WHY MATHEMATICS?

Why Mathematics? | Youden’s J Statistic for Diagnostic Thresholds: Sensitivity, Specificity, ROC Geometry and Prevalence

How one threshold becomes a vertical ROC separation, with a complete J calculation, selection limits and responsible interpretation.

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

Compute J from sensitivity and specificity
Verify the equivalent TPR minus FPR form
Understand empirical threshold selection
Separate equal weighting from real decision consequences

Chapter map

1. Threshold
2. Sensitivity
3. Specificity
4. Add rates
5. Subtract one
6. False-positive rate
7. ROC form
8. Diagonal baseline
9. Perfect point
10. Reversed rule
11. Threshold search
12. Ties
13. Prevalence
14. Consequences
15. Intervals
16. Claim

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

1. Threshold

A small example makes the mechanism visible. Choose one declared score cutoff.

A concrete situation

Use matched outcomes.

What the mathematics does not license

A different cutoff changes both rates. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record value. 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. Sensitivity

This is where modelling choices begin to affect evidence. Use TP=16,FN=4.

A concrete situation

Sensitivity=.80.

What the mathematics does not license

Conditional on actual positives. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Show count. 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. Specificity

The central mathematical idea is concrete. Use TN=72,FP=8.

A concrete situation

Specificity=.90.

What the mathematics does not license

Conditional on actual negatives. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Show count. 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. Add rates

Here the geometry becomes operational. Compute .80+.90.

A concrete situation

Sum=1.70.

What the mathematics does not license

The two denominators differ. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Keep labels. 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. Subtract one

This step deserves a careful pause. Compute 1.70-1.

A concrete situation

J=.70.

What the mathematics does not license

Do not report as seventy per cent accuracy. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Name 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.

6. False-positive rate

The useful question is not only how to calculate, but what the calculation assumes. Compute 1-.90.

A concrete situation

FPR=.10.

What the mathematics does not license

Check class direction. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Retain value. 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. ROC form

A small example makes the mechanism visible. Compute TPR-FPR.

A concrete situation

.80-.10=.70.

What the mathematics does not license

Both formulas must agree. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Audit. 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. Diagonal baseline

This is where modelling choices begin to affect evidence. Set TPR=FPR.

A concrete situation

J=0.

What the mathematics does not license

Zero means no vertical separation at that point. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Avoid universal claim. 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. Perfect point

The central mathematical idea is concrete. Set sensitivity=specificity=1.

A concrete situation

J=1.

What the mathematics does not license

This is a benchmark. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Do not imply attainability. 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. Reversed rule

Here the geometry becomes operational. Allow TPR below FPR.

A concrete situation

J can be negative.

What the mathematics does not license

Reversing labels may improve direction. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Check coding. 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. Threshold search

This step deserves a careful pause. Compute J at every candidate cutoff.

A concrete situation

Select maximum only in development data.

What the mathematics does not license

Same-sample selection is optimistic. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Validate separately. 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. Ties

The useful question is not only how to calculate, but what the calculation assumes. Declare score-tie threshold handling.

A concrete situation

Count consistently.

What the mathematics does not license

Hidden equality rules move the point. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Document. 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. Prevalence

A small example makes the mechanism visible. Note J uses conditional rates.

A concrete situation

Predictive values still change with prevalence.

What the mathematics does not license

Threshold utility is not prevalence-free. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Add context. 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. Consequences

This is where modelling choices begin to affect evidence. Compare false-positive and false-negative costs.

A concrete situation

Use decision analysis when unequal.

What the mathematics does not license

Maximum J encodes symmetry, not every goal. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

State purpose. 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. Intervals

The central mathematical idea is concrete. Resample threshold selection and evaluation correctly.

A concrete situation

Show uncertainty.

What the mathematics does not license

Optimised J has selection bias. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Separate data. 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. Name threshold and validation design.

A concrete situation

Teaching J=.70.

What the mathematics does not license

Do not call the cutoff universally optimal. 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

Youden’s J statistic adds sensitivity and specificity and subtracts one. The same quantity is true-positive rate minus false-positive rate, so it is the vertical separation between an ROC operating point and the diagonal no-discrimination line at the same false-positive rate. J ranges from minus one to one under the usual definitions, with one representing perfect separation.

At one threshold, suppose sensitivity=.80 and specificity=.90. Then J=.80+.90-1=.70. Because false-positive rate is 1-.90=.10, the ROC form gives .80-.10=.70. These two calculations should agree and provide a useful audit of label direction and denominators.

Across several candidate thresholds, an analyst may choose the threshold with the largest empirical J. That rule treats sensitivity and specificity symmetrically on their conditional scales. It does not automatically reflect real prevalence, costs, benefits or capacity constraints, and selecting and evaluating the same threshold on one sample gives optimistic evidence.

J itself is algebraically built from sensitivity and specificity rather than predictive values, but a threshold called “optimal” still depends on the intended decision. A responsible report shows the whole ROC curve, counts, intervals, threshold-selection process and decision analysis. Students learn how a compact index can be mathematically clear yet operationally conditional.

Frequently asked questions

What is Youden’s J?

Sensitivity plus specificity minus one.

What is its ROC form?

True-positive rate minus false-positive rate at the same threshold.

What does J=0 mean?

No vertical separation from the ROC diagonal at that operating point.

Is maximum J universally optimal?

No. It gives symmetric emphasis to conditional class rates, not every consequence model.

Does prevalence never matter?

J uses conditional rates, while predictive values and threshold utility still depend on prevalence and context.

What must be reported?

Threshold, counts, rates, curve, selection data, intervals, prevalence and consequences.

Primary source and fact-check note

William J. Youden, “Index for Rating Diagnostic Tests,” Cancer 3(1), 32–35 (1950), DOI 10.1002/1097-0142(1950)3:1<32::AID-CNCR2820030106>3.0.CO;2-3. 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: threshold evidence and responsible use

Target audit

Freeze class definitions, target condition, score direction, threshold, reference mask and evaluation population before calculation.

Table audit

Write the full two-by-two table or set cardinalities with every row total, column total, intersection and union.

Formula audit

Show each classwise recall, mean, cross-product, likelihood ratio, set operation, adjustment and rounding step.

Threshold audit

Select any score cutoff in development data and evaluate it on properly separated cases.

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.

Case-mix audit

Report prevalence, class support, object size, missing cases, sample weights, clustering and repeated observations.

Edge audit

Declare zero denominators, empty sets, zero cells, continuity corrections, label reversals and software conventions.

Uncertainty audit

Use intervals or resampling that respects threshold selection, paired predictions, clusters and study design.

Final claim

State the exact balanced recall, ROC separation, odds ratio or overlap evidence and stop before universal or guaranteed-performance claims.

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: Threshold

Begin with this mechanism: Choose one declared score cutoff. Recreate the concrete case: Use matched outcomes. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: A different cutoff changes both rates. Complete the transfer step: Record value. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 2: Sensitivity

Begin with this mechanism: Use TP=16,FN=4. Recreate the concrete case: Sensitivity=.80. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Conditional on actual positives. Complete the transfer step: Show count. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 3: Specificity

Begin with this mechanism: Use TN=72,FP=8. Recreate the concrete case: Specificity=.90. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Conditional on actual negatives. Complete the transfer step: Show count. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 4: Add rates

Begin with this mechanism: Compute .80+.90. Recreate the concrete case: Sum=1.70. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: The two denominators differ. Complete the transfer step: Keep labels. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish 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: Subtract one

Begin with this mechanism: Compute 1.70-1. Recreate the concrete case: J=.70. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not report as seventy per cent accuracy. Complete the transfer step: Name index. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 6: False-positive rate

Begin with this mechanism: Compute 1-.90. Recreate the concrete case: FPR=.10. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Check class direction. Complete the transfer step: Retain value. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 7: ROC form

Begin with this mechanism: Compute TPR-FPR. Recreate the concrete case: .80-.10=.70. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Both formulas must agree. Complete the transfer step: Audit. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 8: Diagonal baseline

Begin with this mechanism: Set TPR=FPR. Recreate the concrete case: J=0. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Zero means no vertical separation at that point. Complete the transfer step: Avoid universal claim. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish 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: Perfect point

Begin with this mechanism: Set sensitivity=specificity=1. Recreate the concrete case: J=1. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: This is a benchmark. Complete the transfer step: Do not imply attainability. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 10: Reversed rule

Begin with this mechanism: Allow TPR below FPR. Recreate the concrete case: J can be negative. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Reversing labels may improve direction. Complete the transfer step: Check coding. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 11: Threshold search

Begin with this mechanism: Compute J at every candidate cutoff. Recreate the concrete case: Select maximum only in development data. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Same-sample selection is optimistic. Complete the transfer step: Validate separately. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 12: Ties

Begin with this mechanism: Declare score-tie threshold handling. Recreate the concrete case: Count consistently. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Hidden equality rules move the point. Complete the transfer step: Document. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish 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: Prevalence

Begin with this mechanism: Note J uses conditional rates. Recreate the concrete case: Predictive values still change with prevalence. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Threshold utility is not prevalence-free. Complete the transfer step: Add context. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 14: Consequences

Begin with this mechanism: Compare false-positive and false-negative costs. Recreate the concrete case: Use decision analysis when unequal. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Maximum J encodes symmetry, not every goal. Complete the transfer step: State purpose. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 15: Intervals

Begin with this mechanism: Resample threshold selection and evaluation correctly. Recreate the concrete case: Show uncertainty. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Optimised J has selection bias. Complete the transfer step: Separate data. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.

Workshop 16: Claim

Begin with this mechanism: Name threshold and validation design. Recreate the concrete case: Teaching J=.70. Write threshold, four cells, sensitivity, specificity, TPR, FPR, J, curve points and uncertainty with identifiers and assumptions. Calculate the smallest non-trivial example by hand and reproduce it independently. Challenge this limitation: Do not call the cutoff universally optimal. Complete the transfer step: Bound conclusion. Change exactly one class label, support, threshold, cell, prevalence, adjustment, likelihood-ratio component, zero-cell correction, pixel, mask, union, averaging rule or validation split. Explain which mathematical object changes, which metric moves and which claim must weaken. Keep target definition, arithmetic, sampling uncertainty and interpretation on separate notebook lines. Explain once to a student and once to a parent. Add a counterexample involving a majority-only classifier, maximum J chosen on evaluation data, identical DOR from different sensitivity-specificity trade-offs, high IoU with a displaced boundary or two empty masks. State the extra evidence needed to distinguish explanations. Finish with one supported sentence, one unsupported sentence and one useful next measurement.