Assurance, Incidents and Safety Leadership

Learning objective. Build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews.

Linked synthetic case. A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices.

What an adequate output contains. Good news in personal injury statistics cannot establish process containment integrity; seek evidence that critical controls function on demand.

Why this unit is needed

Process safety concerns preventing and limiting catastrophic releases of hazardous substances or energy, rather than merely maintaining a low count of routine injuries. A real project cannot infer its safety from a single score, accident-free period or calculation. For this unit, build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews. The basis of this reasoning is the physical configuration of equipment and operations, not the appearance of a matrix or dashboard. Assign each critical assertion an owner and identify how field observations could contradict it.

The case used across the lessons is deliberately hypothetical: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. It contains no named real chemicals, equipment certificates, actual incident frequencies or approved operating limits. That distinction is central to interpreting models. For a plant, process safety information must describe what can be hazardous, how the process can deviate, and which barriers genuinely remain functional in abnormal conditions. A model may reveal sensitivity while omitting the mechanism that actually dominates a serious accident.

The five-layer engineering argument

1. Hazard and inventory: Identify substances, energy and populations or environmental receptors that could be exposed. Document what remains unknown before assigning a numerical risk category.

2. Initiating mechanism: State a credible cause, including external utilities, human-machine interfaces, maintenance, abnormal startup and non-routine conditions; separate initiating cause from the loss-of-control top event.

3. Prevention: Identify physical design limits, independently verified interlocks, procedures and maintenance controls. Do not assume independence without checking power, instruments and common personnel.

4. Mitigation: Assess containment, shutdown, relief and emergency arrangements in the context of the actual consequence mechanism. A mitigation may reduce severity without reducing the initiating frequency.

5. Assurance: Specify evidence, proof tests, field observations and responsible action owners. Controls that are unavailable, bypassed or degraded must not be credited as if fully available.

How to use the three numerical laboratories

These interactive models intentionally isolate individual relationships rather than claiming to predict a hazardous plant. Before moving a slider, write the governing equation and the physical units. Decide which outcomes should rise or fall when one parameter changes. After moving it, check whether the response and the plotted curve agree with that expectation. Export numerical data and calculate one point independently. If a curve seems favorable, identify a missing mechanism that could reverse the real decision.

This unit's laboratories are:

• [Lab 22: Critical-action overdue fraction](../labs/lab-22.html) — Explore what an aggregate overdue percentage hides about high-severity controls. Boundary: Counts not weighted by criticality; late cannot exceed due and is capped in model.
• [Lab 23: Inspection completion and coverage](../labs/lab-23.html) — Track inspection completion without confusing it with equipment condition. Boundary: Scheduling indicator; does not measure detection sensitivity, defect severity or fitness-for-service.
• [Lab 24: Binomial success uncertainty](../labs/lab-24.html) — Use a Wilson confidence interval for illustrative demand-test success, not a safety integrity claim. Boundary: Tests may be nonindependent and not representative of hazardous demands; binomial assumptions may fail.

How to review evidence

For each lesson prepare a three-column evidence ledger: declared fact, supporting observation/source, and status (measured, specified, assumed, inferred, or unresolved). Do not promote an assumption to a measurement when an equation happens to give a precise answer. In a team review, challenge the source of every probability and time constant. If an engineering standard is cited, verify the correct edition and legal applicability; standards differ by jurisdiction and project date. For serious events, methods used by qualified specialists should replace these elementary illustrations.

The course encourages deliberate disagreement: two reviewers should independently identify the hazardous deviation and potentially defeated safeguards before reconciling their lists. Record both the agreement and remaining uncertainty. The presence of a highly consequential unresolved scenario should not be hidden inside a low composite average. Closure means that a response has been implemented and verified to work, not that a spreadsheet cell has been colored green.

Progression and assessment

Complete the four lessons in order. Each assessment demands a short derivation, a sensitivity or limitation check, a scenario analysis and a documented decision. A numerical result without a stated boundary, units and credible evidence should receive no engineering credit. The final unit project asks for an auditable decision memo that separates a screening observation from an approval to build or operate. Peer review should specifically challenge common-cause failures and untested assumptions.

Lesson 1: Leading versus lagging safety indicators

Lagging outcomes capture events that occurred; leading indicators track the health of critical activities that may prevent them. A low event count is not enough when barrier impairments and deferred actions rise. Key analytical anchor: Activity completion = completed safety-critical actions / due actions. Investigate the following case: A month with zero major incidents and 60% overdue barrier actions does not demonstrate strong control assurance. The most likely conceptual error is comparing raw event counts across sites without exposure, definitions or reporting culture. In the lesson, assemble pse classification, critical maintenance backlog, audit actions and control tests before drawing the operational conclusion: use a balanced dashboard that includes leading barrier integrity evidence. Further discussion: Specify one metric whose improvement could perversely mean underreporting and a countermeasure.

Lesson 2: Incident investigation and learning loops

An incident review should preserve evidence, reconstruct a timeline and identify weak controls and organizational contributors. Recommendations need owners, deadlines, effectiveness checks and cross-site learning. Key analytical anchor: Learning loop = event → causal analysis → corrective action → effectiveness test. Investigate the following case: After a release, replacing the failed sensor without investigating common power loss may permit recurrence. The most likely conceptual error is stopping at proximate operator action rather than causes of barrier vulnerability. In the lesson, assemble event evidence, historian data, maintenance records, training logs and interviews before drawing the operational conclusion: test whether corrective actions remove or control the identified failure pathway. Further discussion: Write a causal tree and propose a measured effectiveness test after 90 days.

Lesson 3: Audits, assurance and governance

Compliance audits, hazard analysis revalidation and equipment inspections have different purposes. Governance assigns credible escalation rights and documents unresolved decisions, rather than equating a closed spreadsheet row to hazard control. Key analytical anchor: Assurance = independent challenge + traceable evidence + closure verification. Investigate the following case: An audit is marked closed because a procedure was revised but no worker has been trained or assessed; the hazard pathway may remain. The most likely conceptual error is confusing document completion with verified control performance. In the lesson, assemble audit reports, implementation evidence, operator competency and independent closure checks before drawing the operational conclusion: audit unresolved critical safeguards and require risk ownership. Further discussion: Create an evidence chain from finding to physical field verification and next review date.

Lesson 4: Safety case integration and practical decision-making

A safety case connects hazard understanding, justified safeguards, evidence and accountable decisions. It is a structured argument whose claims are limited by the available evidence and applicable law. Key analytical anchor: Claim → argument → evidence → residual uncertainty. Investigate the following case: A proposed expansion raises inventory and changes operating shifts; list safety-case assertions that must be revisited. The most likely conceptual error is treating calculated low risk from generic inputs as permission to operate. In the lesson, assemble validated chemistry and equipment data, specialist assessments, management systems and jurisdiction review before drawing the operational conclusion: make proceed/no-go conditions and remaining uncertainties explicit. Further discussion: Produce a five-page decision memo stating what is proven, conditional and still unknown.

Unit completion checkpoint

The minimum acceptable submission contains an annotated process boundary for assurance, incidents and safety leadership, a cause-to-consequence pathway, a register of assumptions, an independent numerical reproduction of each model, and an unresolved-information list. In the synthetic case, a site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Identify which missing single item is most likely to change your recommendation and why. Then state who would have authority to accept or reject the real project after a specialist review.

Unit 8 · Lesson 1: Leading versus lagging safety indicators

[Unit overview](../units/unit-08.html) · [Interactive lab 22](../labs/lab-22.html) · [Assessment](../assessments/unit-08.html) · [Applied project](../projects/unit-08.html)

The engineering question

Lagging outcomes capture events that occurred; leading indicators track the health of critical activities that may prevent them. A low event count is not enough when barrier impairments and deferred actions rise. This lesson belongs to Assurance, Incidents and Safety Leadership, where the course objective is to build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews. The governing issue is not whether one can calculate a number; it is whether that number describes the failure mechanism under study. A competent analysis separates a claim about what happens, an argument about why it would happen, and evidence that the safeguards could withstand real operation.

Specific thought experiment. A month with zero major incidents and 60% overdue barrier actions does not demonstrate strong control assurance. Work from the equipment boundary outward. Name the initiating event, the operating mode, the top event, the potentially exposed people or environment, and the time over which the consequences could develop. If any one of these is unspecified, maintain the uncertainty rather than inventing a value.

Principles and analytical relationship

The working relation or structured method is Activity completion = completed safety-critical actions / due actions. It is useful because it forces an analyst to state inputs and ask whether the units, time basis and conditioning assumptions match. A rate per year cannot be multiplied by an arbitrary probability without asking what event that probability conditions upon. Similarly, an ordinary fraction is not a mass fraction until the relevant mass balance has been declared. In a causal diagram, an arrow represents a proposed physical or organizational dependency; it is not, by itself, proof of causality.

The physical interpretation for leading versus lagging safety indicators is the following: Lagging outcomes capture events that occurred; leading indicators track the health of critical activities that may prevent them. A low event count is not enough when barrier impairments and deferred actions rise. Thus the result needs an audit trail back to observations, design documents or a controlled estimate. There is no universal conversion from a neat calculation to safe operation. Where the relationship is qualitative, a complete answer explains the conditions for each step and which alternative cause could produce the same observed symptom. When the relation is quantitative, a complete answer supplies units, ranges and a check against a limiting case.

For the course case, a site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Consider three boundaries before deciding whether the model is applicable: the system under study, the observation period, and the intervention boundary. System boundaries determine which inventories and energy flows are included. Observation periods distinguish transient loss of control from a steady process average. Intervention boundaries identify where automatic protection, human response or passive containment can plausibly act. A safeguard that acts after the harm does not prevent the initiating event, although it may limit the consequence.

Worked reasoning example

Unverified numerical or qualitative illustration: A month with zero major incidents and 60% overdue barrier actions does not demonstrate strong control assurance. Begin with the stated conditions only. Write the relationships that must hold, then identify at least one variable not specified in the example. If an exact arithmetic result is given, reproduce it independently and add a clear qualifier about the simplified model. If the example is qualitative, draw a two-branch argument: one where the existing barrier is available and one where it is defeated by a plausible condition. Do not fill missing branches with assumed success simply to obtain a final score.

A useful way to test the claim is to ask what change would produce the opposite conclusion. For this topic, the specific analytical trap is comparing raw event counts across sites without exposure, definitions or reporting culture. Try to construct a counterexample with the same nominal process equipment but an altered startup, maintenance, utility, human or instrument condition. A single plausible counterexample indicates that the original claim was conditional. It does not automatically establish that the process is unsafe; it establishes which fact must be investigated before any stronger judgment.

Evidence and traceability

The minimum relevant evidence set is pse classification, critical maintenance backlog, audit actions and control tests. For each item record source, revision, date, scope, owner and confidence. Separate field-measured values from vendor guarantees, industry generic estimates and deliberately fictional learning inputs. For example, two pumps sold at the same nominal capacity do not necessarily share seal arrangements, failure modes or instrument interfaces. Even an authoritative standard does not substitute for actual testing or determine whether a particular facility falls within its legal scope.

An effective review checks whether evidence is current for the operating mode. Startup differs from steady production; temporary operation after a maintenance outage may differ from both. Procedures, alarms and emergency equipment must be evaluated for those modes explicitly. A reliability assertion based on one short interval must not silently be extended to years of service. Missing evidence should generate a specific work request instead of a generic recommendation to “improve safety.”

Laboratory investigation and sensitivity

Use [Lab 22: Critical-action overdue fraction](../labs/lab-22.html). The printed equation is Overdue fraction = N_overdue / N_due. Explore what an aggregate overdue percentage hides about high-severity controls. Start from each slider's default, record the displayed metric and vary the first parameter over five or more settings. The graph and CSV should agree with your hand calculations. Next, change one other parameter while keeping the first fixed; this distinguishes a mathematical sensitivity from a trend caused by simultaneous input changes.

Investigation target: If only two tasks involve emergency trips, why might the overall percentage mislead? The major simplification is Counts not weighted by criticality; late cannot exceed due and is capped in model.. A good lab report predicts the sign of the derivative or change before observing the plot, then explains what physical process that derivative represents. If an output is nonphysical or a limiting denominator becomes zero, it should be reported as an invalid domain rather than manipulated into a finite reassuring result. Place the lesson's analytical relationship alongside the lab relation; explain whether they describe the same physical quantity or different levels of abstraction.

As a challenge, compare this model with [Lab 23: Inspection completion and coverage](../labs/lab-23.html). The second model's relation is Coverage = inspected_due / inspections_due, and its stated limitation is Scheduling indicator; does not measure detection sensitivity, defect severity or fitness-for-service.. The comparison is not an invitation to combine numbers blindly. Explain what would need to be measured so the two simplified relationships could appear in one defensible engineering argument.

Failure-first decision review

The unacceptable shortcut is comparing raw event counts across sites without exposure, definitions or reporting culture. Make a table with columns assumption, possible failure, observation that would expose it, cheapest credible check, owner, and decision consequence. For this particular topic the review should prioritize the evidence sources above rather than produce a broad list of every conceivable failure. Then ask what is the most important remaining uncertainty if all routine checks pass. Low-frequency high-consequence scenarios often cannot be validated simply by waiting for an incident.

The recommended analytical action is: Use a balanced dashboard that includes leading barrier integrity evidence. This recommendation is a learning exercise, not a control-system change or site-specific directive. Describe its boundaries. What would have to be established by independent qualified review? How would an operator learn that the intended protective measure had become unavailable? Which record would demonstrate closure after a modification? A paper action with no verification is not the same as an effective protective function.

Interpreting uncertainty and competing explanations

Uncertainty enters from input data, model structure, operating variability and human or organizational factors. Input uncertainty might be reduced by a calibrated measurement, whereas model-structure uncertainty demands a different physical model or field trial. Increasing the number of decimal places addresses neither. Separate uncertainty that could reverse the conclusion from uncertainty that changes only the reported numerical precision. The severity of a credible major-accident scenario should influence how urgently an evidence gap is reviewed, even when its probability is difficult to quantify.

For leading versus lagging safety indicators, an apparently successful barrier test may demonstrate only the modes exercised by the test. It might not probe lost utilities, common-cause failure, maintenance bypass or the specific initiating condition. A useful counterfactual question is: If the expected result were wrong tomorrow, which observation would reveal it first? Identify an observable precursor and a clear escalation action. This turns an educational calculation into a disciplined evidence requirement without representing it as a plant certification.

Independent exercises

1. Boundary and mechanism: Specify one metric whose improvement could perversely mean underreporting and a countermeasure. Label at least one fact, one assumption and one unknown.

2. Quantitative check: Reproduce one result from Lab 22. Show units, the governing relationship and a simple limiting-case or conservation check.

3. Scenario variation: Replace one favorable assumption with an unfavorable but plausible operating condition. Explain which safeguarding claim is weakened and why.

4. Decision note: Summarize in 200–300 words whether use a balanced dashboard that includes leading barrier integrity evidence. Specify the additional observation that could change your conclusion.

5. Critical review: Explain why comparing raw event counts across sites without exposure, definitions or reporting culture would be misleading when a high-consequence scenario is involved.

Lesson outcome checklist

• A credible scenario chain is drawn, without conflating initiating cause, top event and outcome.
• Model assumptions and units are explicit and the numerical output is checked independently.
• Any barrier credit is tied to function, independence, availability, evidence and an owner.
• The decision specifies what remains unverified; it does not present an educational model as a safe operating limit.

Unit 8 · Lesson 2: Incident investigation and learning loops

[Unit overview](../units/unit-08.html) · [Interactive lab 23](../labs/lab-23.html) · [Assessment](../assessments/unit-08.html) · [Applied project](../projects/unit-08.html)

The engineering question

An incident review should preserve evidence, reconstruct a timeline and identify weak controls and organizational contributors. Recommendations need owners, deadlines, effectiveness checks and cross-site learning. This lesson belongs to Assurance, Incidents and Safety Leadership, where the course objective is to build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews. The governing issue is not whether one can calculate a number; it is whether that number describes the failure mechanism under study. A competent analysis separates a claim about what happens, an argument about why it would happen, and evidence that the safeguards could withstand real operation.

Specific thought experiment. After a release, replacing the failed sensor without investigating common power loss may permit recurrence. Work from the equipment boundary outward. Name the initiating event, the operating mode, the top event, the potentially exposed people or environment, and the time over which the consequences could develop. If any one of these is unspecified, maintain the uncertainty rather than inventing a value.

Principles and analytical relationship

The working relation or structured method is Learning loop = event → causal analysis → corrective action → effectiveness test. It is useful because it forces an analyst to state inputs and ask whether the units, time basis and conditioning assumptions match. A rate per year cannot be multiplied by an arbitrary probability without asking what event that probability conditions upon. Similarly, an ordinary fraction is not a mass fraction until the relevant mass balance has been declared. In a causal diagram, an arrow represents a proposed physical or organizational dependency; it is not, by itself, proof of causality.

The physical interpretation for incident investigation and learning loops is the following: An incident review should preserve evidence, reconstruct a timeline and identify weak controls and organizational contributors. Recommendations need owners, deadlines, effectiveness checks and cross-site learning. Thus the result needs an audit trail back to observations, design documents or a controlled estimate. There is no universal conversion from a neat calculation to safe operation. Where the relationship is qualitative, a complete answer explains the conditions for each step and which alternative cause could produce the same observed symptom. When the relation is quantitative, a complete answer supplies units, ranges and a check against a limiting case.

For the course case, a site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Consider three boundaries before deciding whether the model is applicable: the system under study, the observation period, and the intervention boundary. System boundaries determine which inventories and energy flows are included. Observation periods distinguish transient loss of control from a steady process average. Intervention boundaries identify where automatic protection, human response or passive containment can plausibly act. A safeguard that acts after the harm does not prevent the initiating event, although it may limit the consequence.

Worked reasoning example

Unverified numerical or qualitative illustration: After a release, replacing the failed sensor without investigating common power loss may permit recurrence. Begin with the stated conditions only. Write the relationships that must hold, then identify at least one variable not specified in the example. If an exact arithmetic result is given, reproduce it independently and add a clear qualifier about the simplified model. If the example is qualitative, draw a two-branch argument: one where the existing barrier is available and one where it is defeated by a plausible condition. Do not fill missing branches with assumed success simply to obtain a final score.

A useful way to test the claim is to ask what change would produce the opposite conclusion. For this topic, the specific analytical trap is stopping at proximate operator action rather than causes of barrier vulnerability. Try to construct a counterexample with the same nominal process equipment but an altered startup, maintenance, utility, human or instrument condition. A single plausible counterexample indicates that the original claim was conditional. It does not automatically establish that the process is unsafe; it establishes which fact must be investigated before any stronger judgment.

Evidence and traceability

The minimum relevant evidence set is event evidence, historian data, maintenance records, training logs and interviews. For each item record source, revision, date, scope, owner and confidence. Separate field-measured values from vendor guarantees, industry generic estimates and deliberately fictional learning inputs. For example, two pumps sold at the same nominal capacity do not necessarily share seal arrangements, failure modes or instrument interfaces. Even an authoritative standard does not substitute for actual testing or determine whether a particular facility falls within its legal scope.

An effective review checks whether evidence is current for the operating mode. Startup differs from steady production; temporary operation after a maintenance outage may differ from both. Procedures, alarms and emergency equipment must be evaluated for those modes explicitly. A reliability assertion based on one short interval must not silently be extended to years of service. Missing evidence should generate a specific work request instead of a generic recommendation to “improve safety.”

Laboratory investigation and sensitivity

Use [Lab 23: Inspection completion and coverage](../labs/lab-23.html). The printed equation is Coverage = inspected_due / inspections_due. Track inspection completion without confusing it with equipment condition. Start from each slider's default, record the displayed metric and vary the first parameter over five or more settings. The graph and CSV should agree with your hand calculations. Next, change one other parameter while keeping the first fixed; this distinguishes a mathematical sensitivity from a trend caused by simultaneous input changes.

Investigation target: Why can 95% coverage be inadequate if one missed item is the only independent protection? The major simplification is Scheduling indicator; does not measure detection sensitivity, defect severity or fitness-for-service.. A good lab report predicts the sign of the derivative or change before observing the plot, then explains what physical process that derivative represents. If an output is nonphysical or a limiting denominator becomes zero, it should be reported as an invalid domain rather than manipulated into a finite reassuring result. Place the lesson's analytical relationship alongside the lab relation; explain whether they describe the same physical quantity or different levels of abstraction.

As a challenge, compare this model with [Lab 24: Binomial success uncertainty](../labs/lab-24.html). The second model's relation is Wilson lower 95% bound for binomial success proportion, and its stated limitation is Tests may be nonindependent and not representative of hazardous demands; binomial assumptions may fail.. The comparison is not an invitation to combine numbers blindly. Explain what would need to be measured so the two simplified relationships could appear in one defensible engineering argument.

Failure-first decision review

The unacceptable shortcut is stopping at proximate operator action rather than causes of barrier vulnerability. Make a table with columns assumption, possible failure, observation that would expose it, cheapest credible check, owner, and decision consequence. For this particular topic the review should prioritize the evidence sources above rather than produce a broad list of every conceivable failure. Then ask what is the most important remaining uncertainty if all routine checks pass. Low-frequency high-consequence scenarios often cannot be validated simply by waiting for an incident.

The recommended analytical action is: Test whether corrective actions remove or control the identified failure pathway. This recommendation is a learning exercise, not a control-system change or site-specific directive. Describe its boundaries. What would have to be established by independent qualified review? How would an operator learn that the intended protective measure had become unavailable? Which record would demonstrate closure after a modification? A paper action with no verification is not the same as an effective protective function.

Interpreting uncertainty and competing explanations

Uncertainty enters from input data, model structure, operating variability and human or organizational factors. Input uncertainty might be reduced by a calibrated measurement, whereas model-structure uncertainty demands a different physical model or field trial. Increasing the number of decimal places addresses neither. Separate uncertainty that could reverse the conclusion from uncertainty that changes only the reported numerical precision. The severity of a credible major-accident scenario should influence how urgently an evidence gap is reviewed, even when its probability is difficult to quantify.

For incident investigation and learning loops, an apparently successful barrier test may demonstrate only the modes exercised by the test. It might not probe lost utilities, common-cause failure, maintenance bypass or the specific initiating condition. A useful counterfactual question is: If the expected result were wrong tomorrow, which observation would reveal it first? Identify an observable precursor and a clear escalation action. This turns an educational calculation into a disciplined evidence requirement without representing it as a plant certification.

Independent exercises

1. Boundary and mechanism: Write a causal tree and propose a measured effectiveness test after 90 days. Label at least one fact, one assumption and one unknown.

2. Quantitative check: Reproduce one result from Lab 23. Show units, the governing relationship and a simple limiting-case or conservation check.

3. Scenario variation: Replace one favorable assumption with an unfavorable but plausible operating condition. Explain which safeguarding claim is weakened and why.

4. Decision note: Summarize in 200–300 words whether test whether corrective actions remove or control the identified failure pathway. Specify the additional observation that could change your conclusion.

5. Critical review: Explain why stopping at proximate operator action rather than causes of barrier vulnerability would be misleading when a high-consequence scenario is involved.

Lesson outcome checklist

• A credible scenario chain is drawn, without conflating initiating cause, top event and outcome.
• Model assumptions and units are explicit and the numerical output is checked independently.
• Any barrier credit is tied to function, independence, availability, evidence and an owner.
• The decision specifies what remains unverified; it does not present an educational model as a safe operating limit.

Unit 8 · Lesson 3: Audits, assurance and governance

[Unit overview](../units/unit-08.html) · [Interactive lab 24](../labs/lab-24.html) · [Assessment](../assessments/unit-08.html) · [Applied project](../projects/unit-08.html)

The engineering question

Compliance audits, hazard analysis revalidation and equipment inspections have different purposes. Governance assigns credible escalation rights and documents unresolved decisions, rather than equating a closed spreadsheet row to hazard control. This lesson belongs to Assurance, Incidents and Safety Leadership, where the course objective is to build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews. The governing issue is not whether one can calculate a number; it is whether that number describes the failure mechanism under study. A competent analysis separates a claim about what happens, an argument about why it would happen, and evidence that the safeguards could withstand real operation.

Specific thought experiment. An audit is marked closed because a procedure was revised but no worker has been trained or assessed; the hazard pathway may remain. Work from the equipment boundary outward. Name the initiating event, the operating mode, the top event, the potentially exposed people or environment, and the time over which the consequences could develop. If any one of these is unspecified, maintain the uncertainty rather than inventing a value.

Principles and analytical relationship

The working relation or structured method is Assurance = independent challenge + traceable evidence + closure verification. It is useful because it forces an analyst to state inputs and ask whether the units, time basis and conditioning assumptions match. A rate per year cannot be multiplied by an arbitrary probability without asking what event that probability conditions upon. Similarly, an ordinary fraction is not a mass fraction until the relevant mass balance has been declared. In a causal diagram, an arrow represents a proposed physical or organizational dependency; it is not, by itself, proof of causality.

The physical interpretation for audits, assurance and governance is the following: Compliance audits, hazard analysis revalidation and equipment inspections have different purposes. Governance assigns credible escalation rights and documents unresolved decisions, rather than equating a closed spreadsheet row to hazard control. Thus the result needs an audit trail back to observations, design documents or a controlled estimate. There is no universal conversion from a neat calculation to safe operation. Where the relationship is qualitative, a complete answer explains the conditions for each step and which alternative cause could produce the same observed symptom. When the relation is quantitative, a complete answer supplies units, ranges and a check against a limiting case.

For the course case, a site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Consider three boundaries before deciding whether the model is applicable: the system under study, the observation period, and the intervention boundary. System boundaries determine which inventories and energy flows are included. Observation periods distinguish transient loss of control from a steady process average. Intervention boundaries identify where automatic protection, human response or passive containment can plausibly act. A safeguard that acts after the harm does not prevent the initiating event, although it may limit the consequence.

Worked reasoning example

Unverified numerical or qualitative illustration: An audit is marked closed because a procedure was revised but no worker has been trained or assessed; the hazard pathway may remain. Begin with the stated conditions only. Write the relationships that must hold, then identify at least one variable not specified in the example. If an exact arithmetic result is given, reproduce it independently and add a clear qualifier about the simplified model. If the example is qualitative, draw a two-branch argument: one where the existing barrier is available and one where it is defeated by a plausible condition. Do not fill missing branches with assumed success simply to obtain a final score.

A useful way to test the claim is to ask what change would produce the opposite conclusion. For this topic, the specific analytical trap is confusing document completion with verified control performance. Try to construct a counterexample with the same nominal process equipment but an altered startup, maintenance, utility, human or instrument condition. A single plausible counterexample indicates that the original claim was conditional. It does not automatically establish that the process is unsafe; it establishes which fact must be investigated before any stronger judgment.

Evidence and traceability

The minimum relevant evidence set is audit reports, implementation evidence, operator competency and independent closure checks. For each item record source, revision, date, scope, owner and confidence. Separate field-measured values from vendor guarantees, industry generic estimates and deliberately fictional learning inputs. For example, two pumps sold at the same nominal capacity do not necessarily share seal arrangements, failure modes or instrument interfaces. Even an authoritative standard does not substitute for actual testing or determine whether a particular facility falls within its legal scope.

An effective review checks whether evidence is current for the operating mode. Startup differs from steady production; temporary operation after a maintenance outage may differ from both. Procedures, alarms and emergency equipment must be evaluated for those modes explicitly. A reliability assertion based on one short interval must not silently be extended to years of service. Missing evidence should generate a specific work request instead of a generic recommendation to “improve safety.”

Laboratory investigation and sensitivity

Use [Lab 24: Binomial success uncertainty](../labs/lab-24.html). The printed equation is Wilson lower 95% bound for binomial success proportion. Use a Wilson confidence interval for illustrative demand-test success, not a safety integrity claim. Start from each slider's default, record the displayed metric and vary the first parameter over five or more settings. The graph and CSV should agree with your hand calculations. Next, change one other parameter while keeping the first fixed; this distinguishes a mathematical sensitivity from a trend caused by simultaneous input changes.

Investigation target: Compare 1/1 success and 99/100 success: why are observed rates not enough? The major simplification is Tests may be nonindependent and not representative of hazardous demands; binomial assumptions may fail.. A good lab report predicts the sign of the derivative or change before observing the plot, then explains what physical process that derivative represents. If an output is nonphysical or a limiting denominator becomes zero, it should be reported as an invalid domain rather than manipulated into a finite reassuring result. Place the lesson's analytical relationship alongside the lab relation; explain whether they describe the same physical quantity or different levels of abstraction.

As a challenge, compare this model with [Lab 22: Critical-action overdue fraction](../labs/lab-22.html). The second model's relation is Overdue fraction = N_overdue / N_due, and its stated limitation is Counts not weighted by criticality; late cannot exceed due and is capped in model.. The comparison is not an invitation to combine numbers blindly. Explain what would need to be measured so the two simplified relationships could appear in one defensible engineering argument.

Failure-first decision review

The unacceptable shortcut is confusing document completion with verified control performance. Make a table with columns assumption, possible failure, observation that would expose it, cheapest credible check, owner, and decision consequence. For this particular topic the review should prioritize the evidence sources above rather than produce a broad list of every conceivable failure. Then ask what is the most important remaining uncertainty if all routine checks pass. Low-frequency high-consequence scenarios often cannot be validated simply by waiting for an incident.

The recommended analytical action is: Audit unresolved critical safeguards and require risk ownership. This recommendation is a learning exercise, not a control-system change or site-specific directive. Describe its boundaries. What would have to be established by independent qualified review? How would an operator learn that the intended protective measure had become unavailable? Which record would demonstrate closure after a modification? A paper action with no verification is not the same as an effective protective function.

Interpreting uncertainty and competing explanations

Uncertainty enters from input data, model structure, operating variability and human or organizational factors. Input uncertainty might be reduced by a calibrated measurement, whereas model-structure uncertainty demands a different physical model or field trial. Increasing the number of decimal places addresses neither. Separate uncertainty that could reverse the conclusion from uncertainty that changes only the reported numerical precision. The severity of a credible major-accident scenario should influence how urgently an evidence gap is reviewed, even when its probability is difficult to quantify.

For audits, assurance and governance, an apparently successful barrier test may demonstrate only the modes exercised by the test. It might not probe lost utilities, common-cause failure, maintenance bypass or the specific initiating condition. A useful counterfactual question is: If the expected result were wrong tomorrow, which observation would reveal it first? Identify an observable precursor and a clear escalation action. This turns an educational calculation into a disciplined evidence requirement without representing it as a plant certification.

Independent exercises

1. Boundary and mechanism: Create an evidence chain from finding to physical field verification and next review date. Label at least one fact, one assumption and one unknown.

2. Quantitative check: Reproduce one result from Lab 24. Show units, the governing relationship and a simple limiting-case or conservation check.

3. Scenario variation: Replace one favorable assumption with an unfavorable but plausible operating condition. Explain which safeguarding claim is weakened and why.

4. Decision note: Summarize in 200–300 words whether audit unresolved critical safeguards and require risk ownership. Specify the additional observation that could change your conclusion.

5. Critical review: Explain why confusing document completion with verified control performance would be misleading when a high-consequence scenario is involved.

Lesson outcome checklist

• A credible scenario chain is drawn, without conflating initiating cause, top event and outcome.
• Model assumptions and units are explicit and the numerical output is checked independently.
• Any barrier credit is tied to function, independence, availability, evidence and an owner.
• The decision specifies what remains unverified; it does not present an educational model as a safe operating limit.

Unit 8 · Lesson 4: Safety case integration and practical decision-making

[Unit overview](../units/unit-08.html) · [Interactive lab 22](../labs/lab-22.html) · [Assessment](../assessments/unit-08.html) · [Applied project](../projects/unit-08.html)

The engineering question

A safety case connects hazard understanding, justified safeguards, evidence and accountable decisions. It is a structured argument whose claims are limited by the available evidence and applicable law. This lesson belongs to Assurance, Incidents and Safety Leadership, where the course objective is to build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews. The governing issue is not whether one can calculate a number; it is whether that number describes the failure mechanism under study. A competent analysis separates a claim about what happens, an argument about why it would happen, and evidence that the safeguards could withstand real operation.

Specific thought experiment. A proposed expansion raises inventory and changes operating shifts; list safety-case assertions that must be revisited. Work from the equipment boundary outward. Name the initiating event, the operating mode, the top event, the potentially exposed people or environment, and the time over which the consequences could develop. If any one of these is unspecified, maintain the uncertainty rather than inventing a value.

Principles and analytical relationship

The working relation or structured method is Claim → argument → evidence → residual uncertainty. It is useful because it forces an analyst to state inputs and ask whether the units, time basis and conditioning assumptions match. A rate per year cannot be multiplied by an arbitrary probability without asking what event that probability conditions upon. Similarly, an ordinary fraction is not a mass fraction until the relevant mass balance has been declared. In a causal diagram, an arrow represents a proposed physical or organizational dependency; it is not, by itself, proof of causality.

The physical interpretation for safety case integration and practical decision-making is the following: A safety case connects hazard understanding, justified safeguards, evidence and accountable decisions. It is a structured argument whose claims are limited by the available evidence and applicable law. Thus the result needs an audit trail back to observations, design documents or a controlled estimate. There is no universal conversion from a neat calculation to safe operation. Where the relationship is qualitative, a complete answer explains the conditions for each step and which alternative cause could produce the same observed symptom. When the relation is quantitative, a complete answer supplies units, ranges and a check against a limiting case.

For the course case, a site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Consider three boundaries before deciding whether the model is applicable: the system under study, the observation period, and the intervention boundary. System boundaries determine which inventories and energy flows are included. Observation periods distinguish transient loss of control from a steady process average. Intervention boundaries identify where automatic protection, human response or passive containment can plausibly act. A safeguard that acts after the harm does not prevent the initiating event, although it may limit the consequence.

Worked reasoning example

Unverified numerical or qualitative illustration: A proposed expansion raises inventory and changes operating shifts; list safety-case assertions that must be revisited. Begin with the stated conditions only. Write the relationships that must hold, then identify at least one variable not specified in the example. If an exact arithmetic result is given, reproduce it independently and add a clear qualifier about the simplified model. If the example is qualitative, draw a two-branch argument: one where the existing barrier is available and one where it is defeated by a plausible condition. Do not fill missing branches with assumed success simply to obtain a final score.

A useful way to test the claim is to ask what change would produce the opposite conclusion. For this topic, the specific analytical trap is treating calculated low risk from generic inputs as permission to operate. Try to construct a counterexample with the same nominal process equipment but an altered startup, maintenance, utility, human or instrument condition. A single plausible counterexample indicates that the original claim was conditional. It does not automatically establish that the process is unsafe; it establishes which fact must be investigated before any stronger judgment.

Evidence and traceability

The minimum relevant evidence set is validated chemistry and equipment data, specialist assessments, management systems and jurisdiction review. For each item record source, revision, date, scope, owner and confidence. Separate field-measured values from vendor guarantees, industry generic estimates and deliberately fictional learning inputs. For example, two pumps sold at the same nominal capacity do not necessarily share seal arrangements, failure modes or instrument interfaces. Even an authoritative standard does not substitute for actual testing or determine whether a particular facility falls within its legal scope.

An effective review checks whether evidence is current for the operating mode. Startup differs from steady production; temporary operation after a maintenance outage may differ from both. Procedures, alarms and emergency equipment must be evaluated for those modes explicitly. A reliability assertion based on one short interval must not silently be extended to years of service. Missing evidence should generate a specific work request instead of a generic recommendation to “improve safety.”

Laboratory investigation and sensitivity

Use [Lab 22: Critical-action overdue fraction](../labs/lab-22.html). The printed equation is Overdue fraction = N_overdue / N_due. Explore what an aggregate overdue percentage hides about high-severity controls. Start from each slider's default, record the displayed metric and vary the first parameter over five or more settings. The graph and CSV should agree with your hand calculations. Next, change one other parameter while keeping the first fixed; this distinguishes a mathematical sensitivity from a trend caused by simultaneous input changes.

Investigation target: If only two tasks involve emergency trips, why might the overall percentage mislead? The major simplification is Counts not weighted by criticality; late cannot exceed due and is capped in model.. A good lab report predicts the sign of the derivative or change before observing the plot, then explains what physical process that derivative represents. If an output is nonphysical or a limiting denominator becomes zero, it should be reported as an invalid domain rather than manipulated into a finite reassuring result. Place the lesson's analytical relationship alongside the lab relation; explain whether they describe the same physical quantity or different levels of abstraction.

As a challenge, compare this model with [Lab 23: Inspection completion and coverage](../labs/lab-23.html). The second model's relation is Coverage = inspected_due / inspections_due, and its stated limitation is Scheduling indicator; does not measure detection sensitivity, defect severity or fitness-for-service.. The comparison is not an invitation to combine numbers blindly. Explain what would need to be measured so the two simplified relationships could appear in one defensible engineering argument.

Failure-first decision review

The unacceptable shortcut is treating calculated low risk from generic inputs as permission to operate. Make a table with columns assumption, possible failure, observation that would expose it, cheapest credible check, owner, and decision consequence. For this particular topic the review should prioritize the evidence sources above rather than produce a broad list of every conceivable failure. Then ask what is the most important remaining uncertainty if all routine checks pass. Low-frequency high-consequence scenarios often cannot be validated simply by waiting for an incident.

The recommended analytical action is: Make proceed/no-go conditions and remaining uncertainties explicit. This recommendation is a learning exercise, not a control-system change or site-specific directive. Describe its boundaries. What would have to be established by independent qualified review? How would an operator learn that the intended protective measure had become unavailable? Which record would demonstrate closure after a modification? A paper action with no verification is not the same as an effective protective function.

Interpreting uncertainty and competing explanations

Uncertainty enters from input data, model structure, operating variability and human or organizational factors. Input uncertainty might be reduced by a calibrated measurement, whereas model-structure uncertainty demands a different physical model or field trial. Increasing the number of decimal places addresses neither. Separate uncertainty that could reverse the conclusion from uncertainty that changes only the reported numerical precision. The severity of a credible major-accident scenario should influence how urgently an evidence gap is reviewed, even when its probability is difficult to quantify.

For safety case integration and practical decision-making, an apparently successful barrier test may demonstrate only the modes exercised by the test. It might not probe lost utilities, common-cause failure, maintenance bypass or the specific initiating condition. A useful counterfactual question is: If the expected result were wrong tomorrow, which observation would reveal it first? Identify an observable precursor and a clear escalation action. This turns an educational calculation into a disciplined evidence requirement without representing it as a plant certification.

Independent exercises

1. Boundary and mechanism: Produce a five-page decision memo stating what is proven, conditional and still unknown. Label at least one fact, one assumption and one unknown.

2. Quantitative check: Reproduce one result from Lab 22. Show units, the governing relationship and a simple limiting-case or conservation check.

3. Scenario variation: Replace one favorable assumption with an unfavorable but plausible operating condition. Explain which safeguarding claim is weakened and why.

4. Decision note: Summarize in 200–300 words whether make proceed/no-go conditions and remaining uncertainties explicit. Specify the additional observation that could change your conclusion.

5. Critical review: Explain why treating calculated low risk from generic inputs as permission to operate would be misleading when a high-consequence scenario is involved.

Lesson outcome checklist

• A credible scenario chain is drawn, without conflating initiating cause, top event and outcome.
• Model assumptions and units are explicit and the numerical output is checked independently.
• Any barrier credit is tied to function, independence, availability, evidence and an owner.
• The decision specifies what remains unverified; it does not present an educational model as a safe operating limit.

Applied Project 08 — Assurance, Incidents and Safety Leadership

Scenario and objective

Scenario: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices.

Target: Build meaningful leading and lagging indicator sets, incident learning loops, action closeout and risk-based management reviews.

The task is a supervised conceptual study. It is not permission to change operating procedures, introduce hazardous chemicals, bypass protection or size relief devices. Assume that a multidisciplinary panel must decide whether to commission additional analysis, not whether this student workbook certifies a plant. Your deliverable must identify the claim being made, what evidence supports it, and what uncertainty remains. The board values clear red flags over cosmetic completion percentages.

Work package A: describe the situation

Draw a block-flow boundary showing process inventory, incoming feeds or energy, manipulated variables, utilities, protective controls and relevant maintenance interfaces. Record the operating mode, unit conventions, material identity gaps and any location- or jurisdiction-dependent regulatory questions. Write one plausible hazardous event without guessing its annual frequency. For each asserted safeguard, identify what action it takes, when it acts relative to the top event, and which elements it shares with other protective controls.

The case provides a starting point only: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices.. Ask how equipment failure and incorrect operating information could combine. Check a routine operation, startup, shutdown and an interrupted utility condition. Treat any missing process safety information as a question with an owner; do not silently assume that a procedure exists or a transmitter is independent. Make a compact register of at least eight assumptions grouped into physical properties, instrumentation, human response and external conditions.

Work package B: model and verify

Lab 22 — [Critical-action overdue fraction](../labs/lab-22.html). Compute the baseline from Overdue fraction = N_overdue / N_due and export a one-parameter sweep. If only two tasks involve emergency trips, why might the overall percentage mislead? The condition Counts not weighted by criticality; late cannot exceed due and is capped in model. must appear as a red-line limitation in the report. Record at least one changed parameter set and reproduce one plotted point without JavaScript. Explain whether a favorable change in this model would genuinely improve the identified hazard or merely change one simplified quantity.

Lab 23 — [Inspection completion and coverage](../labs/lab-23.html). Compute the baseline from Coverage = inspected_due / inspections_due and export a one-parameter sweep. Why can 95% coverage be inadequate if one missed item is the only independent protection? The condition Scheduling indicator; does not measure detection sensitivity, defect severity or fitness-for-service. must appear as a red-line limitation in the report. Record at least one changed parameter set and reproduce one plotted point without JavaScript. Explain whether a favorable change in this model would genuinely improve the identified hazard or merely change one simplified quantity.

Lab 24 — [Binomial success uncertainty](../labs/lab-24.html). Compute the baseline from Wilson lower 95% bound for binomial success proportion and export a one-parameter sweep. Compare 1/1 success and 99/100 success: why are observed rates not enough? The condition Tests may be nonindependent and not representative of hazardous demands; binomial assumptions may fail. must appear as a red-line limitation in the report. Record at least one changed parameter set and reproduce one plotted point without JavaScript. Explain whether a favorable change in this model would genuinely improve the identified hazard or merely change one simplified quantity.

Work package C: argue the controls

Prepare a bow-tie-style diagram with three credible threats and two possible consequence pathways. Describe at least one prevention design measure, one operating or inspection control, one possible protective-system layer and one mitigation measure. Do not treat all of these as independent or calculate a cumulative protection factor without justified evidence. Identify a shared-cause vulnerability (for example, common power supply, blocked sensing line or maintenance isolation) and show how your recommended verification would detect it.

Create a control register with columns control purpose, covered initiating mechanism, physical element, mode of demand, availability evidence, testing method, owner, and unresolved dependency. A valid control should have a documented performance claim that can be tested in the field. If field testing would introduce danger, propose a qualified nonintrusive verification method or documentation review instead of suggesting an uncontrolled trial.

Work package D: decision, uncertainty and follow-up

Deliver a 600–900-word decision memo with a recommendation of further analysis, redesign for inherent safety, or insufficient evidence to proceed. The answer must not state that a student's simulation proves safety or regulatory compliance. Begin with the most serious potentially unmitigated scenario. Include a one-page assumptions register and a prioritized list of the five most valuable additional data items. Differentiate uncertainty in inputs from uncertainty in the basic model structure. Show how one new piece of evidence could reverse your recommendation.

The expected project artifact is: Good news in personal injury statistics cannot establish process containment integrity; seek evidence that critical controls function on demand. Describe at least three acceptance tests, responsible roles and the evidence each would produce. Finish with a handover list for a qualified engineer or hazard-analysis team. A statement like “we will improve training” is insufficient without specifying which error pathway it is intended to reduce and how the effect would be checked.

Assessment criteria

• Process and time boundaries are explicit and aligned with the chosen hazardous event.
• Equations reproduce the baseline, limits are visible, and CSV data agree with an independent computation.
• Claimed safeguards have separate mechanisms and credible evidence; common-cause weaknesses are challenged.
• Unknown chemical properties, actual frequencies and code requirements are not invented.
• The conclusion is proportional to evidence and contains accountable next actions rather than a blanket approval.

Unit 08 — Assessment: Assurance, Incidents and Safety Leadership

Scenario: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. This assessment contains 16 separately numbered, multi-part tasks, four per lesson. Students should complete the numerical parts independently and explicitly distinguish illustrative assumptions from measured or regulated inputs. There is no site-specific pass/fail safety conclusion.

Learning standard: Good news in personal injury statistics cannot establish process containment integrity; seek evidence that critical controls function on demand.

Scoring rubric

Each task is worth ten marks: physical or causal boundary (2), correct relationship or defensible logic (2), evidence and units (2), explicit limitation (2), and transparent action or interpretation (2). A risk claim without necessary provenance receives no numerical credit regardless of apparent precision. Students may challenge an ill-posed input as part of the solution if they explain which information is missing.

Lesson 1 — Leading versus lagging safety indicators

The four problems below require different kinds of evidence; do not copy a single stock paragraph as four answers.

Q01 — Physical reconstruction (10 marks). Start from Activity completion = completed safety-critical actions / due actions in the following case: A month with zero major incidents and 60% overdue barrier actions does not demonstrate strong control assurance. Identify each term or causal link, the relevant units or event definition, and a reasonable limiting case. Describe at least one operational state in which the relationship would no longer be adequate. For numerical examples, reproduce the printed result with no more significant figures than the inputs justify.

Q02 — Interactive computation and falsification (10 marks). Run [Lab 22](../labs/lab-22.html), Critical-action overdue fraction. Tabulate at least seven points for its first variable, calculate one independent output, then change its second variable. The lab asks: If only two tasks involve emergency trips, why might the overall percentage mislead? Evaluate the specific restriction Counts not weighted by criticality; late cannot exceed due and is capped in model.; provide a case where this restriction matters and state what additional physical model would be needed.

Q03 — Hazard and barrier chain (10 marks). Examine leading versus lagging safety indicators using this hazard background: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Name an initiating mechanism, the loss-of-control top event, one possible consequence, an inherently safer alternative and an independent barrier if credible. The analysis must explicitly challenge the misleading assumption comparing raw event counts across sites without exposure, definitions or reporting culture. Which shared utility, maintenance activity or human action could defeat the claimed protections?

Q04 — Evidence-led technical memo (10 marks). Prepare a 250-word memo answering: Specify one metric whose improvement could perversely mean underreporting and a countermeasure. The recommendation should be use a balanced dashboard that includes leading barrier integrity evidence but you must justify or challenge it. List the real-world records needed: PSE classification, critical maintenance backlog, audit actions and control tests. Separate a hypothetical mathematical statement from an engineering action and identify the competent approver and a falsifiable follow-up test.

Lesson 2 — Incident investigation and learning loops

The four problems below require different kinds of evidence; do not copy a single stock paragraph as four answers.

Q05 — Physical reconstruction (10 marks). Start from Learning loop = event → causal analysis → corrective action → effectiveness test in the following case: After a release, replacing the failed sensor without investigating common power loss may permit recurrence. Identify each term or causal link, the relevant units or event definition, and a reasonable limiting case. Describe at least one operational state in which the relationship would no longer be adequate. For numerical examples, reproduce the printed result with no more significant figures than the inputs justify.

Q06 — Interactive computation and falsification (10 marks). Run [Lab 23](../labs/lab-23.html), Inspection completion and coverage. Tabulate at least seven points for its first variable, calculate one independent output, then change its second variable. The lab asks: Why can 95% coverage be inadequate if one missed item is the only independent protection? Evaluate the specific restriction Scheduling indicator; does not measure detection sensitivity, defect severity or fitness-for-service.; provide a case where this restriction matters and state what additional physical model would be needed.

Q07 — Hazard and barrier chain (10 marks). Examine incident investigation and learning loops using this hazard background: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Name an initiating mechanism, the loss-of-control top event, one possible consequence, an inherently safer alternative and an independent barrier if credible. The analysis must explicitly challenge the misleading assumption stopping at proximate operator action rather than causes of barrier vulnerability. Which shared utility, maintenance activity or human action could defeat the claimed protections?

Q08 — Evidence-led technical memo (10 marks). Prepare a 250-word memo answering: Write a causal tree and propose a measured effectiveness test after 90 days. The recommendation should be test whether corrective actions remove or control the identified failure pathway but you must justify or challenge it. List the real-world records needed: Event evidence, historian data, maintenance records, training logs and interviews. Separate a hypothetical mathematical statement from an engineering action and identify the competent approver and a falsifiable follow-up test.

Lesson 3 — Audits, assurance and governance

The four problems below require different kinds of evidence; do not copy a single stock paragraph as four answers.

Q09 — Physical reconstruction (10 marks). Start from Assurance = independent challenge + traceable evidence + closure verification in the following case: An audit is marked closed because a procedure was revised but no worker has been trained or assessed; the hazard pathway may remain. Identify each term or causal link, the relevant units or event definition, and a reasonable limiting case. Describe at least one operational state in which the relationship would no longer be adequate. For numerical examples, reproduce the printed result with no more significant figures than the inputs justify.

Q10 — Interactive computation and falsification (10 marks). Run [Lab 24](../labs/lab-24.html), Binomial success uncertainty. Tabulate at least seven points for its first variable, calculate one independent output, then change its second variable. The lab asks: Compare 1/1 success and 99/100 success: why are observed rates not enough? Evaluate the specific restriction Tests may be nonindependent and not representative of hazardous demands; binomial assumptions may fail.; provide a case where this restriction matters and state what additional physical model would be needed.

Q11 — Hazard and barrier chain (10 marks). Examine audits, assurance and governance using this hazard background: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Name an initiating mechanism, the loss-of-control top event, one possible consequence, an inherently safer alternative and an independent barrier if credible. The analysis must explicitly challenge the misleading assumption confusing document completion with verified control performance. Which shared utility, maintenance activity or human action could defeat the claimed protections?

Q12 — Evidence-led technical memo (10 marks). Prepare a 250-word memo answering: Create an evidence chain from finding to physical field verification and next review date. The recommendation should be audit unresolved critical safeguards and require risk ownership but you must justify or challenge it. List the real-world records needed: Audit reports, implementation evidence, operator competency and independent closure checks. Separate a hypothetical mathematical statement from an engineering action and identify the competent approver and a falsifiable follow-up test.

Lesson 4 — Safety case integration and practical decision-making

The four problems below require different kinds of evidence; do not copy a single stock paragraph as four answers.

Q13 — Physical reconstruction (10 marks). Start from Claim → argument → evidence → residual uncertainty in the following case: A proposed expansion raises inventory and changes operating shifts; list safety-case assertions that must be revisited. Identify each term or causal link, the relevant units or event definition, and a reasonable limiting case. Describe at least one operational state in which the relationship would no longer be adequate. For numerical examples, reproduce the printed result with no more significant figures than the inputs justify.

Q14 — Interactive computation and falsification (10 marks). Run [Lab 22](../labs/lab-22.html), Critical-action overdue fraction. Tabulate at least seven points for its first variable, calculate one independent output, then change its second variable. The lab asks: If only two tasks involve emergency trips, why might the overall percentage mislead? Evaluate the specific restriction Counts not weighted by criticality; late cannot exceed due and is capped in model.; provide a case where this restriction matters and state what additional physical model would be needed.

Q15 — Hazard and barrier chain (10 marks). Examine safety case integration and practical decision-making using this hazard background: A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices. Name an initiating mechanism, the loss-of-control top event, one possible consequence, an inherently safer alternative and an independent barrier if credible. The analysis must explicitly challenge the misleading assumption treating calculated low risk from generic inputs as permission to operate. Which shared utility, maintenance activity or human action could defeat the claimed protections?

Q16 — Evidence-led technical memo (10 marks). Prepare a 250-word memo answering: Produce a five-page decision memo stating what is proven, conditional and still unknown. The recommendation should be make proceed/no-go conditions and remaining uncertainties explicit but you must justify or challenge it. List the real-world records needed: Validated chemistry and equipment data, specialist assessments, management systems and jurisdiction review. Separate a hypothetical mathematical statement from an engineering action and identify the competent approver and a falsifiable follow-up test.

Synthesis: no-go evidence check

Construct an evidence table for assurance, incidents and safety leadership. Include the case A site reports zero recordable injuries but has overdue safety-critical inspections and several bypassed protective devices.. State one safety-critical condition that cannot be closed by the calculations above. Propose a technically reasonable way to obtain missing evidence, an owner and a hold point. Explain why a numerically appealing dashboard could still hide an unacceptable unresolved scenario.