15-Session Programme · 25 Minutes Per Day · Free · For regulators and regulated companies in Nigeria’s petroleum industry
WEEK 3 OF 3 · TRUST, COLLABORATION AND THE LEADERSHIP GAME PLAN · SESSIONS 11–15 · WORK AT YOUR OWN PACE
Part Three · Intelligence in Regulatory PracticeSESSION 11
Gas Markets and Downstream Consumer Protection
Why This Matters
Somewhere in your organisation a complaint, supply dispute or quality concern is sitting in a queue, and the person who raised it does not know what happens next. This session is about recourse: how a consumer, customer or counterparty is heard, the evidence that decides their case, and the honesty of what they are told while they wait. Chapter 11 is written around gas and downstream markets, but its discipline travels. It keeps four things apart that are easily merged: an operational constraint and possible misconduct, a complaint and a confirmed breach, a pattern and a finding, an acknowledgement and a resolution.
25-Minute Module
0:00 – 5:00
Name what is being kept apart
- Gas and downstream markets involve quality, access, reliability, measurement, pricing information and consumer treatment. The leadership problem is telling an operational constraint from possible market misconduct, and a complaint from a confirmed breach. AI can help connect records; the legal and technical tests remain essential.
- NMDPRA's public service information identifies licensing and permit channels, import-related services and gas-network administration. NCELAS is described as supporting licensing and administration under the Nigerian Gas Transportation Network Code. Those services are context for the chapter's proposed workflows, not evidence of AI in use.
Kept apart From Why it matters Operational constraint Possible market misconduct A transport constraint may affect several users without anyone acting improperly Complaint Confirmed breach A report starts an inquiry; it does not end one Analytical pattern Legal finding A pattern may justify further inquiry; it is not a legal finding Projected supply volume Proof of delivery A projection is not proof that a customer can receive it Fast acknowledgement Resolved case A complaint can receive an immediate reply and remain unresolved for months - If your activity is not gas or downstream, find the nearest equivalent: the complaint, recourse or market question closest to the decision you chose in Session 1.
Before anyone describes a case, ask which side of each of these five lines the evidence actually sits.
5:00 – 11:00
Treat gas as a chain of dependent decisions
- Upstream availability, processing, transport, nominations, balancing, delivery and payment are different events, like handovers in a relay: a dropped baton at one handover tells you nothing about the others. The regulator needs evidence about the relevant boundary and rule, not only a national total.
- AI can compare nominations, deliveries and reported constraints and flag inconsistencies. Engineers and market specialists assess pressure, quality, capacity, contractual context and access arrangements. A system should not issue curtailment instructions or reinterpret rights from a forecast alone.
- THE DOCTRINE
Establish comparable periods and bases before comparing reported deliveries. Keep contractual and regulatory questions distinct and assign each to its proper route.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional downstream supervision team receives an industrial customer's delivery complaint. It retains nominations, operating constraints, measurement records and the commercial documents. AI assembles the index and flags mismatches; specialists decide what the records establish, what remains uncertain and which party supplies the next evidence.
AT THE COMPANY · THE PARALLEL- At the company. A fictional gas supplier answering the same complaint assembles the same four record classes on the same basis and period, and separates the contractual question with its customer from the regulatory question. The same loop seen twice: both sides should be able to name the same outstanding piece of evidence and the party who owes it.
If nobody can say which party supplies the next piece of evidence, the case has no next step.
11:00 – 16:00
Quality needs physical evidence; patterns need restraint
- Complaint classification can identify patterns around a product, location or period. It cannot establish quality from language alone. Preserve sample identity, sampling method, laboratory result and chain of custody; a model may help compare these with the applicable specification.
- Hold both possibilities open: an apparent product problem may have a cause unrelated to the supplier, and several credible reports may reveal a serious issue. Intake routes urgent concerns through approved channels and records what evidence will establish the next decision.
- Similar prices or simultaneous changes can result from common costs, logistics, policy or market conditions. Use models to identify questions, not accusations. Record the comparison group, data source, time period and alternative explanations; avoid a convenient comparator that ignores geography, product grade or transport cost. Competition-related decisions require the applicable legal test, adequate evidence and coordination with competent authorities where required.
- THE DOCTRINE
An analytical pattern may justify further inquiry, but it is not a legal finding. A credible investigation should survive scrutiny beyond the model's chart.
When someone shows you a chart of matching prices, ask what the comparison group was before you ask what it means.
16:00 – 21:00
Make recourse understandable, then judge the whole response
- The NMDPRA support service publicly offers general enquiries, licence support, consumer-affairs assistance and whistleblowing channels. That AI could assist routing and acknowledgement there, subject to proper design and authorisation, is an author recommendation, not a claim about the service's current implementation.
- Tell the complainant whether the matter is received, awaiting evidence, under review or determined, and explain the next step in ordinary language. Provide human escalation for material uncertainty and accessibility needs. A chatbot that restates policy without resolving the case creates friction, not service.
Evaluate What the book asks Measure Correct routing, response time, unresolved age and the quality of the outcome Test Unclear language, duplicate reports and urgent conditions Check fairness Whether some users receive poorer responses because their writing or language differs from the training examples Refuse to claim Success merely because automated acknowledgement is fast - A long but necessary technical investigation should be explained, not hidden behind reassuring automated updates. For a company, the same rule governs what you tell a customer or counterparty.
Read your last five status messages as the complainant would. Could they say what happens next?
21:00 – 25:00
Report to the board and write Section 11
- The board should see material quality concerns, unresolved access or supply disputes, repeat complaints and verified corrective action. Allegations, investigations and determinations stay separate in public and executive reporting.
- THE DOCTRINE
Track the burden placed on consumers and businesses, particularly where repeated evidence requests arise from internal handover failures.
- Before you close the laptop, write down the concern connected to your decision that has waited longest, and the status its originator would be given today. Today's assignment produces Section 11 of your plan: recourse, with constraint, complaint, anomaly and breach kept apart, and a response people can understand.
Market intelligence earns its place when it improves the quality of intervention and the credibility of recourse, not when it adds observations.
Key Terms
Operational constraint
A physical or operating limit, such as a transport constraint, that may affect several users; distinct from possible market misconduct.
Analytical pattern
A regularity found in data, such as similar or simultaneous price changes; it may justify further inquiry but is not a legal finding.
Chain of custody
Documented handling of evidence from capture through use, appropriate to its purpose.
Complaint status
The plain-language position of a matter (received, awaiting evidence, under review or determined) with the next step explained.
Self-Check — Before You Submit
1. A market-monitoring model shows that several fictional retailers changed prices on the same day. A director proposes announcing an investigation into collusion. What should you do first?
- A. Announce the investigation, because the pattern is strong
- B. Treat the pattern as a question: record comparison group, source, period and alternative explanations, and route any competition question to the applicable legal test
- C. Ask the model to rank the retailers by likelihood of collusion
- D. Close the matter, because simultaneous changes are always explained by common costs
B. Treat the pattern as a question: record comparison group, source, period and alternative explanations, and route any competition question to the applicable legal test — Similar or simultaneous prices can result from common costs, logistics, policy or market conditions, so a pattern justifies inquiry, not an accusation. Closing it automatically is the opposite error: the chapter keeps both possibilities open.
2. Several consumers describe the same apparent product problem in similar words. What does the complaint classification establish?
- A. That the product failed its specification
- B. That the supplier is responsible
- C. Nothing of use; similar wording is noise to be filtered out
- D. A pattern around a product, location or period that may warrant examination; quality needs sample identity, sampling method, laboratory result and chain of custody
D. A pattern around a product, location or period that may warrant examination; quality needs sample identity, sampling method, laboratory result and chain of custody — Classification cannot establish quality from language alone, yet several credible reports may reveal a serious issue. Saying it shows nothing ignores the pattern's value in directing attention.
3. A new complaints assistant acknowledges every complaint within a minute. The sponsor wants to report the service as a success. What is the right test?
- A. Correct routing, response time, unresolved age and outcome quality, tested with unclear language, duplicates and urgent conditions
- B. Acknowledgement speed, because it is what complainants notice first
- C. The number of complaints the assistant handled
- D. Customer satisfaction with the acknowledgement message
A. Correct routing, response time, unresolved age and outcome quality, tested with unclear language, duplicates and urgent conditions — The book warns against claiming success because automated acknowledgement is fast: a complaint can receive an immediate reply and remain unresolved for months. Volume handled is activity, not outcome.
Today’s Assignment
Write Section 11 of your plan: how a complaint, supply question or quality concern connected to your decision moves from intake to determination in your organisation, and how the person who raised it is kept informed.
- Choose the complaint, supply question or quality concern nearest your Session 1 decision: for an institution, the case it receives; for a company, the matter it must answer or raise.
- Set out the steps from intake to determination and the physical and documentary evidence each needs, including comparable bases and, for quality, sample identity, method, laboratory result and chain of custody.
- Show how allegation, investigation and determination are kept apart in your organisation's records and reporting, and how an operational constraint is distinguished from possible misconduct.
- Write the status messages your organisation would send at received, awaiting evidence, under review and determined, in ordinary language, and the route for urgent and accessibility needs to a human.
- State how you would evaluate the response beyond acknowledgement speed, and what your organisation gets wrong today. If no AI is in use, work from the process in front of you.
YOUR 90-DAY PLAN
Section 11 — Recourse: constraint, complaint, anomaly and breach kept apart, and a response people can understand.
How this is marked — the rubric
| Traces intake to determination with the evidence each step needs, on comparable bases | 30 |
| Keeps constraint, complaint, anomaly, allegation, investigation and determination visibly apart | 25 |
| Gives plain-language status messages and a working human route for urgent and accessibility needs | 25 |
| Evaluates the response by routing, unresolved age and outcome quality, not acknowledgement speed, and names a current weakness | 20 |
| 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
Recourse designEvidence for supply disputesProduct-quality evidencePattern restraintPlain-language statusBoard reporting
Part Four · Trust, Control and SecuritySESSION 12
AI Governance and Administrative Fairness
Why This Matters
An erroneous internal summary wastes time. An erroneous adverse determination can affect rights, investment and reputation. Both need controls, but not the same controls. This session is about governance that fits the consequence: knowing every AI service your organisation uses, making human review real rather than a click, testing fairness where it can be seen, and writing an acceptance case before anyone shows you a score. It also gives the people your decisions affect a route to correct the record.
25-Minute Module
0:00 – 5:00
Start with the inventory and its dependency chain
- Identify the decision and affected parties before choosing the assurance approach. The NIST AI Risk Management Framework is voluntary guidance structured around Govern, Map, Measure and Manage; it is not a Nigerian petroleum regulation and confers no authority to decide. Chapter 12's controls are the author's proposals.
- Record every AI service in use, including small team tools and those that classify, summarise or translate: purpose, model or service version, data, permitted users, outputs, accountable owner and consequences.
- Then map dependencies. A service may rest on OCR (turning an image of text into candidate text), retrieval, a language model and a workflow engine. As in a supply chain, failure can arise at any link: a correct extracted value does not prove the source was the current revision.
- THE DOCTRINE
Governance needs to inspect the chain. A good model answer does not prove that the correct source was retrieved.
Ask each department head which AI tools their teams actually use. The inventory usually grows.
5:00 – 10:00
Make human review meaningful and the record reconstructable
- Human review is meaningful when the reviewer has competence, time, source access and authority to disagree. A required click at the end of a rushed process is not sufficient; the interface should show evidence and uncertainty.
- Reviewers state reasons in terms of evidence and applicable criteria, verifying any AI draft they use. The system recommended it is not a professional reason.
- Explainability must fit the user: an engineer needs the measurement basis; an applicant the evidence gap and route to respond; an assurance reviewer source, version, permissions and decision history.
- For consequential cases, preserve sources, relevant AI outputs, model version, final reasoning, overrides and corrections under the applicable retention policy. The objective is reconstruction, not unlimited logging.
If your organisation could not explain a determination without quoting the model's confidence, the review was not meaningful.
10:00 – 15:00
Test fairness visibly and keep the route to correction open
- Test whether the service performs differently across application classes, facility types, languages and evidence formats. A model may penalise poor scans or unfamiliar wording while appearing to assess the activity.
- Not every difference proves unfairness; investigate and document the cause. Include smaller applicants and less polished submissions, so a digital advantage does not quietly become preferential treatment.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional licensing department's classifier marks an application high-concern; the applicant disputes it. The reviewer retrieves the actual criteria, inputs, source quality and proposed explanation, and asks whether the signal concerns operating evidence or only presentation quality. It records disposition, correction and reason, keeping the model output separately for evaluation.
AT THE COMPANY · THE PARALLEL- At the company. A fictional operator whose submission was classified this way identifies the disputed fact in ordinary language and uses the correction route, without accepting an unsupported classification as fact. Internally, it applies the same test to its own pack-checking tools.
- THE DOCTRINE
Distinguish a correction request, a complaint and a formal appeal. AI can help route the matter but should not collapse them into a single automated response.
Put your least polished case into every test set. That is where unfairness hides.
15:00 – 21:00
Write the acceptance case, then measure
- The proposed acceptance case argues that a defined service can be used for a stated purpose under identified controls: task, consequence of error, required evidence, professional review, fallback and remaining restriction. The design is the author's. A favourable result on clerical detection cannot justify a wider permission.
- Use authorised records representative of the real queue, not only legible ones. Competent reviewers set reference judgements (the agreed answer for each case); uncertain cases stay uncertain, and development material stays apart from the final set.
- Of everything flagged, how much was right? That is precision: the proportion of flagged gaps confirmed by reference review. Of everything it should have found, how much did it find? That is recall: the proportion of reference gaps detected. Illustrative teaching inputs only: if 40 flagged gaps include 30 confirmed, precision is 30 ÷ 40 = 75 percent; if reviewers found 36 reference gaps and 30 were detected, recall is 30 ÷ 36 ≈ 83.3 percent. A zero denominator makes the metric undefined, not perfect.
Proposed measure Denominator or comparison Decision informed Confirmed-gap precision Flagged gaps independently reviewed Burden of unnecessary queries Material-gap recall Material reference gaps Consequential omissions Supported-assertion rate Material assertions reviewed Fidelity of drafting Correct-source-version rate Source uses checked Suitability of retrieval Abstention quality Declined answers assessed Usefulness of escalation Net review effort Baseline and full service effort Capacity effect Boundary-test outcome Defined permission and failure tests Permitted operating scope - These are author metric definitions, not regulator targets; there is no universal acceptance threshold. Test abstention (declining, with the missing evidence named) and recovery (remove a source, stop the AI component, check the case stays intelligible). Deliberately altered test documents never enter a submitted pack or an investigation record.
An aggregate percentage can conceal the one consequential omission. Ask for material gaps on their own line.
21:00 – 25:00
Monitor after release and write Section 12
- After release, track source changes, record quality, overrides, corrected assertions and downstream errors, and diagnose the component before prescribing retraining. Re-run a regression collection when sources, prompts, models, permissions or workflows change; the voluntary NIST Generative AI Profile addresses pre-deployment testing and continuing monitoring.
- The board sees evaluation failures by consequence, unsupported-source rates, override patterns, complaints, overdue corrective actions and the ability to stop or roll back. Accepting one release does not accept every later version.
- THE DOCTRINE
Your organisation should be able to say who owns the service, what it may do, how it was tested and how a person can obtain correction.
- Write down the AI tool in your organisation that nobody has inventoried. Today's assignment produces Section 12: the inventory, meaningful review, fairness tests, the acceptance case and the route to correction.
A stable dashboard colour cannot replace an executive decision about an exception.
Key Terms
Human review
Competent assessment with source access, time and authority to reject an AI output.
Acceptance case
The author's proposed argument that a defined service can be used for a stated purpose under identified controls: task, consequence of error, required evidence, review, fallback and remaining restriction.
Precision and recall
Precision: the proportion of flagged gaps confirmed by reference review. Recall: the proportion of reference gaps detected. Each needs a stated denominator; a zero denominator makes it undefined.
Drift
A change in data or performance that can undermine the service's previous evaluation.
Self-Check — Before You Submit
1. In an evaluation, the reference reviewers found no material gaps in the sample, so material-gap recall has a zero denominator. The supplier reports recall as 100 percent. What is correct?
- A. Accept 100 percent, since no material gap was missed
- B. Report recall as 0 percent
- C. Report the metric as undefined and define the next evidence task
- D. Replace recall with precision, which is defined
C. Report the metric as undefined and define the next evidence task — The book is explicit: a zero denominator makes the metric undefined; it does not establish perfect performance. Substituting precision answers a different question, the burden of unnecessary queries.
2. A reviewer approves an adverse classification, recording the reason as: the system recommended it. Under Chapter 12, what is wrong?
- A. The reason must be stated in terms of the evidence and applicable criteria; a recommendation is not a professional reason
- B. Nothing, if the model's confidence was high
- C. The reviewer should have attached the model's confidence value
- D. The decision should have been taken by the model owner
A. The reason must be stated in terms of the evidence and applicable criteria; a recommendation is not a professional reason — Reasons must be tied to evidence and criteria. Attaching a confidence value is the tempting answer, but the book says confidence is not a substitute for that work.
3. A fictional applicant writes in to say a figure in their record is wrong. The new routing assistant files it as a formal appeal. Why is that a problem?
- A. Appeals are always handled faster
- B. The applicant should have used technical language
- C. AI should never be used to route correspondence
- D. A correction request, a complaint and a formal appeal may have different authorities and procedures; routing should not collapse them
D. A correction request, a complaint and a formal appeal may have different authorities and procedures; routing should not collapse them — AI may help route the matter but not collapse distinct routes into one automated response. The book does not ban AI routing, which is the tempting overcorrection.
Today’s Assignment
Write Section 12 of your plan: the governance for the AI service your decision would use or already uses, from inventory entry to the route by which an affected party corrects the record.
- Write the inventory entry (purpose, version, data, users, outputs, owner, consequences) and map the dependency chain with the failure each link could cause.
- State what would make human review meaningful in your organisation and where today's review is closer to a required click.
- Set the fairness tests across application classes, facility types, languages and evidence formats, including how smaller or less polished submissions enter the test set.
- Write the acceptance case: task, consequence of error, required evidence, professional review, fallback and remaining restriction. Choose the measures from the book's table you would report and state each denominator. Do not invent an acceptance threshold; say who would set it.
- Describe how an affected party questions and corrects a record, keeping correction request, complaint and formal appeal apart. A company describes the route it uses and the test it applies to its own tools. If no AI is in use, govern the proposal in front of you.
YOUR 90-DAY PLAN
Section 12 — Governance and fairness: the inventory, meaningful review, fairness tests, the acceptance case and the route to correction.
How this is marked — the rubric
| Inventory entry complete, with a dependency chain whose links and failure points are named | 20 |
| Meaningful review defined by competence, time, source access and authority, with an honest current gap | 20 |
| Fairness tests across classes, facility types, languages and formats, including less polished submissions | 20 |
| Acceptance case with all six elements and measures from the book's table with stated denominators and no invented threshold | 25 |
| A usable correction route that keeps correction request, complaint and formal appeal apart | 15 |
| 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
AI inventoryMeaningful reviewFairness testingAcceptance casePrecision and recallRoute to correction
Part Four · Trust, Control and SecuritySESSION 13
Enforcement, Cybersecurity and Controlled Autonomy
Why This Matters
This session is about power: the power to investigate, the power an attacker gains by compromising a system, and the power you hand to software when it can act rather than only advise. An investigation that follows a model's suspicion can harm an innocent party. An administrator with an incidental permission can issue what looks like a determination. An agent that retries a timed-out message can send it twice. Chapters 13 and 14 and Appendix F give you the boundaries: what an inquiry record keeps apart, how confidential reports and systems are protected, which actions software may take, and the five gates a service passes before release.
25-Minute Module
0:00 – 5:00
Begin enforcement with evidence and authority
- An anomaly, allegation or risk ranking can support inquiry. It is not itself a finding of breach. A model can connect events that look suspicious but have innocent explanations, or miss a material issue because history holds few examples.
- An investigation record holds the issue, authority, evidence sources, scope, authorised personnel and questions to be resolved, and keeps allegations, working hypotheses, established facts and professional conclusions apart. A generated chronology identifies its sources and gaps.
- THE DOCTRINE
The purpose of investigation is to establish what happened, not to find words that confirm the first suspicion.
- Test alternative explanations: in the book's examples, a payment pattern may reflect a contractual milestone, a repeated meter adjustment an instrument defect, a document similarity a common template. AI may index and compare within the authorised scope, never as broad surveillance; material extracts are verified at source.
Ask for the innocent explanation of every pattern before you ask for the guilty one.
5:00 – 10:00
Protect the report, the system and the work
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional inspection team finds a component certificate that conflicts with its inspection evidence. It establishes component identity, revision, time and scope, preserves the originals and records the investigation. A discrepancy warrants examination; it is not an automatic finding of falsification.
AT THE COMPANY · THE PARALLEL- At the company. A fictional operator receiving the query preserves its originals and answers on identity, revision, time and scope, rather than defending or withdrawing the document in haste.
- Whistleblowing information needs restricted access. A search assistant must not expose a confidential report to unrelated users; test indirect queries too, because a summary can identify a person through context. Routing must not automatically notify the party implicated, and fails safely when permissions are uncertain.
- A compromised system can alter, expose or withhold records. Limit privileges, monitor unusual access, test recovery, and separate administrators' technical powers from decision-makers' professional authority. Submitted files may hide instructions aimed at the AI, known as prompt injection; model instructions alone are not a security boundary.
- In an incident, have a tested route to suspend affected services, retain evidence and continue essential work. A backup is useful only if recovery has been demonstrated. Distinguish confirmed effects from possible exposure.
- THE DOCTRINE
Do not measure enforcement success by the number or value of penalties alone; those incentives can distort institutional judgement.
Ask who in your organisation could change a case status without being authorised to decide it.
10:00 – 15:00
Define the action before discussing autonomy
- An agent uses tools and takes steps: retrieving a file, creating a task, drafting a message. The benefit is less coordination effort; the exposure is acting on a wrong assumption, an untrusted instruction or excessive permission. Specify tools, users, destinations and changes, then test each boundary.
- Begin with preparation: a case briefing from approved records and proposed tasks, with the case owner confirming the next step. Good drafting does not earn authority to submit a legal interpretation, impose a condition or publish an allegation.
- Then, where appropriate, bounded execution: a narrow reversible action, such as an internal reminder to a verified recipient, with rate limits, cancellation and duplicate prevention. Verify external recipients from the controlled case record, never from a document instruction.
- The book's bounded reminder service checks the task is still open, avoids duplicates, preserves delivery evidence and stops when destination or case state is unresolved. A reminder does not create a statutory deadline or settle a case.
A stop button that changes the screen while background tasks continue is not a stop button.
15:00 – 20:00
Treat each action as a controlled transaction
Stage What happens Proposal Purpose, case, recipient, payload, source basis and required permission identified Validation A deterministic validator checks fields and current state at execution, not only at proposal Authorisation The accountable role authorises according to consequence Dispatch Sent through the permitted interface under a stable identifier Confirmed result Receipt retained; an uncertain outcome stays uncertain until reconciled - Evaluate the whole workflow: an agent can retrieve a superseded source, draft a correct-looking query and send it to the wrong recipient. If an interface times out, establish whether the action was accepted before retrying.
- Treat applicant files as evidence, never instructions. Test controlled copies that try to disclose another applicant's records, ignore review, change a destination or mark a case approved.
- THE DOCTRINE
The institution cannot transfer accountability to a supplier or model by describing it as autonomous. Capability and control should be evaluated together.
Ask whether a retry in your process could send the same message twice.
20:00 – 25:00
Pass the five gates and write Section 13
- Appendix F is the author's procurement discipline, not a prescribed procurement procedure. A compelling demonstration substitutes for no gate.
Gate What must be shown 1. Define the service Process, decision, scope, authority, owner; why AI rather than conventional software or process improvement; benefit, consequence of failure, continuing cost 2. Validate the evidence Sources, identities, revisions, access, classification, reference judgements; ordinary and difficult cases; limitations 3. Test the controls Permissions, source fidelity, logging, interruption, export, recovery; poor scans, conflicting sources, malicious instructions; who can stop and restore 4. Prove the operating workflow The complete process in shadow mode, with review effort, rework, competence, support and fallback 5. Authorise a bounded release Permitted users, records, tasks and actions; restrictions; approvals; monitoring and withdrawal; no permanent acceptance of future versions - Ask the supplier who owns and can export the evidence, what it retains, how restrictions hold in generated answers, and what happens at outage, contract expiry or staff turnover. An assurance without a practical test is weak evidence.
- Today's assignment produces Section 13. Pop Quiz 3 opens now: five questions on Sessions 11 to 13, closed book.
The procurement should leave your organisation with a sustainable service and a clear exit route.
Key Terms
Agentic AI
AI that uses tools or performs steps toward a goal, requiring explicit controls over actions.
Controlled transaction
The author's proposed separation of an action into proposal, validation, authorisation, dispatch and confirmed result.
Prompt injection
An attempt to make a model follow instructions embedded in untrusted material, such as a submitted file.
Shadow mode
Evaluation alongside the existing process without determining external action.
Self-Check — Before You Submit
1. A model flags that two fictional companies' technical reports share long identical passages. An officer wants to record a finding of falsification. What should happen?
- A. Record the finding, since the similarity is exact
- B. Ask the model to estimate the probability of falsification
- C. Test alternative explanations, such as a common template, preserve the originals and establish the facts first
- D. Notify both companies they are under investigation for falsification
C. Test alternative explanations, such as a common template, preserve the originals and establish the facts first — Document similarity may reflect a common template, and a discrepancy is not an automatic finding of falsification. Notifying the parties is the tempting shortcut that turns intelligence into accusation.
2. An agent sends an evidence request and the external interface times out. What should the service do?
- A. Use the action's stable identifier to establish whether it was accepted, and keep the outcome uncertain until reconciled
- B. Retry immediately, since the first attempt probably failed
- C. Retry three times, then stop
- D. Mark the request as sent
A. Use the action's stable identifier to establish whether it was accepted, and keep the outcome uncertain until reconciled — A blind retry can create duplicate messages and an unreliable case history. Marking it sent is the opposite error: it records a result nobody confirmed.
3. An operator's board report says that a reminder issued by a bounded reminder service has itself created a regulatory deadline. Is that right?
- A. Yes, because it came from the regulator's system
- B. Yes, if the reminder used an approved template
- C. Only if the agent was in bounded execution
- D. No; a reminder does not create a statutory deadline or settle a case
D. No; a reminder does not create a statutory deadline or settle a case — The book says so expressly. An approved template governs the reminder's wording, not its legal effect, which is the trap in the second option.
Today’s Assignment
Write Section 13 of your plan: the action boundary for the service your decision would use, and which of Appendix F's five gates it has passed.
- State what an inquiry record connected to your decision must keep apart (allegation, hypothesis, established fact, conclusion) and the alternative explanations it must test. Companies write it for an internal inquiry or a regulator's query. Name no real person or company.
- Describe how confidential reports and the system itself are protected in your organisation, including separation of technical from decision authority and handling of instructions hidden in files, and how essential work continues during an incident.
- List which actions an agent may prepare, which narrow reversible actions it may execute and which it must never take, and how each is validated, authorised, de-duplicated and stopped.
- Mark each of Appendix F's five gates as passed, partly passed or not passed, with the evidence, and name the gate your organisation is most tempted to skip. If no AI is in use, apply the gates to the proposal in front of you.
YOUR 90-DAY PLAN
Section 13 — Investigation, security and controlled autonomy: the action boundary and the five acceptance gates.
How this is marked — the rubric
| Inquiry record keeps allegation, hypothesis, fact and conclusion apart and tests alternative explanations without implying guilt | 25 |
| Protection and continuity are specific: confidential reports, separated powers, untrusted instructions and recovery honestly assessed | 20 |
| Action boundary lists prepare, execute and never-take actions, with validation, authorisation, de-duplication and stop controls | 30 |
| Each gate assessed against evidence, with the gap most likely to be skipped named | 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 — Recourse, Governance and Controlled Autonomy 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
Investigation recordsWhistleblower protectionCyber and continuityControlled autonomyAcceptance gatesSupplier challenge
Part Five · Capital, Capability and CollaborationSESSION 14
Better Regulatory Partners and Nigerian Capability
Why This Matters
Many company-regulator difficulties begin inside the company: a schedule that omits approvals, an engineering revision that never reaches compliance, a board told green because a file was submitted. Chapter 15 is written to the company. If you lead one, apply it directly. If you lead a regulatory institution, apply it in reverse: what you ask of companies, how you make good evidence easier to give, and how you recognise a reliable partner without granting preferential treatment. Then Chapter 16 asks both sides one question: which capability must stay in your own hands?
25-Minute Module
0:00 – 5:00
Build the obligation map and one controlled pack
- Map each activity to its applicable obligation, authority, trigger, evidence, owner and dependency, using the controlled source process from Session 3, and link it to engineering, operations, finance, HSEQ and project management. Compliance should not reconstruct the whole company's evidence at the last moment.
- Identify the correct legal entity and activity: a group-level permit may not establish that a subsidiary's project is authorised. Qualified staff determine applicability; AI should not infer entitlement from similar past work.
- One controlled submission pack: revisions and superseded documents identified and a consistent technical basis across drawings, narrative, calculations and schedules. Retain the exact submitted version and receipt; a working folder that later changes is not a record of what the regulator received.
- THE DOCTRINE
A polished AI submission cannot compensate for missing technical work or an inaccurate statement. The responsible professionals still certify the content within their roles.
Ask your compliance lead which submitted pack they could reproduce exactly, today, with its receipt.
5:00 – 10:00
Draft to Appendix E and keep the model honest
- Appendix E offers drafting structures, not prescribed letters: adapt them to the institution's current form and process.
Appendix E structure Elements Company submission cover Decision requested for entity, facility and activity; controlled basis and revision; evidence index; limitations or pending matters; changes and superseded documents; verified contacts; authorised signatory Evidence index Document identifier; title and revision; requirement addressed; status and responsible reviewer Regulatory query Case reference; specific gap; evidence reviewed with source and revision; validated criterion; response required; timing under the applicable verified process; responsible officer Company query response Answer to the actual issue; supporting evidence and revision; change made; dependent documents; remaining uncertainty; professional sign-off - THE DOCTRINE
Do not describe the activity as approved or fully compliant unless that statement is accurate and supported. If a test remains pending, say so and explain the effect on the request.
- Prohibit fabricated citations, invented test results and unsupported assurances; keep plant drawings, personnel records and commercial material out of uncontrolled services. The signatory should know what was generated and what was checked.
- For institutions: a query should preserve uncertainty, using the records appear inconsistent where that is the known position, and explain why the evidence matters to the decision.
A response should make the review easier without obscuring the problem. Unrelated bulk is not evidence.
10:00 – 15:00
Run queries as project work and classify waiting by cause
- Give each query an owner, response requirement, supporting documents and internal target, and distinguish the regulatory deadline from the company's planning target. Avoid a general awaiting regulator label when the company owes evidence.
- For a contractor technical package, verify each item against the controlling requirement and accepted revision. AI can index and compare; it must not turn missing evidence into a generated certificate.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional licensing department reviews its open cases with one question per case: what is outstanding, and who owes it? It publishes evidence expectations through the proper route, offers every company the same ones, and recognises a reliable partner by evidence quality and commitments kept, not by access.
AT THE COMPANY · THE PARALLEL- At the company. A fictional operator's weekly review lists, for each case, the decision requested, current issued basis, company action, institutional action, project dependency and next owner. It ends with evidence tasks. Its report separates time awaiting its own evidence from time awaiting institutional review, and treats a meeting note as guidance, not approval.
- Record receipt, query, response, assessment and decision events, and classify waiting by actual cause. Do not hide a completed company response while institutional review is pending, and do not let a model invent a probable approval date.
- THE DOCTRINE
Relationship quality comes from reliable commitments and evidence, not informal access that bypasses criteria.
Ask whether the regulator and your company would name the same outstanding item on the same case.
15:00 – 21:00
Keep the capability in your own hands
- A programme becomes dependent on a supplier when the organisation buys a system without the capability to question, operate and improve it. Local capability is more than hosting location or attendance.
- Build multidisciplinary teams: operational users, legal and technical reviewers, information management, security and privacy expertise, with independent assurance at gates. A developer should not decide whether a rule applies to a licence.
- Train through real work and assess through performance: can the reviewer identify a superseded source, or the case owner recover a task after an outage? Attendance counts, certificates and photographs document participation; they do not establish capability.
- Procure transfer: documentation, configuration records, exports that preserve relationships, revisions, permissions and decision history, and staff demonstrations without the supplier narrating. Where an external specialist is needed, make the dependency explicit and fund it.
- Nigerian content obligations require their own applicable legal treatment; the book does not equate hiring a local supplier with fulfilment of every such requirement. Supervised apprenticeships, joint training and defined work packages, with local teams given meaningful responsibility, are author recommendations.
Ask your team to restore a sampled case from the export, without the supplier in the room.
21:00 – 25:00
Report honestly and write Section 14
- The company board sees approvals by stage, open conditions, company-owned response delays, repeat queries, evidence quality and project dependencies, with submitted, under review, approved and conditions fulfilled kept separate. Exposure uses stated assumptions, not invented regulatory forecasts.
- Both boards see demonstrated competence, unresolved supplier dependencies, export tests and continuity after personnel changes. Joint workshops can clarify definitions and interfaces without discussing confidential cases; consultation stays distinct from final decisions.
- Today's assignment produces Section 14: the submission and query discipline, waiting by cause, and the capability you keep. Practice Exam 3 opens now: fifteen questions on the full syllabus to date, closed book.
The best regulatory partner makes its evidence easier to assess and its commitments easier to verify.
Key Terms
Company obligation map
Each activity linked to its applicable obligation, authority, trigger, evidence, owner and dependency, built from controlled sources.
Controlled submission pack
The exact version sent, with cover, evidence index, revisions, superseded documents and receipt retained.
Waiting by cause
Elapsed time classified by who owed the next step, so time awaiting company evidence is not reported as regulator delay.
Capability transfer
Evidence that the organisation's own staff can operate, govern and correct the service, shown by demonstration rather than attendance or certificates.
Self-Check — Before You Submit
1. After an informal meeting, a fictional operator's project manager reports to the board that the regulator has approved the modification. What is wrong?
- A. Nothing, if the meeting was with a senior official
- B. A meeting note can support understanding but is not an approval unless the required decision has been issued through the proper process
- C. The report should have waited for the minutes to be signed
- D. The project manager should have asked for an informal email confirmation
B. A meeting note can support understanding but is not an approval unless the required decision has been issued through the proper process — Chapter 15 is explicit that a meeting note is not an approval. Signed minutes or an informal email are the tempting fixes, but neither substitutes for a decision issued through the proper process.
2. A submission assistant drafts the line: the facility is fully compliant with all applicable requirements. One test is still pending. What should the company write?
- A. Keep the line, since the test is expected to pass
- B. Soften it to substantially compliant
- C. Replace it with the actual evidence and its limits, stating that the test is pending and its effect on the request
- D. Remove any mention of the test
C. Replace it with the actual evidence and its limits, stating that the test is pending and its effect on the request — The book replaces such statements with actual evidence and limits, and says to state a pending test. Substantially compliant is still an unsupported assurance in softer words.
3. A fictional institution's supplier shows forty staff certificates as proof of capability transfer. What evidence does Chapter 16 ask for instead?
- A. Staff demonstrating source updates, access changes, export and fallback on the actual service, unaided
- B. A larger number of certificates
- C. Photographs of the training sessions
- D. A longer operations manual
A. Staff demonstrating source updates, access changes, export and fallback on the actual service, unaided — Attendance counts, certificates and photographs document participation but do not establish capability. A large manual does not demonstrate transfer either if nobody can apply it.
Today’s Assignment
Write Section 14 of your plan: how your organisation's side of the regulatory interface works for your decision, and the capability your organisation must keep in its own hands.
- Describe the obligation map (company) or review expectation (institution) your decision relies on, including the legal entity, activity and authority, and where applicability is still for a qualified reviewer.
- Rewrite one real submission cover, query or query response to Appendix E, confidential details removed, and say what your organisation previously overstated or left out.
- Show how queries and waiting are classified by actual cause in your organisation, and how your board report separates submitted, under review, approved and conditions fulfilled, or the equivalent states on your side.
- List what the AI tools in use are prohibited from generating (for example fabricated citations, invented test results, unsupported assurances) and the data that must never go to an uncontrolled service.
- Name the capability your organisation must keep: the competencies, the demonstration that proves them and the supplier dependency to remove. Institutions add how they recognise a reliable partner without preferential treatment. If no AI is in use, work from the interface as it runs today.
YOUR 90-DAY PLAN
Section 14 — The partnership and the capability: the submission and query discipline, waiting by cause, and the capability we keep.
How this is marked — the rubric
| Obligation map or review expectation names entity, activity, authority and open applicability questions | 20 |
| One Appendix E structure applied to a real document, with confidential details removed and the previous weakness named | 25 |
| Waiting classified by cause and reporting statuses kept separate, with no invented approval dates | 20 |
| Clear prohibitions on AI-generated content and data handling | 15 |
| Capability to keep defined by competencies, a performance demonstration and a named supplier dependency to remove | 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.ASSESSMENT NOW OPEN
Practice Exam 3 — Full Syllabus 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
Obligation mappingControlled submissionsQuery disciplineWaiting by causeCapability transferFair partnership
CapstoneSESSION 15
The First Ninety Days: Your Signed 90-Day Regulatory AI Plan
Why This Matters
Fourteen sessions ago you wrote one sentence about one decision that costs your organisation time, quality or trust. Since then you have built one section of the plan each session. Today you turn those fourteen sections into one signed 90-Day Regulatory AI Plan: a pilot charter, a decision-rights matrix fitted to real authority, ninety days stage by stage, the measures you will report and the acceptance decision you will ask for. The plan is credible because it can stop. Then you sit the Final Exam.
25-Minute Module
0:00 – 6:00
Lay out the ninety days
- Chapter 17's timetable is an author planning framework, not a legal deadline or a promise that any organisation will reach maturity within a fixed period.
- Days 1 to 15, establish the decision. Trace representative cases from receipt to outcome; set the baseline (active work, waiting, rework, quality); confirm legal sources and decision rights; name the executive sponsor, process owner and service owner. Write the pilot charter before purchasing.
- Days 16 to 30, prepare the evidence. Resolve identities, revisions and access; build an evaluation set of routine, difficult and incomplete cases, with a separate final set. The first milestone is a trustworthy record. Some processes improve enough through foundations that further AI expenditure becomes unnecessary; that is a legitimate result.
- Days 31 to 60, run in shadow mode. Outputs determine no external action. Record each failure with consequence and cause, test poor scans, conflicting revisions and hostile instructions, and measure reviewer burden. A new prompt is not a fix until tested.
- Days 61 to 90, decide on controlled use. Review against the acceptance gates with professional and operational sign-off. A limited deployment may be justified when broader use is not; if the pilot fails, record why and what reconsideration would need.
- THE DOCTRINE
A programme should be able to stop a weak pilot without calling that responsible decision a failure of ambition.
Do not release a high-consequence service because the budget is spent or a launch date was announced.
6:00 – 10:00
Plan the year and the investment case
- Months 4 to 6, stabilise. Ask whether rework fell and decisions became more explainable, allowing for changes in staffing and case mix. Months 7 to 12, expand by dependency, reusing identities, source governance and evaluation discipline, renewing acceptance where the population or task changes, and stopping expansion that multiplies an unresolved weakness.
- Build the investment case from verified inputs (currency, basis date, source; committed, quoted or uncertain), with implementation, operation, transition and exit costs apart. Test it under lower benefit and higher review effort.
- THE DOCTRINE
If a redesigned process and AI are introduced together, the observed improvement belongs initially to the complete service. Three reliable services are stronger than twelve unsupported demonstrations.
A proposal should remain understandable when its unsupported benefits are removed.
10:00 – 15:00
Complete the Appendix A charter
- Appendix A is an author template, not a prescribed regulator form. Its fields: pilot title; executive sponsor, process owner and AI service owner; decision to improve; problem and baseline; public value or company benefit; controlling sources; permitted AI role; excluded actions; evidence collection; human review; evaluation; stop conditions; fallback and recovery; cost and support; and the acceptance decision, adopt within defined scope, revise or stop, with responsible sign-off and reasons.
- The worked example checks one application class for evidence presence and conflicting revisions; it may not certify technical adequacy, determine legal entitlement or send an external query without approval.
- Acceptance requires the institution to set its own tolerable error and review burden. No universal accuracy threshold is supplied. A missed critical omission can matter more than several harmless extraction errors.
Write the excluded actions as carefully as the permitted role. That is where a pilot stays bounded.
15:00 – 20:00
Fit decision rights to lawful authority
Activity AI contribution Accountable human responsibility Index a submission Propose identifiers and evidence links Case officer verifies identity and scope Retrieve a provision Return source and relevant passage Qualified reviewer establishes applicability Assess technical evidence Identify candidate inconsistencies Competent professional assesses adequacy Prioritise supervision Propose rank with evidence and limits Authorised supervisor sets priorities Draft a query Prepare source-linked working text Responsible officer verifies and authorises Issue a determination Assemble verified supporting record Lawful decision-maker determines and issues Close a finding Compare response with requirement Authorised reviewer verifies closure Change the AI service Support comparison and evaluation Service owner controls change and approval - Appendix C's author matrix must be adapted to actual lawful authority: job titles alone do not establish delegated powers. Its decision record runs from decision requested and authority, through facts, limitations, professional assessments, AI assistance and alternatives, to determination, recourse and outcome review.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional licensing director replaces each role with the officer holding the relevant delegation, and leaves a row visibly unresolved where the delegation is unclear.
AT THE COMPANY · THE PARALLEL- At the company. A fictional operator reads the same matrix from its side: who verifies the pack, who assesses adequacy, who authorises a query response and who signs.
- The review question: could a qualified person who was not involved locate the sources, understand the uncertainty and see why the decision follows?
Where you cannot name the authority for a row, you have found your first Days 1 to 15 task.
20:00 – 25:00
Choose the measures, sign and sit the Final Exam
Appendix D measure Interpretation limit Median case age Does not reveal oldest difficult cases Tail case age Needs consistent case scope Avoidable rework rate Requires agreed cause classification Source fidelity Sample and consequence matter Open material actions Risk should remain visible Correction rate Distinguish AI and other causes Net capacity released Is not automatically cash saving Recovery readiness A policy alone is insufficient - These are author measures, not official targets; agree definitions and denominators first. In the benefits ledger, a capacity entry names baseline, comparable population, service effort and added review, claiming cash only when an actual cash-cost change is established; a reconciliation entry separates corrected uncertainty from recovered product, accepted settlement and received cash. The monthly page opens with decisions requiring leadership. The rhythm: weekly tasks and blockers, monthly benefit, quality and control, quarterly portfolio and capability, and independent assurance.
- THE DOCTRINE
The practical test is whether a decision can be explained, its action followed and its outcome verified.
- Assemble Sections 1 to 14 under a one-page summary, sign and date it. Then sit the Final Exam: fifteen questions, three rebuilt from your own earlier mistakes, closed book, pass mark 70.
Sign only what you would be prepared to review honestly on day ninety.
Key Terms
90-Day Regulatory AI Plan
Sections 1 to 14 assembled around one decision under a one-page summary, signed and dated.
Pilot charter
Appendix A's author template defining a pilot before procurement, from the decision and owners to stop conditions and the acceptance decision.
Decision-rights matrix
Appendix C's author table of what AI contributes and which accountable human decides, adapted to actual lawful authority.
Benefits ledger
Appendix D's record of claimed benefit with baseline, comparable population and effort, keeping capacity apart from established cash effect.
Self-Check — Before You Submit
1. On day 85 a fictional pilot is still missing material omissions in shadow mode, but the budget is spent and a launch was announced. What should the acceptance decision be?
- A. Adopt, because the launch was announced
- B. Adopt in full and fix the omissions in Months 4 to 6
- C. Adopt within a scope the evidence supports, or revise or stop, recording why
- D. Extend shadow mode indefinitely without a decision
C. Adopt within a scope the evidence supports, or revise or stop, recording why — Chapter 17 says not to release a high-consequence service because the budget is spent or a launch announced. Fixing it after release makes adoption unconditional.
2. A fictional company fills Appendix C's accountable-human column by copying job titles from its organisation chart. What does Appendix C warn?
- A. Job titles alone do not establish delegated powers
- B. The matrix is a prescribed regulator form and must not be changed
- C. Only the regulator may complete the matrix
- D. Companies should leave the AI column blank
A. Job titles alone do not establish delegated powers — The matrix is an author proposal adapted to actual lawful authority. Treating it as a prescribed form is the opposite error.
3. A new service and a redesigned process launched together; case age fell. The sponsor credits the model. What does Chapter 17 say?
- A. The model caused it, since it was the new element
- B. Half the improvement can be attributed to the model
- C. Nothing can be credited until Months 7 to 12
- D. The improvement belongs initially to the complete service; crediting the model alone needs further evidence
D. The improvement belongs initially to the complete service; crediting the model alone needs further evidence — Attribution follows evidence. Splitting the credit in half is a guess presented as prudence.
Today’s Assignment
Assemble Sections 1 to 14 into one signed 90-Day Regulatory AI Plan for your organisation, with a one-page summary on which your leadership can take the acceptance decision.
- Assemble all fourteen sections in order, noting where a later session changed an earlier one. Write a one-page summary: the decision, its owner by role and the lawful authority behind it.
- Complete every field of the Appendix A pilot charter, including excluded actions, stop conditions and the tolerable error your organisation would set.
- Adapt the Appendix C matrix to your organisation's actual authority, mark any unclear delegation, and outline the decision record.
- Set out Days 1 to 15, 16 to 30, 31 to 60 and 61 to 90, with the evidence each review must see and the decision it must take.
- Choose the Appendix D measures you will report, with definitions and denominators, and write one benefits-ledger entry with capacity and cash apart.
- State the acceptance decision you will ask for (adopt within defined scope, revise or stop), with sign-off and conditions for Months 4 to 12. Sign and date, and name the commitment you are least confident your organisation will keep.
YOUR 90-DAY PLAN
The signed 90-Day Regulatory AI Plan.
How this is marked — the rubric
| Sections 1 to 14 assembled and consistent, with a one-page summary naming decision, owner and lawful authority, signed and dated | 20 |
| Appendix A charter complete, with excluded actions, stop conditions and a tolerable error set by the organisation | 20 |
| Appendix C matrix adapted to actual authority, unclear delegations marked, decision-record outline given | 15 |
| The ninety days stage by stage with the evidence and decision at each review | 20 |
| Appendix D measures with definitions and denominators, one ledger entry, and an acceptance decision with sign-off and Months 4 to 12 conditions | 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.FINAL EXAM
Your Final Exam is sat in this session. 15 questions, of which 3 are rebuilt from questions you got wrong on the practice papers. Sit it once your capstone is submitted.
Sit the Final Exam →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Plan assemblyPilot charteringDecision rightsStaged adoptionExecutive scorecardAcceptance decision