Chapter map

1. 1. Ordered data
2. 2. Ordinary score
3. 3. Tie correction
4. 4. Ordinary variance
5. 5. Ranks
6. 6. Trend removal
7. 7. Lag series
8. 8. Autocorrelation
9. 9. Lag screening
10. 10. Combinatorial weight
11. 11. Denominator
12. 12. Correction factor
13. 13. Modified variance
14. 14. Modified Z
15. 15. Sensitivity grid
16. 16. Independent validation

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

---

1. Ordered data

The useful question is not only how to calculate, but what the calculation assumes. Preserve values and timestamps.

A concrete situation

Six observations are ordered.

What the mathematics does not license

Revisions change ranks. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Version 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.

---

2. Ordinary score

A small example makes the mechanism visible. Calculate all later-earlier signs.

A concrete situation

S is fixed before correction.

What the mathematics does not license

A coding error propagates. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Audit pairs. 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. Tie correction

This is where modelling choices begin to affect evidence. Record equal-value rank groups.

A concrete situation

Ties reduce ordinary variance.

What the mathematics does not license

Rounding changes ties. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

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

---

4. Ordinary variance

The central mathematical idea is concrete. Calculate Var(S) under independence.

A concrete situation

Illustrative variance is 20.

What the mathematics does not license

It is only a baseline. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Label the null. 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. Ranks

Here the geometry becomes operational. Replace values with average ranks for ties.

A concrete situation

Ranks 1 through 6 are formed.

What the mathematics does not license

Average ranks must be consistent. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Verify software. 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. Trend removal

This step deserves a careful pause. Use the method’s declared detrending step where required.

A concrete situation

Residual ranks are analysed.

What the mathematics does not license

Leaving trend in can inflate autocorrelation. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Document the sequence. 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. Lag series

The useful question is not only how to calculate, but what the calculation assumes. Pair ranks separated by each lag.

A concrete situation

Lag one has five pairs.

What the mathematics does not license

Large lags have little information. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Cap lags. 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. Autocorrelation

A small example makes the mechanism visible. Estimate rank autocorrelation at each candidate lag.

A concrete situation

rho1=.5 in the illustration.

What the mathematics does not license

Sample correlations are noisy. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Give intervals. 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. Lag screening

This is where modelling choices begin to affect evidence. Apply the declared significance rule.

A concrete situation

Only lag one is retained.

What the mathematics does not license

Screening many lags is unstable. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Run alternatives. 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. Combinatorial weight

The central mathematical idea is concrete. Compute (n-i)(n-i-1)(n-i-2).

A concrete situation

At n=6,i=1 the weight is 60.

What the mathematics does not license

Index errors change correction. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show arithmetic. 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. Denominator

Here the geometry becomes operational. Compute n(n-1)(n-2).

A concrete situation

The denominator is 120.

What the mathematics does not license

Small n makes asymptotics weak. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Use simulation. 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. Correction factor

This step deserves a careful pause. Add the weighted autocorrelation term to one.

A concrete situation

Factor=1.5.

What the mathematics does not license

A factor is not a p-value. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report components. 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. Modified variance

The useful question is not only how to calculate, but what the calculation assumes. Multiply ordinary variance by the factor.

A concrete situation

20 becomes 30.

What the mathematics does not license

Negative estimates need careful handling. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Check assumptions. 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. Modified Z

A small example makes the mechanism visible. Standardize S with corrected variance and continuity rule.

A concrete situation

Evidence weakens when variance rises.

What the mathematics does not license

Choice of correction affects tails. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Compare versions. 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. Sensitivity grid

This is where modelling choices begin to affect evidence. Vary lag cap, screening and detrending.

A concrete situation

The direction persists but p changes.

What the mathematics does not license

One convenient setting is biased. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Publish the 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.

---

16. Independent validation

The central mathematical idea is concrete. Test another period or mechanism.

A concrete situation

A held-out interval shows similar order.

What the mathematics does not license

Replication still does not prove cause. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Investigate drivers. 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

The Hamed-Rao modified Mann-Kendall test retains the familiar rank-based trend score but changes its uncertainty when observations are autocorrelated. The ordinary Mann-Kendall variance assumes independently ordered data. Positive persistence can create too many concordant pairs under that null, making an unmodified test look more certain than it should.

The method estimates autocorrelation from the ranks and converts the ordinary variance into an effective variance using a correction factor. In a common expression, n divided by effective n equals 1 plus a weighted sum of significant rank autocorrelations. Early lags receive more weight because they connect more ordered pairs.

With n=6 and only lag-one rank autocorrelation 0.5 retained for illustration, the lag-one weight is (6-1)(6-2)(6-3)=60. The denominator n(n-1)(n-2) is 120. The correction is 1+2(60 x 0.5)/120=1.5, so an ordinary variance of 20 becomes 30. The score has not changed; its standardized evidence becomes more cautious.

The correction depends on how trend, ranks, lags and significant autocorrelations are estimated. Screening many noisy lags can itself be unstable, and nonstationary seasonality or long memory may need a different design. The method adjusts inference; it does not identify cause or guarantee that every dependence structure is handled.

---

Frequently asked questions

What does the Hamed–Rao modification change?

It changes the variance used to standardize the Mann–Kendall score when serial correlation is present.

Does the pairwise score change?

No. The rank-based S statistic remains the same; its estimated uncertainty changes.

Why use rank autocorrelation?

The trend statistic is rank based, so dependence in ranks is relevant to its null variance.

What is effective sample size?

It is an information-equivalent size implied by the weighted dependence correction, not a count of deleted observations.

Can negative autocorrelation reduce variance?

Under the formula it can, although the dependence model and retained lags need careful justification.

Does variance correction establish causality?

No. It calibrates trend evidence under a dependence model and says nothing by itself about mechanism.

---

Primary source and fact-check note

Khaled H. Hamed and A. Ramachandra Rao, “A Modified Mann-Kendall Trend Test for Autocorrelated Data,” Journal of Hydrology 204(1–4), 182–196 (1998), DOI 10.1016/S0022-1694(97)00125-X. 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

• Mann–Kendall trend test
• Yue–Pilon trend-free prewhitening
• How Mathematics Works

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

Score audit

Reproduce ordinary S and tie correction.

Rank audit

Verify ranks and detrending order.

Lag audit

Publish candidate and retained lags.

Weight audit

Show combinatorial weights.

Factor audit

Reproduce effective-sample correction.

Variance audit

Compare ordinary and modified uncertainty.

Sensitivity audit

Vary lag rules and dependence models.

Final claim

Describe corrected trend evidence only.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Ordered data

Begin with this mechanism: Preserve values and timestamps. Recreate the concrete case: Six observations are ordered. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Revisions change ranks. Complete the transfer step: Version data. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 2: Ordinary score

Begin with this mechanism: Calculate all later-earlier signs. Recreate the concrete case: S is fixed before correction. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: A coding error propagates. Complete the transfer step: Audit pairs. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 3: Tie correction

Begin with this mechanism: Record equal-value rank groups. Recreate the concrete case: Ties reduce ordinary variance. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Rounding changes ties. Complete the transfer step: Keep resolution. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 4: Ordinary variance

Begin with this mechanism: Calculate Var(S) under independence. Recreate the concrete case: Illustrative variance is 20. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: It is only a baseline. Complete the transfer step: Label the null. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 5: Ranks

Begin with this mechanism: Replace values with average ranks for ties. Recreate the concrete case: Ranks 1 through 6 are formed. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Average ranks must be consistent. Complete the transfer step: Verify software. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 6: Trend removal

Begin with this mechanism: Use the method’s declared detrending step where required. Recreate the concrete case: Residual ranks are analysed. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Leaving trend in can inflate autocorrelation. Complete the transfer step: Document the sequence. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 7: Lag series

Begin with this mechanism: Pair ranks separated by each lag. Recreate the concrete case: Lag one has five pairs. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Large lags have little information. Complete the transfer step: Cap lags. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 8: Autocorrelation

Begin with this mechanism: Estimate rank autocorrelation at each candidate lag. Recreate the concrete case: rho1=.5 in the illustration. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Sample correlations are noisy. Complete the transfer step: Give intervals. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 9: Lag screening

Begin with this mechanism: Apply the declared significance rule. Recreate the concrete case: Only lag one is retained. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Screening many lags is unstable. Complete the transfer step: Run alternatives. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 10: Combinatorial weight

Begin with this mechanism: Compute (n-i)(n-i-1)(n-i-2). Recreate the concrete case: At n=6,i=1 the weight is 60. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Index errors change correction. Complete the transfer step: Show arithmetic. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 11: Denominator

Begin with this mechanism: Compute n(n-1)(n-2). Recreate the concrete case: The denominator is 120. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Small n makes asymptotics weak. Complete the transfer step: Use simulation. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 12: Correction factor

Begin with this mechanism: Add the weighted autocorrelation term to one. Recreate the concrete case: Factor=1.5. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: A factor is not a p-value. Complete the transfer step: Report components. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 13: Modified variance

Begin with this mechanism: Multiply ordinary variance by the factor. Recreate the concrete case: 20 becomes 30. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Negative estimates need careful handling. Complete the transfer step: Check assumptions. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 14: Modified Z

Begin with this mechanism: Standardize S with corrected variance and continuity rule. Recreate the concrete case: Evidence weakens when variance rises. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Choice of correction affects tails. Complete the transfer step: Compare versions. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 15: Sensitivity grid

Begin with this mechanism: Vary lag cap, screening and detrending. Recreate the concrete case: The direction persists but p changes. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: One convenient setting is biased. Complete the transfer step: Publish the grid. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 16: Independent validation

Begin with this mechanism: Test another period or mechanism. Recreate the concrete case: A held-out interval shows similar order. Write observations, ranks, pair signs, S statistics, tie groups, ordinary variances, lags, rank autocorrelations, lag weights, correction factors, effective sample sizes, modified variances and standardized statistics with units, denominators and assumptions. Calculate the smallest non-trivial example by hand and reproduce it in a spreadsheet or short program. Challenge this limitation: Replication still does not prove cause. Complete the transfer step: Investigate drivers. Change exactly one observation, timestamp, season label, missing-value rule, tie rule, slope estimate, autocorrelation estimate, retained lag, rank, pair sign, variance formula, covariance term, filter convention, null distribution, simulation seed, reference period or validation window. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, statistic, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.