Chapter map

1. 1. Observation stream
2. 2. Initialization
3. 3. Weight λ
4. 4. Update equation
5. 5. Weight sequence
6. 6. Effective memory
7. 7. Large λ
8. 8. Small λ
9. 9. Target mean
10. 10. Variance model
11. 11. Long-run variance
12. 12. Finite-time limits
13. 13. Control multiplier
14. 14. Simulation calibration
15. 15. Raw-series check
16. 16. Investigation

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

---

1. Observation stream

This is where modelling choices begin to affect evidence. Define comparable measurements over time.

A concrete situation

Daily count Yt is recorded.

What the mathematics does not license

Coverage changes create false shifts. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Track exposure. 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. Initialization

The central mathematical idea is concrete. Set Z0 before monitoring.

A concrete situation

Z0=4, the stable target.

What the mathematics does not license

A poor start distorts early values. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report the choice. 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. Weight λ

Here the geometry becomes operational. Choose a number between zero and one.

A concrete situation

λ=.2.

What the mathematics does not license

Tuning after seeing alarms overfits. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Freeze it. 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. Update equation

This step deserves a careful pause. Compute λYt+(1−λ)Zt−1.

A concrete situation

.2(10)+.8(4)=5.2.

What the mathematics does not license

Rounding accumulates. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Keep adequate precision. 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. Weight sequence

The useful question is not only how to calculate, but what the calculation assumes. Expand the recursion into geometric weights.

A concrete situation

Recent weights are .2,.16,.128 and so on.

What the mathematics does not license

Initialization retains residual weight. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show the expansion. 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. Effective memory

A small example makes the mechanism visible. Interpret how quickly weights decay.

A concrete situation

.8^k declines with lag k.

What the mathematics does not license

Memory is not a fixed window. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Avoid false equivalence. 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. Large λ

This is where modelling choices begin to affect evidence. Emphasize new observations.

A concrete situation

λ=.8 reacts sharply.

What the mathematics does not license

It also follows noise. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Compare operating curves. 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. Small λ

The central mathematical idea is concrete. Retain longer history.

A concrete situation

λ=.1 moves gradually.

What the mathematics does not license

Abrupt shifts may be delayed. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Match the target shift. 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. Target mean

Here the geometry becomes operational. Define the stable centre.

A concrete situation

The expected level is 4.

What the mathematics does not license

Drift makes a static target obsolete. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Review by protocol. 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. Variance model

This step deserves a careful pause. Specify σ² or count variance.

A concrete situation

Poisson variance may equal the mean.

What the mathematics does not license

Real counts can overdisperse. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Test the assumption. 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. Long-run variance

The useful question is not only how to calculate, but what the calculation assumes. Use λ/(2−λ) times σ² when assumptions fit.

A concrete situation

With λ=.2, factor=.2/1.8≈.1111.

What the mathematics does not license

This omits start-up adjustment. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

State the horizon. 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. Finite-time limits

A small example makes the mechanism visible. Include the geometric start-up factor.

A concrete situation

Early limits differ from steady state.

What the mathematics does not license

Software defaults may hide it. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Verify equations. 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. Control multiplier

This is where modelling choices begin to affect evidence. Choose the limit width.

A concrete situation

A wider limit reduces alerts.

What the mathematics does not license

It increases detection delay. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Calibrate jointly. 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. Simulation calibration

The central mathematical idea is concrete. Replay realistic quiet and shifted streams.

A concrete situation

Alarm rates and delays are estimated.

What the mathematics does not license

Idealized independence can mislead. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Model calendars. 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. Raw-series check

Here the geometry becomes operational. Plot Yt beside Zt.

A concrete situation

A spike and sustained rise look different.

What the mathematics does not license

Smoothing can hide detail. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Keep both. 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. Investigation

This step deserves a careful pause. Examine causes after a signal.

A concrete situation

Reporting, exposure and system changes are checked.

What the mathematics does not license

The chart is not causal evidence. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record findings. 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

An exponentially weighted moving average gives the newest observation the greatest weight while retaining geometrically decreasing memory of all earlier observations. The recursion Zt=λYt+(1−λ)Zt−1 is simple enough to calculate by hand, yet it creates a tunable detector for small sustained shifts that may never produce one dramatic point.

If λ=.2, the previous EWMA is 4 and today’s count is 10, then Zt=.2(10)+.8(4)=5.2. The observation rises by six, but the smoothed statistic rises by 1.2 because eighty per cent of the previous state is retained. A smaller λ remembers longer; λ=1 discards that memory and returns the current observation.

Control limits require a process model. Under independent observations with stable variance σ², the long-run EWMA variance is approximately [λ/(2−λ)]σ²; finite-time limits include the start-up factor. Count surveillance may need Poisson or negative-binomial variance, time-varying exposure and simulation-based limits instead of importing normal-theory limits without checking.

EWMA is a monitoring device rather than an explanation. A signal indicates that the weighted history is unusual under the declared baseline. Good practice preserves the raw series, states initialization and λ, calibrates false-alarm and delay trade-offs, checks dependence and seasonality, and investigates reporting or system changes alongside substantive causes.

---

Frequently asked questions

What is an EWMA?

It is a recursively weighted average in which weights on older observations decrease geometrically.

What does λ control?

It controls responsiveness and memory: large λ emphasizes the newest point, while small λ retains history longer.

How is EWMA initialized?

A declared target, historical mean or another documented starting value is used.

Why can EWMA detect small shifts?

Repeated moderate departures accumulate in the smoothed statistic even when no single point is extreme.

Are normal-theory limits always valid for counts?

No. Discrete, overdispersed or dependent counts may require a different variance model or simulation.

Does an EWMA signal explain the change?

No. It marks unusual weighted history under the baseline and must be investigated.

---

Primary source and fact-check note

S. W. Roberts, “Control Chart Tests Based on Geometric Moving Averages,” Technometrics 1(3), 239–250 (1959), DOI 10.1080/00401706.1959.10489860. 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

• Poisson CUSUM surveillance
• Wombling boundary detection
• 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

Stream audit

Verify measurements and exposure.

Start audit

Publish Z0 and start-up treatment.

Weight audit

Explain λ and geometric decay.

Variance audit

Match limits to the data model.

Calibration audit

Measure false alarms and delay.

Plot audit

Preserve raw points beside the EWMA.

Transfer audit

Test frozen settings on new periods.

Final claim

Describe unusual weighted history, not its cause.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Observation stream

Begin with this mechanism: Define comparable measurements over time. Recreate the concrete case: Daily count Yt is recorded. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Coverage changes create false shifts. Complete the transfer step: Track exposure. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 2: Initialization

Begin with this mechanism: Set Z0 before monitoring. Recreate the concrete case: Z0=4, the stable target. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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 poor start distorts early values. Complete the transfer step: Report the choice. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 3: Weight λ

Begin with this mechanism: Choose a number between zero and one. Recreate the concrete case: λ=.2. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Tuning after seeing alarms overfits. Complete the transfer step: Freeze it. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 4: Update equation

Begin with this mechanism: Compute λYt+(1−λ)Zt−1. Recreate the concrete case: .2(10)+.8(4)=5.2. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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 accumulates. Complete the transfer step: Keep adequate precision. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 5: Weight sequence

Begin with this mechanism: Expand the recursion into geometric weights. Recreate the concrete case: Recent weights are .2,.16,.128 and so on. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Initialization retains residual weight. Complete the transfer step: Show the expansion. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 6: Effective memory

Begin with this mechanism: Interpret how quickly weights decay. Recreate the concrete case: .8^k declines with lag k. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Memory is not a fixed window. Complete the transfer step: Avoid false equivalence. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 7: Large λ

Begin with this mechanism: Emphasize new observations. Recreate the concrete case: λ=.8 reacts sharply. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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 also follows noise. Complete the transfer step: Compare operating curves. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 8: Small λ

Begin with this mechanism: Retain longer history. Recreate the concrete case: λ=.1 moves gradually. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Abrupt shifts may be delayed. Complete the transfer step: Match the target shift. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 9: Target mean

Begin with this mechanism: Define the stable centre. Recreate the concrete case: The expected level is 4. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Drift makes a static target obsolete. Complete the transfer step: Review by protocol. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 10: Variance model

Begin with this mechanism: Specify σ² or count variance. Recreate the concrete case: Poisson variance may equal the mean. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Real counts can overdisperse. Complete the transfer step: Test the assumption. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 11: Long-run variance

Begin with this mechanism: Use λ/(2−λ) times σ² when assumptions fit. Recreate the concrete case: With λ=.2, factor=.2/1.8≈.1111. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: This omits start-up adjustment. Complete the transfer step: State the horizon. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 12: Finite-time limits

Begin with this mechanism: Include the geometric start-up factor. Recreate the concrete case: Early limits differ from steady state. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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 defaults may hide it. Complete the transfer step: Verify equations. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 13: Control multiplier

Begin with this mechanism: Choose the limit width. Recreate the concrete case: A wider limit reduces alerts. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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 increases detection delay. Complete the transfer step: Calibrate jointly. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 14: Simulation calibration

Begin with this mechanism: Replay realistic quiet and shifted streams. Recreate the concrete case: Alarm rates and delays are estimated. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Idealized independence can mislead. Complete the transfer step: Model calendars. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 15: Raw-series check

Begin with this mechanism: Plot Yt beside Zt. Recreate the concrete case: A spike and sustained rise look different. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: Smoothing can hide detail. Complete the transfer step: Keep both. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 16: Investigation

Begin with this mechanism: Examine causes after a signal. Recreate the concrete case: Reporting, exposure and system changes are checked. Write observations, target levels, starting values, lambda weights, lag weights, EWMA values, variance estimates, start-up factors, control limits, alarm flags, run lengths and review outcomes 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: The chart is not causal evidence. Complete the transfer step: Record findings. Change exactly one observed count, reporting date, historical window, baseline mean, standard deviation, guard-band length, dispersion factor, trend term, seasonal factor, robust weight, predictive quantile, baseline rate, target rate, likelihood score, reset value, decision limit, EWMA weight, starting value, variance model, calendar effect, exposure denominator, random seed, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, baseline design, model, computation, threshold, uncertainty, interpretation and independent investigation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.