Chapter map

1. 1. Series order
2. 2. Rank count
3. 3. Cumulative t
4. 4. Null mean
5. 5. Null variance
6. 6. UF standardization
7. 7. Significance band
8. 8. Reverse order
9. 9. UB alignment
10. 10. Curve crossing
11. 11. Endpoint evidence
12. 12. Tie correction
13. 13. Autocorrelation
14. 14. Seasonality
15. 15. Sensitivity
16. 16. Attribution

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

---

1. Series order

Here the geometry becomes operational. Freeze observations and dates.

A concrete situation

Values are 1,3,2,5,4.

What the mathematics does not license

Sorting destroys the process. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Lock timestamps. 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. Rank count

This step deserves a careful pause. For each xi count earlier xj below it.

A concrete situation

Counts are 0,1,1,3,3.

What the mathematics does not license

Ties need a rule. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Show comparisons. 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. Cumulative t

The useful question is not only how to calculate, but what the calculation assumes. Add counts progressively.

A concrete situation

t is 0,1,2,5,8.

What the mathematics does not license

One error persists forward. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Recalculate independently. 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. Null mean

A small example makes the mechanism visible. Compute i(i-1)/4.

A concrete situation

At i=5 mean=5.

What the mathematics does not license

Indexing conventions differ. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

State i. 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. Null variance

This is where modelling choices begin to affect evidence. Compute i(i-1)(2i+5)/72.

A concrete situation

At i=5 variance=4.167.

What the mathematics does not license

Dependence changes it. 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.

---

6. UF standardization

The central mathematical idea is concrete. Subtract mean and divide by standard deviation.

A concrete situation

UF5≈1.470.

What the mathematics does not license

Small i approximations are coarse. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Plot from stable i. 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. Significance band

Here the geometry becomes operational. Choose a predeclared normal band.

A concrete situation

±1.96 marks nominal five percent.

What the mathematics does not license

Repeated points are not independent tests. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Avoid pointwise overclaim. 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. Reverse order

This step deserves a careful pause. Run the same counts from the end.

A concrete situation

Reverse values are 4,5,2,3,1.

What the mathematics does not license

Sign alignment is easy to mishandle. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Archive code. 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. UB alignment

The useful question is not only how to calculate, but what the calculation assumes. Reverse and orient the retrograde statistics.

A concrete situation

UB is plotted on original dates.

What the mathematics does not license

Software conventions vary. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Verify with a toy series. 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. Curve crossing

A small example makes the mechanism visible. Locate sign changes in UF minus UB.

A concrete situation

A crossing lies between dates.

What the mathematics does not license

Interpolation invents precision. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report an interval. 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. Endpoint evidence

This is where modelling choices begin to affect evidence. Read the final UF with its band.

A concrete situation

1.470 stays inside ±1.96.

What the mathematics does not license

No crossing threshold is proof of no change. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report uncertainty. 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. Tie correction

The central mathematical idea is concrete. Adjust rank counting where equal values occur.

A concrete situation

Rounded values tie.

What the mathematics does not license

Ignoring ties inflates structure. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Preserve 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.

---

13. Autocorrelation

Here the geometry becomes operational. Assess persistence before using ordinary bands.

A concrete situation

Positive lag-one correlation remains.

What the mathematics does not license

Naive bands are optimistic. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Simulate dependence. 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. Seasonality

This step deserves a careful pause. Use comparable seasonal groups.

A concrete situation

Monthly cycles alter rank counts.

What the mathematics does not license

Pooling seasons confuses trend. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

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

---

15. Sensitivity

The useful question is not only how to calculate, but what the calculation assumes. Repeat justified windows and conventions.

A concrete situation

Candidate timing moves one year.

What the mathematics does not license

Outcome-selected windows bias results. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Publish all planned variants. 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. Attribution

A small example makes the mechanism visible. Compare candidate timing with records.

A concrete situation

A sensor change occurs nearby.

What the mathematics does not license

Coincidence is not cause. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Validate mechanisms. 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 sequential Mann–Kendall procedure watches a rank-based trend statistic evolve through time instead of reporting only one end-of-series score. A progressive curve is calculated from the beginning, and a retrograde curve is calculated from the reversed series. Their paths help locate intervals where a persistent change may have begun, while significance bands show when the cumulative evidence becomes unusually large.

For each observation xi, count how many earlier observations are smaller and accumulate those counts into ti. Under the no-trend null, E(ti)=i(i-1)/4 and Var(ti)=i(i-1)(2i+5)/72. The standardized progressive statistic is [ti-E(ti)] divided by the square root of that variance. The reverse calculation uses the same logic in the opposite direction.

For 1,3,2,5,4, the new-lower counts are 0,1,1,3,3 and cumulative t values are 0,1,2,5,8. At i=5, E(t5)=5 and Var(t5)=300/72≈4.167, so UF5=(8-5)/sqrt(4.167)≈1.470. That endpoint is below 1.96, so the small example does not show five-percent normal-approximation evidence.

A crossing of progressive and retrograde curves is an approximate timing clue, not a proven change date. Serial dependence, seasonality, ties, short samples and repeated inspection affect calibration. Plot the original series and validate candidate intervals with independent records.

---

Frequently asked questions

What does sequential Mann–Kendall add?

It shows how standardized rank evidence develops through time in forward and reverse directions.

What is UF?

UF is the standardized progressive cumulative count of earlier observations lower than each new observation.

What is UB?

UB is the corresponding retrograde curve derived from the reversed series and aligned to forward time.

What does a curve crossing mean?

It is an approximate candidate interval for a change, subject to calibration and context.

Why use significance bands?

They show when cumulative evidence is unusually far from the no-trend expectation under the declared approximation.

Does a crossing prove a change point?

No. Dependence, seasonality and repeated inspection can create crossings; independent evidence is needed.

---

Primary source and fact-check note

R. Sneyers, On the Statistical Analysis of Series of Observations, WMO Technical Note No. 143, WMO-No. 415 (1990). 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
• Hamed–Rao modified Mann–Kendall test
• 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

Count audit

Reproduce every progressive rank count.

Moment audit

Verify null means and variances.

UF audit

Archive standardized forward values.

UB audit

Verify reverse alignment and sign.

Band audit

Separate pointwise bands from repeated testing.

Dependence audit

Simulate realistic persistence.

Timing audit

Report intervals rather than exact dates.

Final claim

Describe approximate timing evidence only.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Series order

Begin with this mechanism: Freeze observations and dates. Recreate the concrete case: Values are 1,3,2,5,4. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Sorting destroys the process. Complete the transfer step: Lock timestamps. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Rank count

Begin with this mechanism: For each xi count earlier xj below it. Recreate the concrete case: Counts are 0,1,1,3,3. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Ties need a rule. Complete the transfer step: Show comparisons. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Cumulative t

Begin with this mechanism: Add counts progressively. Recreate the concrete case: t is 0,1,2,5,8. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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 error persists forward. Complete the transfer step: Recalculate independently. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Null mean

Begin with this mechanism: Compute i(i-1)/4. Recreate the concrete case: At i=5 mean=5. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Indexing conventions differ. Complete the transfer step: State i. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Null variance

Begin with this mechanism: Compute i(i-1)(2i+5)/72. Recreate the concrete case: At i=5 variance=4.167. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Dependence changes it. Complete the transfer step: Check assumptions. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: UF standardization

Begin with this mechanism: Subtract mean and divide by standard deviation. Recreate the concrete case: UF5≈1.470. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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 i approximations are coarse. Complete the transfer step: Plot from stable i. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Significance band

Begin with this mechanism: Choose a predeclared normal band. Recreate the concrete case: ±1.96 marks nominal five percent. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Repeated points are not independent tests. Complete the transfer step: Avoid pointwise overclaim. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Reverse order

Begin with this mechanism: Run the same counts from the end. Recreate the concrete case: Reverse values are 4,5,2,3,1. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Sign alignment is easy to mishandle. Complete the transfer step: Archive code. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: UB alignment

Begin with this mechanism: Reverse and orient the retrograde statistics. Recreate the concrete case: UB is plotted on original dates. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Software conventions vary. Complete the transfer step: Verify with a toy series. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Curve crossing

Begin with this mechanism: Locate sign changes in UF minus UB. Recreate the concrete case: A crossing lies between dates. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Interpolation invents precision. Complete the transfer step: Report an interval. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Endpoint evidence

Begin with this mechanism: Read the final UF with its band. Recreate the concrete case: 1.470 stays inside ±1.96. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: No crossing threshold is proof of no change. Complete the transfer step: Report uncertainty. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Tie correction

Begin with this mechanism: Adjust rank counting where equal values occur. Recreate the concrete case: Rounded values tie. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Ignoring ties inflates structure. Complete the transfer step: Preserve resolution. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Autocorrelation

Begin with this mechanism: Assess persistence before using ordinary bands. Recreate the concrete case: Positive lag-one correlation remains. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Naive bands are optimistic. Complete the transfer step: Simulate dependence. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Seasonality

Begin with this mechanism: Use comparable seasonal groups. Recreate the concrete case: Monthly cycles alter rank counts. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Pooling seasons confuses trend. Complete the transfer step: Apply a seasonal design. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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

Begin with this mechanism: Repeat justified windows and conventions. Recreate the concrete case: Candidate timing moves one year. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Outcome-selected windows bias results. Complete the transfer step: Publish all planned variants. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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: Attribution

Begin with this mechanism: Compare candidate timing with records. Recreate the concrete case: A sensor change occurs nearby. Write observations, dates, progressive counts, cumulative t values, null means, null variances, UF values, reverse counts, UB values, crossings, significance bands and candidate intervals 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: Coincidence is not cause. Complete the transfer step: Validate mechanisms. Change exactly one observation, date, split point, site, tie rule, lag, covariance, slope convention, significance band, null distribution, simulation seed, region boundary or validation period. 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.