inner interval
4. Inner miss
5. Inner score
6. Outer interval
7. Outer coverage
8. Weights
9. Weighted sum
10. Normalisation
11. Pinball identity
12. CRPS link
13. Decomposition
14. Aggregation
15. Missing cases
16. Claim

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

1. Median

A small example makes the mechanism visible. Record the predictive median.

A concrete situation

Use m=100.

What the mathematics does not license

The median is not the mean by default. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Name functional. 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. Outcome

This is where modelling choices begin to affect evidence. Observe y=120.

A concrete situation

Median absolute error is 20.

What the mathematics does not license

One outcome cannot establish calibration. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep case label. 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. Inner interval

The central mathematical idea is concrete. Use 50% interval [90,110].

A concrete situation

Alpha is .50 and width is 20.

What the mathematics does not license

Intervals must be nested for the intended summary. 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. Inner miss

Here the geometry becomes operational. Outcome exceeds 110 by 10.

A concrete situation

Penalty factor is 4.

What the mathematics does not license

Use 2/alpha, not 1/alpha. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Show formula. 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. Inner score

This step deserves a careful pause. Add width and miss penalty.

A concrete situation

20+4×10=60.

What the mathematics does not license

This is one component. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep component. 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. Outer interval

The useful question is not only how to calculate, but what the calculation assumes. Use 90% interval [70,130].

A concrete situation

Alpha is .10 and width is 60.

What the mathematics does not license

Wide intervals are not free. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Count width. 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. Outer coverage

A small example makes the mechanism visible. Place 120 inside [70,130].

A concrete situation

Outer score is 60.

What the mathematics does not license

Inside does not mean score zero. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Explain width. 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. Weights

This is where modelling choices begin to affect evidence. Use .5,.25,.05.

A concrete situation

They match median and alpha/2 choices.

What the mathematics does not license

Different grids create different summaries. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Freeze 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.

9. Weighted sum

The central mathematical idea is concrete. Calculate numerator.

A concrete situation

10+15+3=28.

What the mathematics does not license

Do not drop median contribution. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

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

10. Normalisation

Here the geometry becomes operational. Divide by K+.5=2.5.

A concrete situation

WIS is 11.2.

What the mathematics does not license

A different scaling changes reported numbers. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

State divisor. 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. Pinball identity

This step deserves a careful pause. Map interval endpoints to quantile losses.

A concrete situation

Use paired lower and upper quantiles.

What the mathematics does not license

Crossing quantiles break interpretation. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Check monotonicity. 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. CRPS link

The useful question is not only how to calculate, but what the calculation assumes. Use a denser equally spaced grid.

A concrete situation

WIS approaches CRPS under stated weights.

What the mathematics does not license

Approximation is not equality for every sparse grid. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Use precise 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.

13. Decomposition

A small example makes the mechanism visible. Separate widths and directional penalties.

A concrete situation

Plot sharpness, underprediction and overprediction.

What the mathematics does not license

One total can hide mechanism. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Open score. 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. Aggregation

This is where modelling choices begin to affect evidence. Average across a fixed case set.

A concrete situation

Stratify by target and horizon.

What the mathematics does not license

Scale differences can dominate. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Normalise only with rationale. 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. Missing cases

The central mathematical idea is concrete. Predefine submission and observation rules.

A concrete situation

Keep denominators visible.

What the mathematics does not license

Selective omission can improve a ranking falsely. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Audit coverage. 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. Report grid, weights and components.

A concrete situation

Teaching WIS is 11.2.

What the mathematics does not license

Do not compare unmatched WIS definitions. 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

The weighted interval score combines a predictive median with several nested central prediction intervals. Bracher, Ray, Gneiting and Reich explain that it is a proper score for interval or quantile forecasts and, under a common weighting scheme, approximates the continuous ranked probability score as the interval grid becomes dense. It can be decomposed into sharpness, underprediction and overprediction components, so the final number can be opened rather than treated as a mysterious leaderboard value.

Use a teaching forecast with median 100, a 50% interval [90,110], a 90% interval [70,130] and outcome 120. The median absolute error is 20. The 50% interval has alpha=.50, width 20 and upper miss 10, so its interval score is 20+4×10=60. The 90% interval has alpha=.10 and contains 120, so its score is 60. With w0=.5, w1=.25 and w2=.05, and divisor K+.5=2.5 for K=2 intervals, WIS=(.5×20+.25×60+.05×60)/2.5=11.2.

The reported WIS depends on which quantiles or intervals are supplied and how they are weighted. Scores computed from different grids are not automatically comparable. Averaging across targets with very different scales can let high-volume series dominate, while ad hoc truncation or target-dependent exclusions can change the ranking. WIS is sensitive to distance and is not the same decision criterion as the logarithmic score. A model can perform well under one score and less well under another because the scores value different forecast features.

A reproducible forecast comparison freezes the median, interval levels, weights, normalisation, targets, horizons and missing-value policy before looking at outcomes. It reports the overall mean alongside width and directional penalty components, checks empirical coverage at every nominal level and adds scale-aware summaries when targets differ. For students, WIS is a practical bridge between percentiles, absolute error, weighted averages and honest uncertainty communication.

Frequently asked questions

What data does WIS need?

A predictive median and one or more central prediction intervals, or the equivalent quantiles.

Why include several interval levels?

They describe more of the predictive distribution than one range.

Why is WIS proper?

With nonnegative fixed weights, truthful quantile and interval reporting minimises expected loss under the stated framework.

How is WIS related to pinball loss?

Each interval score can be written using its lower and upper quantile losses.

Is WIS always identical to CRPS?

No. A dense, suitably weighted interval grid approximates CRPS; sparse grids retain discretisation differences.

What must be reported?

Median, interval levels, weights, normalisation, components, targets, horizons, missing cases and uncertainty.

Primary source and fact-check note

Johannes Bracher, Evan L. Ray, Tilmann Gneiting and Nicholas G. Reich, “Evaluating Epidemic Forecasts in an Interval Format,” PLOS Computational Biology 17(2), e1008618 (2021), DOI 10.1371/journal.pcbi.1008618. 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: score-specific validation and responsible use

Grid audit

Freeze the median, lower and upper quantiles, central interval levels and ordering before evaluation.

Weight audit

Declare every nonnegative weight and divisor; do not compare scores built from unmatched grids as if they were identical.

Component audit

Retain absolute median error, interval widths, lower penalties and upper penalties before aggregation.

Coherence audit

Check nested intervals and noncrossing quantiles; repair only with a declared method and preserve the original diagnosis.

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.

Scale audit

Stratify by target scale and horizon; use normalisation only when its meaning and reference quantity are defensible.

Missingness audit

Record every forecast and observation omission; selective case availability can change model rankings.

Uncertainty audit

Use dependence-aware comparisons when cases share dates, locations, forecast origins or ensemble members.

Final claim

Report the score definition and decomposition; never call WIS a universal measure independent of grid and decision context.

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: Median

Begin with this mechanism: Record the predictive median. Recreate the concrete case: Use m=100. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The median is not the mean by default. Complete the transfer step: Name functional. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Outcome

Begin with this mechanism: Observe y=120. Recreate the concrete case: Median absolute error is 20. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One outcome cannot establish calibration. Complete the transfer step: Keep case label. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Inner interval

Begin with this mechanism: Use 50% interval [90,110]. Recreate the concrete case: Alpha is .50 and width is 20. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Intervals must be nested for the intended summary. Complete the transfer step: Check order. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Inner miss

Begin with this mechanism: Outcome exceeds 110 by 10. Recreate the concrete case: Penalty factor is 4. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Use 2/alpha, not 1/alpha. Complete the transfer step: Show formula. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that 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: Inner score

Begin with this mechanism: Add width and miss penalty. Recreate the concrete case: 20+4×10=60. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is one component. Complete the transfer step: Keep component. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: Outer interval

Begin with this mechanism: Use 90% interval [70,130]. Recreate the concrete case: Alpha is .10 and width is 60. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Wide intervals are not free. Complete the transfer step: Count width. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: Outer coverage

Begin with this mechanism: Place 120 inside [70,130]. Recreate the concrete case: Outer score is 60. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Inside does not mean score zero. Complete the transfer step: Explain width. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Weights

Begin with this mechanism: Use .5,.25,.05. Recreate the concrete case: They match median and alpha/2 choices. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Different grids create different summaries. Complete the transfer step: Freeze grid. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that 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: Weighted sum

Begin with this mechanism: Calculate numerator. Recreate the concrete case: 10+15+3=28. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not drop median contribution. Complete the transfer step: Audit terms. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Normalisation

Begin with this mechanism: Divide by K+.5=2.5. Recreate the concrete case: WIS is 11.2. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A different scaling changes reported numbers. Complete the transfer step: State divisor. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Pinball identity

Begin with this mechanism: Map interval endpoints to quantile losses. Recreate the concrete case: Use paired lower and upper quantiles. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Crossing quantiles break interpretation. Complete the transfer step: Check monotonicity. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: CRPS link

Begin with this mechanism: Use a denser equally spaced grid. Recreate the concrete case: WIS approaches CRPS under stated weights. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Approximation is not equality for every sparse grid. Complete the transfer step: Use precise language. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that 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: Decomposition

Begin with this mechanism: Separate widths and directional penalties. Recreate the concrete case: Plot sharpness, underprediction and overprediction. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: One total can hide mechanism. Complete the transfer step: Open score. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Aggregation

Begin with this mechanism: Average across a fixed case set. Recreate the concrete case: Stratify by target and horizon. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Scale differences can dominate. Complete the transfer step: Normalise only with rationale. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Missing cases

Begin with this mechanism: Predefine submission and observation rules. Recreate the concrete case: Keep denominators visible. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Selective omission can improve a ranking falsely. Complete the transfer step: Audit coverage. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: Report grid, weights and components. Recreate the concrete case: Teaching WIS is 11.2. Write outcome units, median, quantile levels, interval endpoints, alpha values, weights, component scores, divisor, WIS, target and horizon with units, signs, identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not compare unmatched WIS definitions. Complete the transfer step: Bound conclusion. Change exactly one interval endpoint, alpha level, outcome, median, quantile grid, interval weight, normalisation, predictive mean, variance, covariance, threshold, weight function, ensemble value or case weight. Explain which mathematical object changes, which score term moves, which ranking survives and which claim must weaken. Keep target definition, calculation, aggregation, uncertainty and interpretation on separate notebook lines. Explain the result once to a student with the full arithmetic and once to a parent with the real decision and original units. Add a counterexample involving lucky coverage, excessive width, quantile crossing, unmatched grids, scale domination, overconfidence, singular covariance, identical moments with different tails, threshold cherry-picking or regional improvement with global deterioration. State the extra evidence that would separate the explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.