Chapter map

1. 1. Measurement definition
2. 2. Reference mean
3. 3. Drift allowance
4. 4. First increment
5. 5. Cumulative value
6. 6. Running minimum
7. 7. PH statistic
8. 8. Second observation
9. 9. Third observation
10. 10. Alarm threshold
11. 11. Downward chart
12. 12. Unknown mean
13. 13. Variance change
14. 14. Calendar effects
15. 15. Operating study
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. Measurement definition

The central mathematical idea is concrete. Fix the quantity and interval.

A concrete situation

Daily score xt is comparable.

What the mathematics does not license

Scale changes create artificial shifts. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Lock units. 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. Reference mean

Here the geometry becomes operational. Estimate or declare μ0.

A concrete situation

Stable mean is 10.

What the mathematics does not license

Contaminated history biases monitoring. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Use quiet 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.

---

3. Drift allowance

This step deserves a careful pause. Choose δ for tolerated movement.

A concrete situation

δ=.5.

What the mathematics does not license

Too large hides small shifts. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Relate it to action. 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. First increment

The useful question is not only how to calculate, but what the calculation assumes. Compute x1−μ0−δ.

A concrete situation

11−10−.5=.5.

What the mathematics does not license

Sign conventions vary. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

State direction. 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. Cumulative value

A small example makes the mechanism visible. Add the increment to mt−1.

A concrete situation

m1=.5.

What the mathematics does not license

Rounding can accumulate. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Keep 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.

---

6. Running minimum

This is where modelling choices begin to affect evidence. Update Mt=min(Mt−1,mt).

A concrete situation

M1 remains zero.

What the mathematics does not license

Initialization matters. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Publish M0. 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. PH statistic

The central mathematical idea is concrete. Compute mt−Mt.

A concrete situation

PH1=.5.

What the mathematics does not license

It is not a p-value. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

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

---

8. Second observation

Here the geometry becomes operational. Continue the recursion.

A concrete situation

x2=12 gives m2=2.0.

What the mathematics does not license

One high point may be noise. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Keep raw data visible. 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. Third observation

This step deserves a careful pause. Measure persistence.

A concrete situation

x3=12 gives PH3=3.5.

What the mathematics does not license

Autocorrelation amplifies runs. Evidence becomes stronger when the limitation is tested instead of hidden.

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.

---

10. Alarm threshold

The useful question is not only how to calculate, but what the calculation assumes. Signal when PH exceeds λ.

A concrete situation

λ is fixed before monitoring.

What the mathematics does not license

Small λ creates alert fatigue. Keeping this boundary clear makes the result easier to audit and transfer.

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.

---

11. Downward chart

A small example makes the mechanism visible. Reverse deviations for decreases.

A concrete situation

A separate lower statistic is maintained.

What the mathematics does not license

Mixing directions confuses thresholds. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Label each chart. 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. Unknown mean

This is where modelling choices begin to affect evidence. Consider online mean estimation only with rules.

A concrete situation

Reference updates gradually.

What the mathematics does not license

Adaptation can learn the change away. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Freeze or bound updates. 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. Variance change

The central mathematical idea is concrete. Check whether spread also changes.

A concrete situation

Noise becomes larger after maintenance.

What the mathematics does not license

Mean-only signals can mislead. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Model variance. 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. Calendar effects

Here the geometry becomes operational. Remove recurring patterns.

A concrete situation

Weekdays have different levels.

What the mathematics does not license

Unmodelled seasonality creates cycles. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Use comparable baselines. 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. Operating study

This step deserves a careful pause. Replay quiet and shifted scenarios.

A concrete situation

False alarms and delays are recorded.

What the mathematics does not license

Synthetic shifts can be too simple. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Use varied shapes. 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

The useful question is not only how to calculate, but what the calculation assumes. Review signal timing and data quality.

A concrete situation

The candidate date guides inspection.

What the mathematics does not license

It is not guaranteed exact. Keeping this boundary clear makes the result easier to audit and transfer.

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.

---

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

Page–Hinkley monitoring looks for a sustained change in the mean by accumulating deviations from a reference level while tracking the most favourable earlier point. A small drift allowance prevents ordinary noise from building the statistic too quickly, and an alarm threshold expresses how much cumulative evidence is required before action.

One common upward-shift form updates a cumulative value mt=mt−1+(xt−μ0−δ), tracks its running minimum Mt=min(Mt−1,mt), and monitors PHt=mt−Mt. When PHt exceeds λ, the chart signals. Equivalent implementations may estimate the reference mean online or use sign-reversed conventions, so formulas and initialization must be stated.

Let μ0=10, δ=0.5 and observations be 11, 12 and 12. Starting at m0=M0=0 gives increments 0.5, 1.5 and 1.5, so m values are 0.5, 2.0 and 3.5. The running minimum stays zero and PH3=3.5. Whether that signals depends on the predeclared threshold λ.

The method does not automatically handle seasonality, autocorrelation, changing variance or time-varying exposure. The baseline, drift allowance and threshold must be calibrated on representative quiet and shifted streams. A signal locates an operational change candidate; confirmation needs data-quality checks and domain evidence.

---

Frequently asked questions

What does Page–Hinkley detect?

It detects sustained shifts in a process mean through cumulative departures from a reference.

What does δ do?

The drift allowance subtracts a small tolerance so routine fluctuations do not accumulate as quickly.

Why track a running minimum?

The distance from the lowest cumulative value identifies the strongest subsequent upward departure.

What is λ?

It is the predeclared alarm threshold for the Page–Hinkley statistic.

Can the method detect downward shifts?

Yes, with a corresponding sign convention or separate lower detector.

Does a signal locate the exact causal moment?

No. It provides a statistical change candidate whose timing and cause require further analysis.

---

Primary source and fact-check note

D. V. Hinkley, “Inference about the Change-Point from Cumulative Sum Tests,” Biometrika 58(3), 509–523 (1971), DOI 10.1093/biomet/58.3.509. 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
• EARS C1, C2 and C3 surveillance
• 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

Reference audit

Justify μ0 and update rules.

Allowance audit

Relate δ to a meaningful shift.

Recursion audit

Reproduce m, M and PH values.

Threshold audit

Calibrate λ before monitoring.

Direction audit

Separate upper and lower charts.

Dependence audit

Stress-test calendar and autocorrelation.

Transfer audit

Freeze settings for later data.

Final claim

Report a change candidate with timing uncertainty.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Measurement definition

Begin with this mechanism: Fix the quantity and interval. Recreate the concrete case: Daily score xt is comparable. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Scale changes create artificial shifts. Complete the transfer step: Lock units. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 2: Reference mean

Begin with this mechanism: Estimate or declare μ0. Recreate the concrete case: Stable mean is 10. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Contaminated history biases monitoring. Complete the transfer step: Use quiet data. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 3: Drift allowance

Begin with this mechanism: Choose δ for tolerated movement. Recreate the concrete case: δ=.5. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Too large hides small shifts. Complete the transfer step: Relate it to action. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 4: First increment

Begin with this mechanism: Compute x1−μ0−δ. Recreate the concrete case: 11−10−.5=.5. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Sign conventions vary. Complete the transfer step: State direction. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 5: Cumulative value

Begin with this mechanism: Add the increment to mt−1. Recreate the concrete case: m1=.5. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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 can accumulate. Complete the transfer step: Keep precision. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 6: Running minimum

Begin with this mechanism: Update Mt=min(Mt−1,mt). Recreate the concrete case: M1 remains zero. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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 matters. Complete the transfer step: Publish M0. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 7: PH statistic

Begin with this mechanism: Compute mt−Mt. Recreate the concrete case: PH1=.5. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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 is not a p-value. Complete the transfer step: Use operational language. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 8: Second observation

Begin with this mechanism: Continue the recursion. Recreate the concrete case: x2=12 gives m2=2.0. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: One high point may be noise. Complete the transfer step: Keep raw data visible. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 9: Third observation

Begin with this mechanism: Measure persistence. Recreate the concrete case: x3=12 gives PH3=3.5. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Autocorrelation amplifies runs. Complete the transfer step: Simulate dependence. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 10: Alarm threshold

Begin with this mechanism: Signal when PH exceeds λ. Recreate the concrete case: λ is fixed before monitoring. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Small λ creates alert fatigue. Complete the transfer step: Calibrate jointly. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 11: Downward chart

Begin with this mechanism: Reverse deviations for decreases. Recreate the concrete case: A separate lower statistic is maintained. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Mixing directions confuses thresholds. Complete the transfer step: Label each chart. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 12: Unknown mean

Begin with this mechanism: Consider online mean estimation only with rules. Recreate the concrete case: Reference updates gradually. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Adaptation can learn the change away. Complete the transfer step: Freeze or bound updates. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 13: Variance change

Begin with this mechanism: Check whether spread also changes. Recreate the concrete case: Noise becomes larger after maintenance. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Mean-only signals can mislead. Complete the transfer step: Model variance. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 14: Calendar effects

Begin with this mechanism: Remove recurring patterns. Recreate the concrete case: Weekdays have different levels. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Unmodelled seasonality creates cycles. Complete the transfer step: Use comparable baselines. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 15: Operating study

Begin with this mechanism: Replay quiet and shifted scenarios. Recreate the concrete case: False alarms and delays are recorded. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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: Synthetic shifts can be too simple. Complete the transfer step: Use varied shapes. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.

Workshop 16: Investigation

Begin with this mechanism: Review signal timing and data quality. Recreate the concrete case: The candidate date guides inspection. Write measurements, reference means, drift allowances, signed increments, cumulative values, running minima, PH statistics, thresholds, stopping times, false-alarm rates, detection delays and investigation 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 is not guaranteed exact. Complete the transfer step: Report uncertainty. Change exactly one observation, timestamp, gap, rank, tie rule, candidate split, reference mean, drift allowance, running minimum, likelihood model, change size, likelihood ratio, initial value, threshold, state multiplier, transition penalty, state ceiling, null model, simulation seed, dependence pattern, monitoring horizon or validation period. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, preprocessing, model, recursion or search, calibration, uncertainty, interpretation and independent validation on separate notebook lines. Reserve complete periods outside tuning when testing transfer.