THE INTELLIGENT PETROCHEMICAL COMPANY LEADERSHIP PROGRAMME
The decision behind the chemical.
20-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigeria’s petrochemical and specialty chemical sector
WEEK 3 OF 4 · RELIABILITY, RESOURCES AND ENTERPRISE EXECUTION · SESSIONS 11–15 · WORK AT YOUR OWN PACE
Part Four · Reliability and ResourcesSESSION 11
Equipment Reliability and Maintenance
Why This Matters
An early warning is worth money only if your company can act on it before the consequence arrives. That action may need an inspection, a suitable spare, competent labour, an approved work window and a funding decision. This session teaches Chapter 10's discipline: start from the failure that matters, make the maintenance record worth learning from, define the failure before anyone builds a predictor, and price the whole response chain rather than the alert. You will also work through the book's assumed reliability investment and see why expected value and available cash remain separate decisions. By the end you will know what to demand before a predictive-maintenance proposal reaches your budget.
25-Minute Module
0:00 – 5:00
Start with the failure consequence
- Choose an asset because its condition affects an important decision, not because its data is easy to collect. Establish its failure modes and the consequence of each: a production interruption, a quality issue, a containment concern and a degraded service condition may need different actions and protections.
- The assets follow your business model. A manufacturer may watch process vessels, pumps, compressors and mixers; a specialist supplier may depend on handling, transport and service equipment at a customer's site.
- Define the question. Detecting an abnormal pattern, diagnosing a fault and estimating remaining useful life are different claims. A vendor's broad predictive-maintenance label does not establish all three; require the evaluation to match the proposed use.
- THE DOCTRINE
A model should not declare an asset fit for continued service because no anomaly appeared in the available signal. Absence of a warning is not evidence that every relevant failure mode has been evaluated.
Ask which failure you are trying to avoid before anyone asks which sensor to buy.
5:00 – 11:00
Make the record useful, then define the failure
- A work order should preserve asset identity, observed condition, action taken, parts used and findings. A closure reading pump repaired teaches little; a model trained on vague closure text may predict your organisation's reporting habits rather than your equipment's condition.
- Distinguish planned, corrective and inspection work, and record whether planned work was completed, deferred or changed, with approval and reason. A closed work order is not proof the original scope was done if the field record says otherwise.
- Now the failure label. In plain terms: before a model can predict an event, someone must define which event counts. An unplanned stoppage, a bearing finding, a loss of required capacity and a protective trip are not interchangeable. Technical review resolves ambiguous descriptions.
- Two traps sit in the history. Completed maintenance can prevent the failure the model would otherwise have seen. And an asset that has not failed by the end of the evaluation period is a right-censored record, like a sample still on test when the study closes: it lasted at least that long, which is not proof it would never fail.
- THE DOCTRINE
Build the failure label before building the predictor. A predictive-maintenance dataset needs a definition of the event it predicts, resolved by technical review where work-order descriptions are ambiguous.
- Evaluate chronologically, and keep readings from one developing fault together, or the model may recognise an episode it has effectively seen before.
If your work orders cannot tell you what failed, no model trained on them can either.
11:00 – 16:00
Evaluate the warning window and the spares behind it
- Measure the useful time a warning gives. An alert two days before an interruption may be valuable if the response needs one day, and too late if a suitable spare takes weeks. If a pump inspection can be arranged within a shift but a part takes weeks, those are different decision windows.
- Report events detected, missed events, false alert episodes and actionable warning time. Count unique episodes: twenty alerts for one persistent issue are not twenty detected faults. Where failure history is sparse, say so; condition monitoring with a structured investigation workflow can still help.
- A spare has value when it is suitable, available and usable. A stock ledger may show a quantity that cannot support the planned work, and a superficially similar part is no answer without the relevant assessment.
- THE DOCTRINE
Avoid indiscriminate reductions in maintenance inventory. A rarely used item may protect a major consequence; decide against the actual risk and service requirement.
A warning that arrives after the last useful intervention time is a report, not a warning.
16:00 – 21:00
The assumed reliability investment and the response chain
- Chapter 10's assumed case: a bounded programme addresses a commercially relevant interruption with a defined financial consequence of ₦12 million. The review assumes a probability of 0.20 in the evaluation period under the current arrangement and 0.08 if the programme is implemented effectively. The book is explicit that these are fictional inputs, not measured probabilities.
Line Assumed calculation Result (recomputed) Baseline expected loss 0.20 × ₦12 million ₦2.4 million Proposed expected loss 0.08 × ₦12 million ₦960,000 Expected reduction ₦2.4 million − ₦960,000 ₦1.44 million Expected net benefit after ₦1 million cost ₦1.44 million − ₦1 million ₦440,000 - The result is sensitive to the assumed probabilities and consequence; it does not establish those inputs or prove the programme works. It does not authorise an unsafe condition; protections are evaluated separately, and lost contribution must not be counted twice. If the ₦1 million is paid now and the avoided loss falls later, treasury funds it first: expected value and available cash remain separate decisions.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional batch formulator's mixer shows an abnormal pattern. The maintenance lead asks whether an inspection can be arranged in time, whether the qualified spare is on site and preserved, and whether the campaign plan has a work window.
IN THE MARGIN · THE PARALLEL- Margin. The finance director asks what the defined interruption would cost and when each sum falls in cash. A replay showing the warning would have appeared does not prove the intervention would have prevented the event; until it is proven, report an expected-loss scenario, not a realised saving.
- That is pricing the response chain: separating the ability to detect from the probability that your organisation can inspect, fund and complete the intervention in time.
When a paper shows ₦440,000 of expected benefit, ask which assumption would have to move for it to disappear.
21:00 – 25:00
Close the review with a decision
- The sponsor ends each review with funded work, a justified deferment or a recorded change in programme scope, and the supplier's acceptance test includes the evidence and timing needed to reach it.
- THE DOCTRINE
A health dashboard without funded work, a justified deferment or a recorded change in programme scope is an incomplete reliability intervention.
- Write down the failure mode behind your decision whose consequence you would least like to explain to your board, and its last useful intervention time.
- Today's assignment produces Section 11 — Failure consequence, the failure label, the warning window and the response chain priced.
The useful alert ends in a funded decision, not in a meeting.
Key Terms
Failure label
The agreed definition of the event a predictive-maintenance model predicts, with asset, time, finding and consequence.
Right-censored record
An observation whose later event time is unknown because the observation period ended first; not proof of permanent health.
Actionable warning time
The useful time between an alert and the deadline for the intended response, given inspection, spare and work-window lead times.
Expected loss
An assumed probability multiplied by a defined financial consequence. A scenario for comparison, not a realised saving.
Self-Check — Before You Submit
1. In the book's assumed reliability investment, what is the expected net financial benefit for the period?
- A. ₦1.44 million
- B. ₦440,000
- C. ₦2.4 million
- D. ₦960,000
B. ₦440,000 — ₦2.4 million less ₦960,000 gives a ₦1.44 million reduction; the ₦1 million programme cost leaves ₦440,000. ₦1.44 million is tempting but ignores the cost.
2. A vendor's replay shows its model would have alerted before three past interruptions. Your operations director wants to book them as avoided. What do you do?
- A. Book the savings; the replay proves it
- B. Reject the model; replays have no value
- C. Treat the replay as evidence of detection only, require diagnosis, work findings and a comparison before claiming prevention, and report an expected-loss scenario meanwhile
- D. Ask for a longer replay to enlarge the savings
C. Treat the replay as evidence of detection only, require diagnosis, work findings and a comparison before claiming prevention, and report an expected-loss scenario meanwhile — A replay shows a warning would have appeared, not that the intervention would have prevented the event. Rejecting the model discards real detection evidence.
3. Why are twenty alerts for one persistent issue a problem for evaluation?
- A. They are not twenty detected faults; count unique episodes and the hours to resolve them
- B. They show welcome sensitivity
- C. They prove the asset must be replaced
- D. They make every record right-censored
A. They are not twenty detected faults; count unique episodes and the hours to resolve them — Alert burden matters: a team chasing many false alarms can become less responsive to the useful alert, so the book asks for unique episodes and resolution hours.
Today’s Assignment
Write Section 11 of your plan: failure consequence, failure label, warning window and response chain for one equipment decision behind your company's recurring decision.
- Name one asset your company's decision depends on (with no plant, use the handling, transport or customer-site equipment you rely on). State its failure modes and the consequence of each.
- Check three recent work orders for it and say honestly what your records get wrong.
- Write the failure label, and state whether you need detection, diagnosis or remaining-life estimation.
- Set out the warning window: the response required, the lead times of inspection, spare and work window, and the last useful intervention time.
- Price the response chain in expected-loss form with your own inputs labelled as assumptions, state the cash timing, and close with the sponsor's decision.
YOUR 90-DAY PLAN
Section 11 — Failure consequence, the failure label, the warning window and the response chain priced.
How this is marked — the rubric
| Names an asset chosen for decision relevance, with failure modes and distinct consequences | 25 |
| Defines a failure label with asset, time, finding and consequence and assesses the record honestly | 25 |
| States the warning window against real lead times, including the last useful intervention time | 25 |
| Prices the response chain with labelled assumptions, separates expected value from cash and ends with a decision | 25 |
| Total · pass mark 70 | 100 |
Submitting session 11
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 11, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Failure-consequence selectionFailure label definitionWarning-window evaluationExpected-loss reasoningResponse-chain pricing
Part Four · Reliability and ResourcesSESSION 12
Utilities, Inventory and Material Availability
Why This Matters
Your plant, warehouse or service operation depends on resources that rarely appear in a board paper until they fail: power, water, cooling, compressed air, packaging and suitable storage. Chapter 11 asks you to make those dependencies explicit, to measure utility use on a denominator that means something, and to make inventory status physical rather than a single reassuring total. An optimisation proposal that ignores its utility and material constraints can recommend an attractive operating point your company cannot reach. This session gives you the questions that expose that gap, and the discipline to plan how your business keeps control of its product when a system or utility is unavailable.
25-Minute Module
0:00 – 5:00
Map the critical dependencies
- For each resource behind your decision, record the resource, the operating stage it supports and the consequence of its loss. The technical team establishes which interruptions need a controlled response and which merely reduce output; recovery follows the site's approved arrangements.
- The requirement follows the product, not a generic checklist. A batch may need water of a stated quality, available cooling and suitable packaging; a supplier may need compatible storage and safe handling equipment.
- THE DOCTRINE
Installed capacity is different from usable capacity under current constraints. A standby unit is not necessarily ready if its fuel, maintenance or switching arrangement is unresolved.
- This extends Session 6's binding constraint to supporting resources. An optimisation that ignores them can recommend an attractive but unavailable operating point.
Ask to see the standby unit's fuel, maintenance and switching status, not its nameplate.
5:00 – 11:00
Measure utility intensity on a useful basis
- Energy per tonne needs a stated denominator (gross, accepted or delivered product) and a stated boundary of energy counted. Mixed boundaries can produce a misleading improvement.
- The plain idea first. Some consumption continues whatever the output, as a delivery round costs about the same whether the truck is full or half-empty. When output falls, that fixed load spreads over fewer tonnes and energy per tonne rises with no loss of equipment efficiency. Normalising compares actual use with what an approved baseline predicts at actual output.
Chapter 11 assumed illustration Calculation Result (recomputed) 1,000 accepted tonnes 100,000 kWh ÷ 1,000 t 100 kWh per tonne 800 accepted tonnes, same electricity 100,000 kWh ÷ 800 t 125 kWh per tonne Baseline at 1,000 t 20,000 + (80 × 1,000) kWh 100,000 kWh Baseline at 800 t 20,000 + (80 × 800) kWh 84,000 kWh Actual at 800 t 90,000 kWh ÷ 800 t 112.5 kWh per tonne Deviation from baseline 90,000 − 84,000 kWh 6,000 kWh - Comparing 100 with 112.5 kWh per tonne would confound the fixed-load effect with the additional 6,000 kWh deviation. These are the book's teaching models, not plant benchmarks; a real baseline needs suitable measurements and evidence that its relationship still holds. A model-derived estimate of shared metering is labelled as an estimate.
- The US Department of Energy's 2015 chemical-manufacturing bandwidth study distinguishes current typical, state-of-the-art, practical-minimum and thermodynamic-minimum energy estimates, calls them macro-scale and excludes the cost of achieving savings. A sector benchmark cannot become your site's funded savings target.
Whenever energy per tonne improves, ask what happened to accepted tonnes.
11:00 – 16:00
Compare the full option against the required service
- Compare options on one service boundary. Power involves energy cost, capital, maintenance, fuel availability, reliability and switching; water involves treatment, quality assurance, storage and waste. A single unit price cannot settle it.
- Use your site's actual contract, bills, generation records and fuel evidence, not a generic tariff, and state any foreign-currency assumption. The book assumes no universal tariff, outage frequency or eligibility for any supply arrangement; an option is compared only where the technical and regulatory owners have established its availability and requirements.
- THE DOCTRINE
A lower energy price can be unattractive if the arrangement cannot support the qualified process. An expensive backup can protect a defined commercial consequence without being the preferred source for all hours.
- Preserve measured interruption time and its effect on accepted output. Not every unavailable hour loses full-rate production, and the full revenue of an affected campaign and its lost contribution are not two benefits.
A tariff comparison that ignores power quality and interruption response is a price comparison, not a supply decision.
16:00 – 21:00
Make inventory status physical
- Separate available, held, committed, in-transit and unusable stock, linked to lot, location, condition, relevant date and commitment. A single total can hide the shortage that matters. Shelf-life rules are product-specific; never one expiry policy for every chemical.
- THE DOCTRINE
An AI forecast cannot compensate for a ledger that records receipt twice or fails to record material consumption. Repair the transaction control first.
- Use demand and service together: separate forecasts from firm commitments, and keep the customer's requested date, agreed date and acceptance conditions. Replenishment options include delay, reduction, a customer deposit or a changed supply arrangement.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional batch formulator's ledger shows enough raw material for next week's campaign. A physical check finds part held pending a test and part committed elsewhere. The usable quantity sets the plan.
IN THE MARGIN · THE PARALLEL- Margin. The commercial director sees the same total as idle cash and proposes cutting replenishment. By status, held and committed stock is not idle, and a cut could force an emergency purchase. Inventory policy is a commercial decision supported by technical suitability and treasury capacity.
- A forecast that lowers cash exposure while creating an unfunded emergency purchase has displaced the problem into another period.
Ask for stock by status before you ask for stock by value.
21:00 – 25:00
Plan degraded operation, then write
- If a digital application is down, staff must still establish material identity, release status and current commitments through the approved fallback. Test it with the operating team.
- THE DOCTRINE
A business should not lose basic control of its product because an analytical service cannot connect.
- Write down the three supporting resources whose loss would stop your decision, and whether each standby is actually ready.
- Today's assignment produces Section 12 — Critical dependencies, physical inventory status and degraded operation planned.
The fallback should be rehearsed before the outage, not improvised during it.
Key Terms
Usable capacity
Capacity actually available under current constraints, including fuel, maintenance and switching, as distinct from installed capacity.
Normalised energy baseline
An approved relationship predicting energy use at actual output, separating the fixed-load effect from genuine deviation.
Physical inventory status
Stock split into available, held, committed, in-transit and unusable, linked to lot, location, condition, date and commitment.
Degraded operation
The approved way a process continues when a system or utility fails, keeping control of identity, release status and commitments.
Self-Check — Before You Submit
1. In the book's assumed normalised baseline, actual use at 800 accepted tonnes is 90,000 kWh. What is the deviation?
- A. 10,000 kWh
- B. 6,000 kWh
- C. 12.5 kWh per tonne
- D. 16,000 kWh
B. 6,000 kWh — The baseline predicts 20,000 + 80 × 800 = 84,000 kWh, so the deviation is 6,000 kWh. Comparing 100 with 112.5 kWh per tonne confounds the fixed load with the deviation.
2. A supply proposal offers a lower energy price. Which question does the book require before you prefer it?
- A. Is the price below a national average tariff
- B. Can a sector benchmark's practical minimum be reached
- C. Does the vendor offer a dashboard
- D. Can the arrangement serve the load, power quality, interruption response, maintenance, fuel, capital and cash timing the qualified process needs
D. Can the arrangement serve the load, power quality, interruption response, maintenance, fuel, capital and cash timing the qualified process needs — Options are compared against the required service on one boundary. The book assumes no universal tariff, and a sector benchmark cannot be a site's funded target.
3. Your team proposes an AI replenishment forecast because shortages keep surprising the plant. A physical check finds receipts recorded twice. What comes first?
- A. Repair the transaction control and reconcile against physical checks
- B. Deploy the forecast with a wider safety margin
- C. Raise stock of every material
- D. Train the model on the duplicate pattern
A. Repair the transaction control and reconcile against physical checks — A forecast cannot compensate for a ledger that records receipt twice. A wider margin hides the defect at a cash cost; training on the error teaches the error.
Today’s Assignment
Write Section 12 of your plan: the critical dependencies behind your company's decision, its physical inventory status and its degraded operation.
- Map the utilities, materials, packaging, storage and handling your company's decision depends on: stage supported, consequence of loss, and installed versus usable capacity.
- State one utility or resource measure your business uses, its denominator and boundary, and what it hides if not normalised.
- Split the relevant stock by physical status, and say honestly whether your ledger can do this today and where physical checks disagree.
- Compare two supply or replenishment options against the required service, with any currency or funding assumption labelled.
- Describe the approved fallback for identity, release status and commitments, and when it was last tested.
YOUR 90-DAY PLAN
Section 12 — Critical dependencies, physical inventory status and degraded operation planned.
How this is marked — the rubric
| Maps each dependency with stage, consequence of loss and usable versus installed capacity | 25 |
| States a measure with explicit denominator and boundary and what normalisation would change | 20 |
| Splits stock by physical status and honestly names ledger or transaction-control weaknesses | 30 |
| Compares options against required service and describes a tested or testable fallback | 25 |
| Total · pass mark 70 | 100 |
Submitting session 12
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 12, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Dependency mappingDenominator and boundary disciplineEnergy normalisationPhysical inventory controlDegraded-operation planning
Part Four · Reliability and ResourcesSESSION 13
Safety, Environmental Evidence and Controlled Change
Why This Matters
A chemical business must know what can go wrong, which protections are required and whether those protections are functioning. AI can organise that evidence and surface exceptions. It cannot establish that your physical system is safe or that your company has met every applicable obligation. This session teaches Chapter 12's position for the executive: keep hazard knowledge specific, make every change visible before execution, including software and model changes, build an obligation register for the activities you actually run, and look directly at the condition of protections rather than at a reassuring score. An unresolved protection can stop a campaign as surely as a missing raw material.
25-Minute Module
0:00 – 5:00
Keep hazard knowledge specific
- Hazards depend on the actual material, process, equipment and conditions. A familiar product or a history without reported incidents does not establish that a proposed change is acceptable.
- The US Chemical Safety Board's Improving Reactive Hazard Management study and recommendations (2002, with associated material released in 2003) emphasise identifying and evaluating reactive hazards, managing change, appropriate design, procedures and training. The book treats it as an international learning source, not Nigerian legislation or evidence that any local process is safe.
- An assistant may retrieve approved hazard information with its source and scope. It should not complete an assessment by filling gaps with plausible generic text.
- THE DOCTRINE
A missing product-specific document should remain a missing document requiring action, not a gap an assistant fills with plausible generic text.
When you hear that the company has made this product for years, ask whether it has made it under these conditions.
5:00 – 10:00
Learn from a documented reactive incident
- The US Chemical Safety Board's investigation of the December 2007 reactive explosion in Jacksonville, Florida (its Runaway Reaction report of September 2009) identified a runaway reaction and a likely loss of sufficient cooling, and addressed inadequate recognition of the reactive hazards. The book uses it as an international learning case, not evidence about any Nigerian company, and omits the business name.
- The lesson: prior production history and normal operating observations cannot establish the safety of an unfamiliar adverse condition, and a model trained on ordinary production may have little evidence about it.
- THE DOCTRINE
A confident normal-condition prediction cannot be used to dismiss an adverse case that the engineering assessment must consider.
- Ask how each required protection works and what it depends on. A warning, a control action and an independent protective function are not separate safeguards because they appear on different screens; shared measurements, power, software or communication create common dependencies. The book assigns no reliability credit or safety-integrity level to an AI model.
Count safeguards by what they depend on, not by how many screens show them.
10:00 – 16:00
Make change visible before execution
- A change can involve material source, formulation, equipment, software, operating conditions or a customer application. The review establishes consequence, approvals, revised documents, training and follow-up evidence. The responsible technical authority decides whether it is routine under an approved procedure or needs a new assessment.
- Software counts when it alters information or actions used in operation: a new model version, a changed data mapping or an expanded execution permission. Record purpose, evaluation and approval.
AI-related change (Chapter 12) What the change review sees A planning recommendation now writes a setpoint Scope has expanded beyond planning A quality estimate is copied into a certificate Scope has expanded beyond early warning A model update changes the meaning of an operator's score The effect needs evaluation before release - Temporary changes keep their expiry or review condition. An application can flag the date; an accountable owner decides the disposition.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional specialty-chemicals producer moves a raw material to a second source with a matching specification. The technical authority asks whether hazard information, the approved procedure and training still apply, and classifies the change before the first batch.
IN THE MARGIN · THE PARALLEL- Margin. Procurement presented the switch as a saving. Until the review closes, the saving is contingent, because the campaign may not yet be feasible. One loop seen twice: the technical classification sets what the margin case may assume.
A temporary change without a review date is a permanent change nobody approved.
16:00 – 21:00
Build the obligation register and preserve the evidence
- List the activities your company actually runs (chemical manufacturing, petroleum-sector activity, importing, storage, waste handling, customer service) and the bodies, permits, standards and contracts relevant to each. Do not assume one regulator or licence covers every activity within a group.
- One instrument the book cites: NESREA's National Environmental (Chemicals, Pharmaceuticals, Soap and Detergent Manufacturing Industries) Regulations, S.I. No. 36 of 2009, covering pollution prevention and controls for that sector. NESREA's August 2025 notice reported a stakeholder review of these and other regulations. A review notice alone does not demonstrate enactment of a replacement text; your compliance owner establishes the operative instruments.
Register field What it records Primary source The instrument, permit, standard or contract Relevant provision The specific part that may apply Applicability assessment The competent compliance and legal judgement Responsible person The named owner Evidence required What shows the duty is met - THE DOCTRINE
A model can assist with retrieval and change detection. The approved applicability judgement belongs to the competent compliance and legal review.
- Environmental values trace to their measurement or approved estimation method. In the book's hypothetical, a monitor is unavailable for a reportable period; filling the gap with an average would create evidence the company does not possess. Establish the permitted treatment with the responsible specialist.
A neat narrative should never conceal an unresolved measurement discrepancy.
21:00 – 25:00
Track protections, prepare for response, then write
- The executive review shows overdue or impaired protections and whether a planned campaign relies on one. Process safety cannot be reduced to personal injury statistics or training attendance, which may be useful but do not establish the condition of the process and its controls, and closure follows verified work, not a changed colour.
- THE DOCTRINE
Avoid averaging a serious concern into a reassuring score. If one critical condition is unresolved, the review should show it directly.
- A generative assistant is not the sole source of emergency instructions. Exercises record what was learned and what was unavailable, and incident learning separates facts, findings and unconfirmed explanations.
- Pop Quiz 3 opens now: five questions on Sessions 11 to 13, closed book. Today's assignment produces Section 13 — Specific hazard knowledge, visible change, the obligation register and the condition of protections.
An attendance sheet shows who came to the exercise, not whether the company can respond.
Key Terms
Management of change
The review of consequence, approvals, documents, training and follow-up evidence for a change in material, formulation, equipment, software, conditions or application, before execution.
Obligation register
An activity-specific record of primary source, relevant provision, applicability assessment, responsible person and evidence required.
Common dependency
A shared measurement, power supply, software or communication link on which several apparent safeguards depend, so they should not be counted as separate safeguards.
Applicability assessment
The competent compliance and legal judgement of whether an instrument applies to a specific activity and facility.
Self-Check — Before You Submit
1. What does NESREA's August 2025 notice establish, as the book reads it?
- A. That S.I. No. 36 of 2009 has been repealed
- B. That a stakeholder review was reported, not that a replacement text has been enacted
- C. That one NESREA licence covers all activities
- D. That generated summaries may be submitted unreviewed
B. That a stakeholder review was reported, not that a replacement text has been enacted — A review notice alone does not demonstrate enactment of a replacement. Reading it as a repeal could lead a compliance owner to drop an instrument that may still be operative.
2. An advisory planning model would, after a vendor update described as minor, write setpoints directly. What should you require?
- A. Accept it as administrative
- B. Accept it once the vendor certifies the model
- C. Treat it as a change: scope has expanded beyond planning, so record purpose, evaluation and technical-authority approval before release
- D. Accept it on night shifts only
C. Treat it as a change: scope has expanded beyond planning, so record purpose, evaluation and technical-authority approval before release — The book says a planning recommendation that writes a setpoint has expanded beyond planning. The book gives that classification to the company's responsible technical authority, not to the vendor.
3. Why does the book use the Jacksonville runaway-reaction investigation?
- A. To show that prior production history cannot establish the safety of an unfamiliar adverse condition
- B. As evidence about Nigerian operating practice
- C. To assign a safety-integrity level to AI models
- D. To show injury statistics measure process safety
A. To show that prior production history cannot establish the safety of an unfamiliar adverse condition — It is an international learning case about reactive hazards and cooling. The book assigns no safety-integrity level to AI and says injury statistics do not establish the condition of the process and its controls.
Today’s Assignment
Write Section 13 of your plan: specific hazard knowledge, visible change, an obligation register and the condition of protections for your company's decision.
- State whether the hazard knowledge your decision touches is specific to your actual material, process and conditions, and name any missing product-specific document.
- Take the last three changes in your company affecting the decision, including any model or data-mapping change, and say honestly whether each was reviewed before execution.
- Draft five obligation-register rows for the activities your company runs, with all five fields. Mark unestablished applicability as open; do not guess.
- Show how your executive review displays impaired protections, and how your emergency arrangement works when digital services are down.
YOUR 90-DAY PLAN
Section 13 — Specific hazard knowledge, visible change, the obligation register and the condition of protections.
How this is marked — the rubric
| Keeps hazard knowledge specific and names missing product-specific documents | 20 |
| Assesses recent changes honestly, including software and model changes | 25 |
| Drafts register rows with all five fields, leaving unestablished applicability open | 30 |
| Makes impaired protections and emergency response visible without averaging into a score | 25 |
| Total · pass mark 70 | 100 |
Submitting session 13
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 13, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Pop Quiz 3 — Reliability, Utilities and Safety opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Pop Quiz 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Hazard specificityManagement of changeObligation registerProtection condition reviewResponse readiness
Part Five · Enterprise ExecutionSESSION 14
Demand, Procurement and Working Capital
Why This Matters
A profitable order can still strain your company's finances. You may pay a supplier before shipment, hold material before the customer needs it, and collect only after acceptance. Chapter 13 asks leadership to manage the full commitment, not the invoice margin alone. This session works through the book's Worked Case Four, a hypothetical imported order showing ₦80 million of contribution that the business cannot fund within its assumed ₦200 million facility on the original terms. You will build the cash calendar, find the funding trough, price financing on actual balances, test a customer deposit, a currency movement and a collection delay, and decide what AI can and cannot improve.
25-Minute Module
0:00 – 5:00
Know what kind of demand you hold
- Separate enquiries, commercial forecasts, firm orders, scheduled call-offs and historical consumption; a model treating them as equivalent can stock for orders that never come. Record why orders were lost or delayed, and never infer a customer commitment from a friendly discussion.
- Compare replenishment alternatives on landed cost, usable-material date, stock exposure and funding, with technical suitability as a prerequisite, and record payment terms and the currency of the obligation.
- THE DOCTRINE
Purchasing authority applies to the substance of the commitment, not merely the format of the transaction. A system must not split an order to evade an approved cash limit.
- A one-time inventory reduction is not recurring profit. Report cash release, write-down, carrying cost and service consequences separately.
Ask of every demand number whether a customer has committed to it.
5:00 – 11:00
Build the cash calendar and find the trough
- The idea first. List each payment and receipt by date, keep a running total and find its lowest point. As with a tank drawn down before the next delivery, the lowest level matters, not the month-end level. That most negative cumulative balance is the funding trough.
- Worked Case Four is hypothetical: its rates, prices, terms and financing costs are invented, not current Nigerian rates or bank offers, and the import is assumed already qualified. A specialist supplier holds a ₦360 million naira order. Product costs US$150,000, or ₦240 million at an assumed ₦1,600 per dollar; landed expenditure ₦30 million (no duty rate implied); delivery and service ₦10 million. Contribution before financing: ₦360 million − ₦280 million = ₦80 million.
- Terms: customer deposit 20 percent (₦72 million); supplier 30 percent at day 0 (₦72 million) and 70 percent at day 30 (₦168 million); customer balance ₦288 million at day 120 after assumed acceptance. No other opening cash.
Day Assumed cash event (₦ million) Cumulative (₦ million) 0 Deposit in +72; supplier deposit −72 0 30 Supplier balance −168 −168 45 Landed expenditure −30 −198 60 Delivery and service −10 −208 120 Customer balance +288 +80 - THE DOCTRINE
The funding trough is ₦208 million before financing. If the approved available facility is ₦200 million, the order is not funded: obtain approved funds, change the terms, or decline or defer.
An attractive contribution tells you the order is worth having, not that you can afford to carry it.
11:00 – 16:00
Finance the actual balances, then test a deposit
- At an assumed 2 percent simple interest per 30-day month, with no compounding, fees or principal repayment before collection, each increment is financed until day 120: ₦168 million × 0.02 × 3 = ₦10.08 million; ₦30 million × 0.02 × 2.5 = ₦1.50 million; ₦10 million × 0.02 × 2 = ₦0.40 million. Total ₦11.98 million; contribution after financing ₦68.02 million.
- The trough excludes the cash needed to pay interest. A facility equal to the pre-financing trough is not thereby adequate.
- Alternative: a 30 percent deposit (₦108 million), balance ₦252 million at day 120. After the supplier deposit, ₦36 million remains, so borrowing is ₦132 million at day 30, ₦162 million at day 45 and ₦172 million at day 60. Financing: ₦7.92 million + ₦1.50 million + ₦0.40 million = ₦9.82 million. Contribution after financing: ₦70.18 million, ₦2.16 million better.
- THE DOCTRINE
The deposit is not additional revenue. Its value comes from earlier cash and lower borrowing; a dashboard that adds it to revenue counts the same customer order twice.
- It appears to fit the facility, subject to interest timing and actual terms, but stays contingent until the deposit arrives. A draft amendment is not available cash.
Interest follows the balance and the days; price it on both.
16:00 – 21:00
Test currency exposure and collection delay
- The unpaid supplier balance is US$105,000. At ₦1,800 instead of ₦1,600 it costs ₦189 million, ₦21 million more, and contribution before financing falls from ₦80 million to ₦59 million.
Book's sensitivity (recomputed) 20% deposit 30% deposit Base trough ₦208m ₦172m Base financing; contribution after ₦11.98m; ₦68.02m ₦9.82m; ₦70.18m ₦1,800 rate: trough ₦229m ₦193m ₦1,800 rate: financing; contribution after ₦13.24m; ₦45.76m ₦11.08m; ₦47.92m Collection 30 days late: extra financing; contribution after ₦4.16m; ₦63.86m ₦3.44m; ₦66.74m - These are sensitivities, not forecasts. The Central Bank of Nigeria's published account of the January 2025 Nigeria FX Code launch describes standards for conduct and governance in the foreign exchange market; it establishes no settlement rate, guaranteed availability or approved hedging product for this order. The full Code text was inaccessible at the book's source check, so no detailed duties are inferred; treasury confirms the executable arrangement with its authorised counterparties.
- Investigate why collection moves. A missing acceptance certificate, a product issue and a payment delay need different responses, so keep delivery, acceptance, invoice, due date and receipt as separate events. A reminder will not resolve a technical dispute.
- IN THE PLANT · THE WORKED CASE
- Plant. In the case, collection depends on completed acceptance, which may require evidence the company has not yet prepared; if it is not ready, the day-120 receipt moves whatever the sales forecast says.
IN THE MARGIN · THE PARALLEL- Margin. Treasury sees the same event as ₦4.16 million of extra financing. A combined currency and collection downside needs its own cash calendar, not two sensitivities added mechanically.
Every date in the cash calendar should have a source and an owner who can say why it moved.
21:00 – 25:00
Decide, define what AI can improve, then write
- The book's hypothetical decision: withhold the unfunded original commitment. The larger-deposit alternative proceeds only once its commercial, technical and funding conditions are met, with the downside calculations kept in the approval paper.
- A readiness assistant can reconcile order, supplier milestones, documents, release, acceptance and funding calendar. A forecast is judged against the event the decision needs: port arrival is not usable availability, and due date is not collection.
- THE DOCTRINE
Neither capability changes a contractual term or confirms a bank balance by inference. The negotiated deposit's value should not automatically be credited to AI.
- Today's assignment produces Section 14 — The cash calendar, the funding trough and the downside tests.
If a model helps the team intervene earlier, record the action it changed; that is the benefit it has earned.
Key Terms
Funding trough
The largest cumulative cash deficit within a stated boundary and timing model; financing charges are added separately.
Cash calendar
Dated payments and receipts, committed or proposed, expected or confirmed, each with its source.
Financing on actual balances
Interest on each drawn amount for the days it is actually outstanding.
Acceptance evidence
The delivery, technical acceptance and documents a customer needs before paying, kept separate from invoice, due date and receipt.
Self-Check — Before You Submit
1. In Worked Case Four on the original terms, what is the funding trough before financing?
- A. ₦168 million
- B. ₦200 million
- C. ₦208 million
- D. ₦280 million
C. ₦208 million — The running balance reaches −₦168 million, −₦198 million and then −₦208 million at day 60. ₦280 million is total cost, but the deposit offsets part of it.
2. Your sales director wants to report the larger deposit as additional revenue, since it improved the case by ₦2.16 million. What do you say?
- A. Agree; it is cash received
- B. No; it is earlier cash and lower borrowing, and adding it would count the same order twice
- C. Agree for the extra ten points only
- D. No; credit it to the readiness assistant
B. No; it is earlier cash and lower borrowing, and adding it would count the same order twice — Revenue stays ₦360 million; the gain is lower financing. Crediting it to AI is the second trap, which the book names explicitly.
3. Why does the ₦172 million trough not prove the ₦200 million facility adequate for the deposit alternative?
- A. The trough excludes interest cash, and the deposit is contingent until received
- B. Facilities must exceed troughs by a fixed percentage
- C. Troughs use monthly net totals
- D. The facility must also cover revenue
A. The trough excludes interest cash, and the deposit is contingent until received — The book says the alternative appears to fit, subject to interest timing and actual terms, and a draft amendment is not available cash. No fixed-percentage rule exists in the case.
Today’s Assignment
Write Section 14 of your plan: the cash calendar for one commitment behind your company's decision, its funding trough and its downside tests.
- Choose one order or commitment linked to your company's decision; if your business imports nothing, use your largest supplier payment due before a customer receipt.
- Build its cash calendar, marking each item committed, proposed, expected or confirmed, with its source.
- Calculate the funding trough and compare it with the facility actually approved, stating your interest assumption and timing.
- Run at least two labelled downside tests, such as currency and a 30-day collection delay.
- Say honestly where your company manages invoice margin rather than the full commitment, and what AI can and cannot improve.
YOUR 90-DAY PLAN
Section 14 — The cash calendar, the funding trough and the downside tests.
How this is marked — the rubric
| Builds a dated cash calendar separating committed, proposed, expected and confirmed items, each sourced | 30 |
| Calculates the trough correctly and tests it against the approved facility, including interest timing | 25 |
| Runs at least two labelled downside tests and explains their effect | 25 |
| States what AI can and cannot improve without crediting negotiated terms to the model | 20 |
| Total · pass mark 70 | 100 |
Submitting session 14
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 14, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.Capabilities Covered
Demand classificationCash-calendar constructionFunding-trough analysisFinancing on actual balancesCurrency and collection stress tests
Part Five · Enterprise ExecutionSESSION 15
Data Foundations, Traceability and Cybersecurity
Why This Matters
Every earlier session has depended on records: lot identities, release statuses, work orders, cash dates. Chapter 14 treats that operating record as an asset of your company and asks you to protect it through precise definitions, controlled access and recovery you have actually tested. Buying a platform before defining the records can leave you with more interfaces and the same uncertainty. This session teaches stable identities, the data contract for your first decision, the separation of reading information from controlling the plant or a transaction, auditable retrieval, the protection of personal and commercial information, and supplier exit. Practice Exam 3 then closes Week 3.
25-Minute Module
0:00 – 5:00
Use stable identities
- Create controlled identities for products, lots, batches, equipment, customers and relevant documents. A supplier trade name may need an alias, not a second uncontrolled product record; a revised specification stays identifiable as a revision.
- Trace material from receipt into a batch and from product to dispatch. Where repacking, mixing or transfer changes the physical arrangement, preserve the genealogy rather than renaming the stock.
- Received, inspected, held, released, allocated and dispatched are different events; record who can change each status and on what evidence.
- THE DOCTRINE
AI should read status events from the source rather than infer them from a narrative description.
If two systems call the same lot by different names, you have two lots until someone proves otherwise.
5:00 – 11:00
Write the data contract for the first decision
- Session 1 introduced the bounded data contract; now write one. Specify fields, units, allowable values, update frequency and owner, which items are mandatory, and how missing or stale data affects permission. Where the contract is not met, the model should abstain.
- Scope follows the decision. The book's lot-readiness contract might include product and lot identity, current specification, receipt status, required documentation, test status and disposition authority; customer credit belongs in an order-readiness application.
- Test source-to-output reconciliation: if the assistant reports ten held lots, staff should find those ten in the controlled source. An unexplained difference is a defect.
- Preserve the record as known at the decision time. When something happened and when you learned of it differ: a sample collected Monday may be reported Tuesday. A replay on today's corrected data may grant the model knowledge the planner did not possess.
- THE DOCTRINE
For consequential fields, the data contract should identify event time, availability time, source revision, status and owner. A change from kilograms to tonnes should not be discovered through erroneous outputs.
Ask of any replay whose knowledge it uses, and as of when.
11:00 – 16:00
Keep information access separate from control
- The plain idea first. Reading a gauge and turning a valve are different permissions. An application can see plant information without any ability to write to the control system: information access is separate from control. Begin with the least access the purpose needs.
- The book cites NIST SP 800-82 Revision 3, Guide to Operational Technology (OT) Security, guidance addressing performance, reliability and safety needs, published in September 2023. At the book's source check NIST had identified an initial public draft of Revision 4; a draft is not a final publication.
- The same holds for transactions. Read through a defined interface; keep the authoritative system for stock, release and commitments. Structured controls check a drafted order's supplier, product, unit, quantity, currency and price, and a retried request must not create a duplicate commitment.
- IN THE PLANT · THE WORKED CASE
- Plant. A fictional continuous producer links a reporting assistant to process data through a cloud service. The security owner and technical team confirm it only reads and cannot become an uncontrolled path to operational execution.
IN THE MARGIN · THE PARALLEL- Margin. The same producer lets an assistant draft purchase orders. Approval binds to the specific proposal: change quantity, currency or supplier and the permission lapses. A conversational confirmation is weak evidence for a consequential action.
- THE DOCTRINE
Connecting an application to plant information does not require permission to write to the plant's control system. Begin with the least access needed for the analytical purpose.
Every new permission is a change; review it as one.
16:00 – 21:00
Control the knowledge collection and protect information
- Retrieval-augmented generation supplies source passages to a generative model before it answers. A fluent answer built on the wrong product's document remains wrong, so evaluate retrieval and answer separately, including missing, conflicting and superseded sources.
- Treat retrieved content as evidence, never instruction: a supplier document can contain incorrect or malicious instructions. For proprietary formulations, restrict access before retrieval.
- The Nigeria Data Protection Act, 2023 establishes the country's personal data protection framework and the Nigeria Data Protection Commission. Personal data in staff, supplier and customer records needs its own assessment by your privacy and legal owners. The complete Act was not accessible at the book's source check, so the book states no filing deadlines or detailed legal tests, and choosing a hosting region alone does not settle every obligation. Formulation data needs contractual and security controls even where it is not personal data.
- Restrict credentials by role and log meaningful access. A supplier's support account needs a defined purpose, approval and termination.
A source-bounded assistant is tested on conflicts and gaps, not only on easy questions.
21:00 – 25:00
Test recovery, register consumers, close the week
- A backup never restored is an unproven arrangement. The supplier agreement should cover data export, documentation, configuration access and termination; a model may be replaceable while an undocumented workflow is not.
- Register the consumers of every model output. Sculley and colleagues (Hidden Technical Debt in Machine Learning Systems, 2015) describe debt from data dependencies, feedback and undeclared consumers; the book uses this to support reviewing the whole deployed arrangement, not as a Nigerian-sector cost estimate. A planning score later copied into a release or credit workflow has changed its consequence.
- Practice Exam 3 opens now: fifteen questions on Sessions 11 to 15, closing Week 3.
- Today's assignment produces Section 15 — Stable identities, the data contract, access kept from control and recovery tested.
Design the fallback before the outage; an improvised one risks duplicate transactions and conflicting release statuses.
Key Terms
Data contract
An agreement on fields, meanings, units, quality, timing, owners and missing-data treatment for one decision.
Event time and availability time
When something occurred and when its information reached the decision. Historical evaluation keeps both.
Information access versus control
Permission to read information is separate from permission to change the plant, a status or a commitment.
Retrieval-augmented generation
Supplying retrieved sources to a generative model; correct selection and faithful use still need evaluation.
Self-Check — Before You Submit
1. What does the book say about NIST SP 800-82?
- A. Revision 4 is final and replaces Revision 3
- B. Revision 3 was published in September 2023; an initial public draft of Revision 4 is a draft, not a final publication
- C. It is Nigerian OT legislation
- D. It assigns safety-integrity levels to AI
B. Revision 3 was published in September 2023; an initial public draft of Revision 4 is a draft, not a final publication — The book keeps the edition distinction visible. Treating the Revision 4 draft as final is the tempting error, and the guidance is not Nigerian law.
2. An assistant reports ten held lots; the controlled source shows eight. The vendor calls it a minor narrative difference. What do you do?
- A. Accept the assistant's figure
- B. Ask the assistant to explain itself
- C. Treat the unexplained difference as a defect and investigate by reconciliation
- D. Accept the warehouse figure and move on
C. Treat the unexplained difference as a defect and investigate by reconciliation — The book calls an unexplained count difference a defect. Asking the assistant to explain substitutes fluent text for reconciliation; moving on leaves the defect in place.
3. Why record availability time as well as event time?
- A. So a replay uses only information available at decision time
- B. Because regulators require two timestamps
- C. Because availability time is more accurate
- D. So corrections can overwrite the original event
A. So a replay uses only information available at decision time — A sample reported Tuesday was not available to Monday's decision. Without availability time, a replay overstates what the model could have known.
Today’s Assignment
Write Section 15 of your plan: stable identities, the first data contract, the access boundary and tested recovery for your company's decision.
- List the identities your company's decision depends on, and say honestly where aliases, renamed stock or duplicate records exist today.
- Write the data contract: fields, units, owner, mandatory items, event and availability time, and when the application must abstain.
- Draw the access boundary: what any application may read, draft and never execute, and who approves a change to it.
- State how personal and formulation information is protected, when your business last restored from backup, and what supplier agreements say about exit. With no AI in use yet, answer for the systems the decision already relies on.
YOUR 90-DAY PLAN
Section 15 — Stable identities, the data contract, access kept from control and recovery tested.
How this is marked — the rubric
| Identifies controlled identities and honestly names aliases or duplicates | 20 |
| Writes a data contract with owner, event and availability time and abstention conditions | 30 |
| Separates reading from drafting and execution, with a named approver | 25 |
| Covers information protection, tested recovery and supplier exit | 25 |
| Total · pass mark 70 | 100 |
Submitting session 15
Write your answer in Word or Google Docs. The first line must read
Write your answer in Word or Google Docs. The first line must read
Session: 15, or the submission is refused. Upload the Word file (.docx), or switch to the Google Doc tab and paste your document’s share link.Marked against the rubric above and returned as a written letter, saved to your progress page. Do not submit until you have passed the self-check above.ASSESSMENT NOW OPEN
Practice Exam 3 — Reliability, Resources and Enterprise Execution opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Practice Exam 3 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Stable identity controlData contract designAccess separated from controlAuditable retrievalRecovery and supplier exit