15-Session Programme · 25 Minutes Per Day · Free · For regulators and regulated companies in Nigeria’s petroleum industry
WEEK 2 OF 3 · INTELLIGENCE IN REGULATORY PRACTICE · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Part Three · Intelligence in Regulatory PracticeSESSION 6
Licensing and Approvals That Can Be Explained
Why This Matters
Week 1 laid the foundations; this week applies them to regulatory work, starting with the approval. An approval is where both sides of the relationship feel delay most and understand it least: the applicant does not know which question is unresolved, and the reviewer is pressed to decide while evidence is still missing. A transparent case record names the actual blocker. By the end you will be able to map the route for your own application class, keep completeness apart from adequacy, write a query that can be answered and set conditions that can be monitored.
25-Minute Module
0:00 – 4:00
Five questions inside one approval
- An approval contains several questions that should not be collapsed into one. Has the applicant supplied the required evidence? Is it internally consistent? Does the proposal satisfy the applicable technical and commercial requirements? Who has authority to decide? What conditions remain after the determination?
- AI can assist with the first questions; a lawful decision still depends on the whole process. The case record should distinguish a necessary technical query from an avoidable administrative repetition.
- THE DOCTRINE
A transparent case record should identify the actual blocker. The applicant and the reviewer both benefit from that clarity.
When someone says the file is with the regulator, ask which of the five questions is still open.
4:00 – 9:00
Completeness is not adequacy
- A document can be present and still be inadequate. A drawing can carry a title and signature but use the wrong design basis; a financial statement can be attached but cover an irrelevant entity.
- Completeness is the presence of required evidence. Adequacy is its professional sufficiency for the decision. An intake assistant should report presence and apparent relevance without certifying sufficiency.
- The book proposes five distinct interface labels: received, readable, relevant candidate, internally consistent and professionally assessed. The last needs an accountable reviewer. Avoid a single complete status that appears to settle the technical question.
- THE DOCTRINE
The words used by a system are part of its control design. A single complete status appears to settle a technical question that no accountable reviewer has assessed.
- For a company, the same labels discipline your own pack: compliance can confirm receipt and readability; only an accountable professional reviewer can mark the evidence professionally assessed.
Look at the status words on your own portal or tracker. One of them is probably promising more than anyone has checked.
9:00 – 14:00
Map the route, then worked process one
- For each application class, establish the statutory authority, reviewer roles, dependencies and escalation route. Where several approvals are involved, show their separate decisions; a combined schedule should not imply that one substitutes for another.
- Do not invent universal clocks. Legal staff validate the applicable timeline and any rules on additional information. Record receipt, requests, responses and decision events separately, so delay can be measured without a false assertion that an approval has automatically arisen.
- Worked process one is the book's proposed review structure for a refining decision, populated from current sources and the actual application. As Session 3 set out, sections 111(1) and 183 of the Petroleum Industry Act 2021 preserve the Minister's refining-licence role on the Authority's recommendation. So a recommendation may be complete while the final decision remains pending. Record them separately, beside the company activity affected.
- A new licence, renewal, modification and existing condition are different events; a route for the initial licence should not be assumed to govern every later change. If your decision is not a refining licence, apply the same route discipline to your own application class.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. In a fictional licensing department, the case owner records the necessary reviews and the next authorised step. Recommendation issued and final decision pending are two entries, not one. The institution's actual delegation controls who decides; the proposed role assignments only coordinate the work.
AT THE COMPANY · THE PARALLEL- At the company. In a fictional applicant, the project manager identifies the company gate affected, compliance and legal establish the route and engineering defines the scope and design basis. Commissioning targets and procurement commitments sit in the planning record with their own status. Same loop, seen from the other side.
If your project plan shows the approval as one bar, it is hiding at least one decision.
14:00 – 20:00
A checkable comparison and an answerable query
- AI can build a candidate evidence index at intake, compare revisions during review and help assemble a decision pack. It should not introduce a new fact: if a capacity figure is absent, the pack identifies the gap.
- AI output should carry source location, candidate value, extraction status and uncertainty. Engineering checks the comparison before it enters a query or decision pack. A candidate extraction is not the accepted design basis.
Record to connect Proposed comparison Responsible verification Process narrative Capacity and operating basis Process engineering Storage basis Dependency on the process change Relevant storage specialist Layout and drawings Revision and changed configuration Design control and engineering Supporting assessment Scope and revision assessed Assessment owner - This is the book's proposed review structure; actual identifiers and acceptance criteria come from the case.
- A good query states the issue, identifies the source or criterion, explains the missing evidence and specifies the response needed. It avoids provide all relevant documents where the gap is narrow, and it does not describe a suspected inconsistency as established misconduct.
- The company answers in the same structure: quote the issue, identify the response document and revision, explain the change and list affected dependencies. A large resubmission without a response index transfers the work back to the reviewer.
A query that cannot be answered in one targeted response is usually two queries, or none.
20:00 – 25:00
Authorise, monitor, measure, and your wrap
- The decision pack keeps facts, assessments, unresolved matters, authority and proposed conditions apart, and preserves disagreement. A status change requires the actual authorised event. The model's narrative is not that event.
- THE DOCTRINE
A condition should name the responsible party, the required evidence, the applicable trigger and the review route. A condition requiring compliance with everything adds little operational clarity.
- Measure preparation time, repeat queries, waiting by cause and later corrections. Time released in drafting is a task benefit. It does not establish faster final approval, lower payroll or avoided commercial delay without separate evidence.
- For the board: oldest unresolved cases by consequence, repeat-query patterns, completeness-assistant errors and conditions lacking closure evidence. I would ask whether the company and institution can identify the same outstanding decision from their records.
- Before you close: sketch the route for your Session 1 decision and mark which of the five labels each piece of evidence has truly reached. That is Section 6 of your 90-Day Regulatory AI Plan. Write your assignment now.
An intelligent approval earns its credibility from the reasoning and evidence behind the decision, not from the speed of the portal.
Key Terms
Completeness
Presence of required evidence, distinct from its professional adequacy.
Adequacy
Whether present evidence is professionally sufficient for the decision; it needs an accountable reviewer, not an intake check.
Case owner
The person coordinating a defined case without replacing specialist or lawful decision authority.
Decision route
For an application class: the statutory authority, reviewer roles, dependencies, escalation and any separate recommendation and final instrument.
Self-Check — Before You Submit
1. Which status labels does the book propose to keep completeness apart from adequacy?
- A. Submitted, complete and approved
- B. Received, readable, relevant candidate, internally consistent and professionally assessed
- C. Open, in progress and closed
- D. Draft, issued and accepted
B. Received, readable, relevant candidate, internally consistent and professionally assessed — The book proposes five distinct labels, with professionally assessed needing an accountable reviewer. The tempting first option contains a single complete status, which the book warns appears to settle the technical question.
2. A fictional refining applicant learns that the Authority's review and recommendation are finished. Its project director wants the dashboard to show the licence as granted so procurement can proceed. What is right?
- A. Show it as granted, because the technical work is done
- B. Remove the approval from the dashboard
- C. Record the recommendation as complete and the final decision as pending, beside the company activity it affects
- D. Ask the vendor to forecast the decision date
C. Record the recommendation as complete and the final decision as pending, beside the company activity it affects — Sections 111(1) and 183 preserve the Minister's refining-licence role on the Authority's recommendation, so a recommendation may be complete while the final decision remains pending. A local dashboard should not turn a recommendation or submission into permission to proceed.
3. An AI drafting tool halves the time officers spend writing review summaries. What can the institution claim from that alone?
- A. A task benefit in drafting time
- B. Faster final approvals
- C. Lower payroll cost
- D. Avoided commercial delay for applicants
A. A task benefit in drafting time — Time released in drafting is a task benefit. Faster final approval, lower payroll and avoided commercial delay each need separate evidence.
Today’s Assignment
Write Section 6 of your plan: the decision route, completeness kept apart from adequacy, an answerable query and conditions that can be monitored.
- Map the route for the application class or approval behind your Session 1 decision: exact decision requested, authority, reviewers, dependencies, escalation and any separate recommendation and final instrument. State no timeline; name who validates it.
- Take one real case and mark each key piece of evidence against the book's labels: received, readable, relevant candidate, internally consistent, professionally assessed. Name the accountable reviewer for the last.
- Rewrite one real query (if you are in a regulatory institution) or query response (if you are in a company) in the book's structure, with confidential details removed.
- List the conditions you would expect to monitor, each with responsible party, evidence, trigger and review route.
- The honest part: where your organisation currently treats receipt as acceptance or a recommendation as a decision. If no AI is in use yet, that is normal; work from the decision or proposal in front of you. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 6 — The decision route, completeness kept apart from adequacy, an answerable query and conditions that can be monitored.
How this is marked — the rubric
| Maps the exact decision, authority, reviewers, dependencies and escalation, keeping any recommendation and final instrument separate and inventing no clock | 30 |
| Applies the five labels to real evidence and names the accountable reviewer for professional assessment | 25 |
| Rewrites a query or response stating issue, source or criterion, missing evidence and response needed | 25 |
| States conditions with party, evidence, trigger and review route, with an honest example from your organisation | 20 |
| Total · pass mark 70 | 100 |
Submitting session 6
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: 6, 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
Decision-route mappingCompleteness versus adequacyAnswerable queriesMonitorable conditionsHonest approval measures
Part Three · Intelligence in Regulatory PracticeSESSION 7
Risk Based Supervision and Inspection
Why This Matters
No regulator can inspect everything, and no company can give every finding the same urgency. The method used to prioritise attention is itself a decision someone must explain. This session is about directing professional attention without letting a score replace inspection judgement, and about following a finding all the way to closure that someone has actually verified. A low incident count is not proof of a safe facility. By the end you will be able to set out a transparent prioritisation method, test it for the knowledge it lacks and define the record behind each step from observation to accepted closure.
25-Minute Module
0:00 – 5:00
Begin with consequence, not history
- Historical schedules may miss changing conditions; a model built on past findings may miss facilities rarely inspected. Risk based supervision needs a disciplined treatment of both observed risk and missing knowledge.
- Begin with consequence: potential harm, exposure, operating condition, known control weaknesses and confidence in the available evidence.
- THE DOCTRINE
A low incident count is not proof of a safe facility. It may reflect low exposure, poor reporting or the absence of an event so far, and a score should not hide those alternatives.
- For an early programme, a documented rule-based method may be more defensible than a complex predictive model: factors, their evidence and their effect on priority. Treat overdue critical actions and material changes as explicit triggers, and give inspectors a route to escalate a concern even when the overall score is low.
- The book's illustration: a facility with uncertain integrity records and a high-consequence service may deserve attention without a recent incident, while one with repeated minor documentation defects may need corrective administrative support instead.
Ask to see the rationale behind the top ten priorities. If there is only a number, there is no method yet.
5:00 – 11:00
What history cannot tell you
- Inspection selection bias, in plain terms: past findings record where inspectors looked and what they wrote down. A model trained only on inspected facilities may learn the institution's earlier choices rather than underlying risk. A region with better reporting can look more problematic simply because more is visible.
- Retain a planned sample outside the highest-ranked cases where appropriate, and compare findings. Record missingness as a factor with a reason. Unknown does not mean safe, and a missing record is not automatically severe wrongdoing.
- Outcome coverage asks how much of the population has a verified outcome at all. An uninspected facility does not become a verified low-risk example because no adverse finding is recorded.
- The rare-event accuracy trap. A model that predicts no events can appear accurate when events are rare. Illustratively, with assumed teaching inputs: if 2 of 100 facilities had a material event, a model that predicts none is right 98 times in 100, or 98%, and detects nothing. Test useful detection, false alerts and the real response capacity instead.
- Define the model's task first: predicting a later adverse finding and prioritising limited capacity are different tasks. Reconstruct only the information available at the ranking date. A score should not be described as a probability without evidence that this interpretation is valid for the deployment population.
When a vendor quotes accuracy for a rare event, ask how many events the model actually caught.
11:00 – 16:00
Worked process two: the evidence boundary
- Worked process two connects an observation to an authorised finding, corrective action and verified closure. Sections 7 and 8 of the Petroleum Industry Act 2021 provide upstream supervisory starting points; the actual inspection must use the applicable instruments, standards and approved procedures. The book supplies no universal engineering threshold, and neither does this session.
- Distinguish factual observation, technical inference and the authorised finding. Preserve photographs, documents and measurements with location, period and revision. A photograph of corrosion without a component identifier may not support the intended conclusion; a sensor alarm without calibration or operating context may mislead.
- AI can build a briefing with source links; the inspector verifies on site and separates observation from a hypothesis generated before the visit. Conflicting edits enter a resolution queue rather than being replaced by the newest timestamp.
- THE DOCTRINE
A priority score neither establishes a finding nor prescribes a sanction. The supervisory lead explains why a visit, desk review or other response is appropriate.
- Where urgent response is required, use the applicable emergency or escalation route. Draft generation must not become a prerequisite for it.
Ask of every photograph in a finding: which component, when, and who says so.
16:00 – 21:00
From finding to verified closure
- The loop closes only when the response is assessed: the reviewer determines whether that evidence resolves the specific finding, and whether a residual condition remains.
Proposed state Record supporting the state Observation captured Identified evidence and context Finding authorised Review basis and accountable determination Company action reported Issued response and supporting records Verification open Reviewer, criteria and unresolved question Closure accepted Authorised conclusion and accepted evidence - These five states are the book's proposal; an institution can use different labels if the distinctions remain clear. Status changes are driven by retained events. A narrative containing completed should not trigger closure.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. In a fictional inspection directorate, the case owner assigns a reviewer, who states which evidence resolves the finding. Earlier incomplete submissions are retained, because they explain elapsed time. Time awaiting company evidence is separated from time awaiting regulatory assessment.
AT THE COMPANY · THE PARALLEL- At the company. In a fictional operator, the maintenance system records work completion while compliance records acceptance of evidence. The two events are linked without implying they happened together. If the response is judged inadequate, the reason is kept and the next evidence task goes to the person able to act.
- A competent reviewer who did not attend the visit should understand why the finding was closed.
Closure is a judgement about the condition and the evidence, not a keyword.
21:00 – 25:00
Capacity to act, and your wrap
- Track every flag to disposition. A model can grow the queue of unresolved work faster than the institution can respond, and that backlog is part of the service's operating effect.
- THE DOCTRINE
A warning that nobody can act on is an incomplete service. Leadership must fund the response as well as the intelligence.
- Keep an urgent route independent of any scheduled ranking cycle. Preserve the professional's reason for an override: it is evaluation evidence, neither automatic proof of model failure nor proof that human judgement was correct.
- For the board: high-consequence coverage, overdue material actions, repeat findings, verified closure time and the condition of evidence at facilities not recently inspected, paired with the team's operational capacity.
- Before you close: write how attention for your Session 1 decision is prioritised today, and who would act on a flag. That is Section 7 of your 90-Day Regulatory AI Plan. Write your assignment now.
Before you buy a better ranking, count how many of last quarter's flags anyone actually acted on.
Key Terms
Risk ranking
Prioritisation for attention, distinct from a finding of breach.
Inspection selection bias
The distortion that arises because historical findings reflect where inspectors looked and what they recorded, not necessarily where risk lies.
Outcome coverage
How much of the population has a verified outcome; an uninspected facility with no recorded finding is not a verified low-risk example.
Verified closure
An authorised conclusion, based on accepted evidence, that the response resolves the specific finding; not a completion claim or keyword.
Self-Check — Before You Submit
1. A ranking model trained on past inspection findings places one region consistently at the top. What is the most important question to ask first?
- A. Whether the region's operators should be sanctioned
- B. Whether the region simply had more inspections and better reporting
- C. Whether the model can be made more complex
- D. Whether the other regions can be removed from the programme
B. Whether the region simply had more inspections and better reporting — Historical findings reflect where inspectors looked and what they recorded, so better reporting can make a region look more problematic. Sanction is the dangerous wrong answer: a priority score neither establishes a finding nor prescribes a sanction.
2. A fictional operator uploads a completion certificate stating that a corrosion finding has been completed. The status tool proposes closing the finding. What should happen?
- A. Close it, because the company has certified completion
- B. Close it and schedule a reminder
- C. Reject the certificate
- D. Move it to verification open, so a reviewer can assess whether the evidence resolves the specific finding
D. Move it to verification open, so a reviewer can assess whether the evidence resolves the specific finding — Company action reported and closure accepted are different states. Closure is a judgement about the condition and evidence, not a keyword.
3. An inspector overrides the model and sends a team to a low-scored facility. How should leadership treat the override?
- A. As evaluation evidence to examine against the accepted outcome and context
- B. As proof the model has failed
- C. As proof the inspector was right
- D. As a disciplinary matter, because the method was bypassed
A. As evaluation evidence to examine against the accepted outcome and context — An override is useful evaluation evidence, neither automatically proof of model failure nor proof that human judgement was correct.
Today’s Assignment
Write Section 7 of your plan: the prioritisation method, the evidence boundary and verified closure for your Session 1 decision.
- Describe honestly how attention for your decision is prioritised in your organisation today, then the transparent method you would use: factors, evidence, explicit triggers and a low-score escalation route.
- Explain how you would check selection bias and missing knowledge, including outcome coverage and where unknown is currently being read as safe.
- Show how your evidence boundary separates observation, inference and authorised finding, with the identity, location, period and revision each item needs.
- Set out the record that supports each state from observation captured to closure accepted, separating company action from verification.
- Name who acts on a flag and the response capacity behind it. If no AI is in use yet, that is normal; work from the decision or proposal in front of you. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 7 — Supervision and closure: the prioritisation method, the evidence boundary and verified closure.
How this is marked — the rubric
| Sets out a transparent method with factors, evidence, explicit triggers and a low-score escalation route | 25 |
| Explains how selection bias, outcome coverage and missing knowledge would be checked | 25 |
| Separates observation, inference and authorised finding and supports each closure state with a named record | 30 |
| Names who acts on a flag and the capacity behind it, with an honest example from your organisation | 20 |
| Total · pass mark 70 | 100 |
Submitting session 7
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: 7, 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
Transparent prioritisationSelection-bias awarenessEvidence boundariesVerified closureResponse capacity
Part Three · Intelligence in Regulatory PracticeSESSION 8
Measurement and Revenue Assurance
Why This Matters
Measurement is where regulatory evidence meets money. A difference between two quantities can affect commercial settlement and public revenue, and it can just as easily arise from timing, reference conditions or allocation as from a real loss. Leaders on both sides are tempted to read a discrepancy as leakage, or to wave it away. Both are technically weak. This session teaches you to define what can legitimately be compared, to keep competing explanations open until evidence settles them and to count each benefit once. Even if your organisation never handles a barrel, it depends on some reconciliation, and the same discipline applies.
25-Minute Module
0:00 – 5:00
A difference is a question, not a verdict
- A discrepancy can arise from timing, reference conditions, meter uncertainty, allocation, inventory movement or an actual loss. The regulatory task is to identify the difference, establish its basis and resolve it through evidence.
- The book states the instruments carefully. NUPRC's gazetted index includes upstream measurement regulations. The NMDPRA Petroleum Measurement Regulations 2023 address measurement and allocation of petroleum and related products within their stated scope. You establish the applicable instrument and boundary for your activity.
- THE DOCTRINE
Treating every discrepancy as leakage is technically weak. Treating every discrepancy as harmless is equally weak. Identify the difference, establish its basis and resolve it through evidence.
- AI fits as a reconciliation assistant: detecting inconsistent units, linking meter identifiers, flagging drift and helping explain a reconciliation already established by qualified staff.
When a report says revenue was lost, ask which of the six possible causes has been ruled out, and by whom.
5:00 – 11:00
The accounting boundary and the balance
- The accounting boundary says what enters, what leaves, what remains in inventory and what is consumed or transformed. Match periods and units.
- The book describes the liquid balance in words. Start with accepted opening inventory; add receipts and subtract dispatches for the same boundary and period, treating other verified movements explicitly. That gives an expected closing inventory, which is compared with the measured closing inventory. The difference is the residual.
Balance term (illustrative, assumed teaching inputs) Quantity, in one consistent assumed unit Accepted opening inventory 10,000 Plus receipts for the same boundary and period 50,000 Less dispatches for the same boundary and period 48,000 Expected closing inventory (10,000 + 50,000 − 48,000) 12,000 Measured closing inventory 11,850 Residual (11,850 − 12,000) −150 - These numbers are invented for teaching; the book gives no worked quantities. The arithmetic is correct and still says nothing about cause. The 150 may reflect reference conditions, a transfer crossing the period cut-off, calibration or allocation.
- THE DOCTRINE
A residual is a question for investigation. Its sign and magnitude alone do not establish loss, theft or an acceptable tolerance, and no generic percentage makes a difference acceptable.
A correct sum over the wrong boundary is still the wrong answer.
11:00 – 17:00
Worked process three: what can legitimately be compared
- Worked process three begins with an actual discrepancy and authenticated records. For upstream work, the Nigerian Upstream Petroleum Measurement Regulations 2023 provide selected anchors: regulation 3 the metering plan, regulation 6 measurement-system requirements, regulation 7 reporting and data exchange, and regulations 15 and 16 production records. Apply the correct instrument for the activity.
- Reference conditions, in plain terms: a quantity is always expressed under stated conditions, and two quantities are comparable only when expressed on the same basis, much as two prices are comparable only in the same currency. Comparing quantities under different conditions can create a difference that does not establish physical loss.
- Identify transfers crossing the cut-off and later corrections; as Session 4 taught, record the period described separately from receipt time.
Comparison layer Proposed record Verification needed Original measurement Source logs and meter identity Authenticity and validity Basis alignment Units, conditions and period Comparable measurement basis Adjustment Reason, formula and authorised inputs Accepted calculation basis Allocation Allocation method and dependencies Applicable approved treatment Reconciliation Original residual and accepted changes Explained and unresolved portions Consequence Issued review or settlement record What outcome is actually established? - This layering is the book's proposal. Use controlled computation for arithmetic; AI can propose an index or candidate explanation but should not invent a correction factor. Retain formula and input versions so another reviewer can reproduce the result.
Ask whether a specialist who did not build the spreadsheet could follow the reconciliation.
17:00 – 21:00
Competing explanations and the fiscal line
- Create an exception for each unresolved comparison. Timing, reference basis, calibration status, inventory movement and allocation answer different questions, so assign each to its competent owner and request the smallest set of evidence sufficient to examine it.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. In a fictional measurement unit, AI locates a later correction to a return. The reviewer checks that it concerns the same boundary and was authorised. The explanation stays a hypothesis until the evidence is accepted, and rejected alternatives are kept with the reason.
AT THE COMPANY · THE PARALLEL- At the company. In a fictional producer, the measurement team answers the specific exception rather than sending every historical production document. The original return, accepted adjustment and reconciled result are kept as separate related records. Neither side rewrites the original source.
- Keep fiscal calculation separate. The book offers no current tax-rate schedule and gives no fiscal advice. Use approved parameters with version, applicability and reviewer, and keep measurement, valuation, assessment, payment and dispute statuses distinct.
- An accepted measurement adjustment does not automatically establish a cash consequence; finance and relevant specialists determine the effect on the correct basis. A model must never silently change an issued return, settlement or determination.
An unexplained residual is not a crime and not a rounding error. It is an open item with an owner.
21:00 – 25:00
Count each benefit once, and your wrap
- Test extraction, basis comparison, arithmetic and explanation separately; a service can calculate correctly from the wrong period or meter.
- THE DOCTRINE
A record correction, an accepted commercial effect and received cash are different benefits. Count each evidenced benefit once, or revenue-assurance reporting becomes an inflated sum of unlike events.
- An assessment is different from collection; a corrected invoice is different from payment; a prevented future loss is different from a resolved historical discrepancy. For the board: unresolved reconciliations by materiality, age and cause, valid measurement coverage, overdue calibration or verification actions where relevant, and approved adjustments and confirmed collections reported separately.
- Before you close: write the boundary and the comparison layers for your Session 1 decision or the nearest reconciliation your organisation depends on. That is Section 8 of your 90-Day Regulatory AI Plan.
- Pop Quiz 2 opens now: five questions, closed book, on Sessions 6 to 8. Sit it before Session 9, then write your assignment.
My executive question is whether the review can explain the quantity on a defensible basis and name the remaining action.
Key Terms
Accounting boundary
What enters, leaves, remains in inventory and is consumed or transformed, for a defined product, period and unit basis.
Residual
The difference between expected and measured closing inventory; a question for investigation, not proof of loss or an acceptable tolerance.
Reference conditions
The stated basis on which quantities are expressed; quantities under different conditions cannot legitimately be compared without an accepted transformation.
Benefits ledger
A record keeping verified record correction, accepted commercial consequence and realised cash as separate benefits, each counted once with its basis.
Self-Check — Before You Submit
1. A monthly reconciliation for a fictional facility shows a negative residual. The finance director wants to report it as product lost. What is right?
- A. Report it as loss, because the sign is negative
- B. Treat it as acceptable if it falls below a standard percentage
- C. Open exceptions for timing, reference basis, calibration, inventory movement and allocation, each with its competent owner
- D. Write it off as measurement uncertainty
C. Open exceptions for timing, reference basis, calibration, inventory movement and allocation, each with its competent owner — A residual's sign and magnitude do not establish loss, theft or an acceptable tolerance. The percentage option is the tempting trap: the book says no generic percentage makes a difference acceptable.
2. Which regulations does the book cite as selected upstream anchors in the 2023 measurement regulations?
- A. Regulations 3, 6, 7, 15 and 16
- B. Regulations 20 to 24, 30 and 35
- C. Regulations 3 to 4 and 18 to 22
- D. Sections 111(1) and 183
A. Regulations 3, 6, 7, 15 and 16 — Regulation 3 concerns the metering plan, 6 measurement-system requirements, 7 reporting and data exchange, and 15 and 16 production records. The other options are anchors from the host-community regulations, the decommissioning regulations and the PIA.
3. A revenue-assurance dashboard adds together corrected returns, issued assessments and payments received into one recovered-revenue figure. What is the flaw?
- A. It should use a larger sample
- B. Payments should be excluded entirely
- C. Corrected returns are not evidence
- D. It counts unlike events as one benefit; record correction, accepted commercial effect and received cash need separate categories
D. It counts unlike events as one benefit; record correction, accepted commercial effect and received cash need separate categories — An assessment is different from collection, and a reporting adjustment is not necessarily recovered product or cash. Each evidenced benefit is counted once in its own category; excluding payments would drop a benefit the book says to report separately, as confirmed collections.
Today’s Assignment
Write Section 8 of your plan: the reconciliation behind your decision, with its boundary, comparable basis, competing explanations and a benefit counted once.
- Define the accounting boundary for your Session 1 decision, or the nearest reconciliation your organisation depends on: product or item, boundary, period, units and which quantities are measured, estimated and allocated.
- Set out the comparison layers from original measurement to consequence, with the verification each needs and who performs it.
- List the competing explanations for a discrepancy and the competent owner of each. Use no generic tolerance.
- Write the benefits-ledger treatment that keeps record correction, accepted commercial effect and received cash apart.
- The honest part: where your organisation currently reports an unverified difference as loss, or as harmless. Any numbers you use must be labelled as assumed. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 8 — The reconciliation: boundary, comparable basis, competing explanations and a benefit counted once.
How this is marked — the rubric
| Defines boundary, period, units and measured, estimated and allocated quantities | 25 |
| Sets out comparison layers to consequence with the verification each needs | 25 |
| Lists competing explanations with competent owners and no generic tolerance | 25 |
| Keeps record correction, commercial effect and cash separate, with an honest example from your organisation | 25 |
| Total · pass mark 70 | 100 |
Submitting session 8
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: 8, 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 2 — Approvals, Inspection and Measurement 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 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Accounting boundariesComparable measurement basisCompeting explanationsFiscal separationBenefit discipline
Part Three · Intelligence in Regulatory PracticeSESSION 9
Safety and Environmental Intelligence
Why This Matters
Safety and environmental supervision needs timely evidence from operations, inspections, incidents and corrective actions. AI can help connect weak signals, but detecting a pattern does not establish a cause or determine a safe response, and engineering controls and emergency procedures must work without the model. This session is about keeping evidence honest under pressure: provisional facts kept provisional, dissent kept visible, figures tied to a measurement basis and long-term obligations kept as separate records. Technology must not create a new delay at the point of danger. By the end you will know which signals should prompt which review, and which routes must never depend on a model.
25-Minute Module
0:00 – 5:00
Leading evidence and the escalation that cannot wait
- NUPRC's public gazetted index includes instruments addressing safety, environmental remediation, flaring, venting and methane. Applicability and the current controlled text require confirmation for the activity.
- Leading evidence is evidence that a condition may be developing before harm occurs: overdue integrity work, impairment of protective systems, repeated leaks, unresolved change-management tasks or poor closure of inspection findings. Its relevance depends on the asset and hazard.
- An increase in one indicator should prompt a defined review, not an automatic conclusion that an incident will occur. The supervisor asks whether the signal is credible, what condition it may indicate, what further evidence is needed and who can act.
- THE DOCTRINE
If the potential consequence is urgent, established escalation procedures take priority over completing an AI analysis. Technology must not create a new delay at the point of danger.
Ask your team to walk you through an urgent escalation with every AI tool switched off. If it stalls, fix that first.
5:00 – 10:00
Incident records without invented certainty
- An incident record should keep known facts, initial estimates, hypotheses and confirmed findings apart. Early reports are often incomplete. AI can organise them and identify inconsistencies, but must preserve provisional status; a later investigation may change the explanation while the original observations stay important.
- THE DOCTRINE
Do not let summarisation erase dissent or uncertainty. A neat narrative can weaken an investigation if it removes the evidence that does not fit.
- If one engineer raises a possible failure mechanism and another disputes it, the summary keeps the issue open for resolution.
- For recurring loss-of-containment evidence, connect component identity, service, operating conditions and maintenance history, and establish record quality and coverage before interpreting recurrence. AI can propose groupings. Specialists verify the affected population and competing explanations. A candidate pattern is an investigation lead, not a proven common cause.
When an incident summary reads too cleanly, ask what was left out of it.
10:00 – 15:00
Measurement basis and community reports
- Distinguish measured emissions, calculated estimates, model inference and reported activity data. Methane detection depends on instrument capability, operating conditions and observation coverage. A remote observation can indicate a plume without establishing an annual emissions total or the responsible source with certainty.
- Environmental specialists validate method, uncertainty and aggregation. Do not convert a model prediction into a compliance measurement without an accepted basis; where approved methods permit estimation, record the method and its assumptions.
- Reports from affected people can reveal conditions a remote system misses. Provide accessible routes to report, acknowledge and follow up, and protect personal information. AI translation or classification may help, but technical meaning and local language need competent review when the consequence is material.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional environmental desk receives a community report and a spike in online posts about the same site. The report is acknowledged, protected and routed for physical evidence. The online volume directs attention; it is not treated as a measure of harm.
AT THE COMPANY · THE PARALLEL- At the company. A fictional operator preparing its own environmental figure states whether it is measured, calculated or inferred, with method and assumptions. It does not quote a remote observation as an annual total, and it keeps its records of a community report linked to the follow-up.
- Social-media volume or sentiment is not a reliable measure of environmental harm; it may be biased, duplicated or manipulated. Physical evidence and the applicable investigation process establish what happened.
Before you quote an environmental number, be able to say whether it was measured, calculated or inferred.
15:00 – 21:00
Long-term obligations as separate records
- A producing asset's record should keep its future closure obligations, plans, financial support and unresolved uncertainties. The Nigerian Upstream Petroleum Decommissioning and Abandonment Regulations 2026 provide a current primary source: regulations 3 and 4 address the plan and its updating; regulations 18 to 22 address the asset database, fund, contributions, use and reporting subjects. These are selected anchors, not a complete statement of the regime.
- Connect each plan revision to its licence or lease, asset scope, cost basis and authenticated financial evidence. A company-wide fund indicator should not hide which plan or asset remains unresolved.
- Keep apart a plan estimate, an approved contribution, a bank balance, an authorised withdrawal and completed work. They answer different questions. A provision in financial accounts does not, by itself, establish the contents of the relevant escrow account.
Question raised Who establishes the answer Has a changed asset or method significantly affected the plan? Engineering Do cost and funding records support it? Financial specialists Does the change require a submission or other authorised action? Legal or regulatory staff Which provisions apply to our existing plan and reporting? The legal owner Did an obligation transfer with an asset interest? The authorised professional, from the approved transaction and legal basis - AI can compare the asset register with the accepted plan and propose scope exceptions. A similarity match should not assign liability or decide that an obligation transferred. If your activity carries no closure obligations, apply this to the nearest long-term commitment your organisation holds or supervises.
Ask to see the plan, the asset list and the fund evidence on one page, each with its own owner and status.
21:00 – 25:00
Measure action, not alarms, and your wrap
- For the board: overdue material safety and environmental actions, repeat incidents, verified closures, evidence quality and response time to credible urgent reports, with alert burden and missed concerns tracked beside them.
- THE DOCTRINE
Do not measure an AI programme by the number of alarms it creates. Measure whether appropriate action occurred in time and whether its outcome was verified.
- Before you close: list the leading signals relevant to your Session 1 decision and the escalation route that must work without the model. That is Section 9 of your 90-Day Regulatory AI Plan. Write your assignment now.
An alert that arrives after the escalation should have happened is a record, not a safeguard.
Key Terms
Leading evidence
Signals that a condition may be developing, such as overdue integrity work or impaired protective systems, each prompting a defined review rather than a conclusion.
Provisional status
The marking of an incident fact, estimate or hypothesis as not yet confirmed, preserved through every summary until investigation settles it.
Measurement basis
Whether an environmental figure is measured, calculated, inferred by a model or reported activity data, with its method, uncertainty and lineage.
Selected anchor
A provision the book cites as a starting point for workflow design, not a complete statement of the regime; applicability needs confirmation.
Self-Check — Before You Submit
1. A fictional operator's protective system shows impairment and the supervisor suspects an urgent hazard. The AI risk analysis will take several hours. What should happen?
- A. Wait for the analysis so the escalation is evidence-based
- B. Use the established escalation procedure now, and let the analysis follow
- C. Ask the model for a shorter analysis
- D. Log the signal for the next weekly review
B. Use the established escalation procedure now, and let the analysis follow — Established escalation procedures take priority over completing an AI analysis where the consequence may be urgent. Waiting for the analysis sounds rigorous, but it creates a new delay at the point of danger.
2. A remote observation indicates a methane plume near a fictional facility. What does it establish on its own?
- A. The facility's annual emissions total
- B. That the facility is the responsible source
- C. A compliance measurement
- D. That a plume may be present, prompting validation of method, coverage and source
D. That a plume may be present, prompting validation of method, coverage and source — A remote observation can indicate a plume without establishing an annual total or the responsible source with certainty. Converting it into a compliance measurement would need an accepted basis the observation alone does not provide.
3. Which items does the book say must be kept apart because they answer different questions about long-term closure?
- A. Plan estimate, approved contribution, bank balance, authorised withdrawal and completed work
- B. Plan, asset and licence only
- C. Fund balance and accounting provision, which can be treated as the same
- D. Only the plan revision and its date
A. Plan estimate, approved contribution, bank balance, authorised withdrawal and completed work — The book separates all five and warns that a provision in financial accounts does not, by itself, establish the contents of the relevant escrow account, which is why the third option is wrong.
Today’s Assignment
Write Section 9 of your plan: safety and environmental evidence for your Session 1 decision.
- Name the leading evidence relevant to your decision and the defined review each signal should prompt, with who can act.
- Describe the escalation route in your organisation that must work without the model, and where today it might wait on an analysis.
- Explain how provisional status and dissent survive summarisation in your incident or case records.
- State the measurement basis behind any environmental figure you rely on, and how community reports are received, acknowledged and protected.
- Where your activity carries long-term closure obligations, show how plan, asset scope, cost basis and funding evidence are kept as separate records with owners; otherwise apply this to your nearest long-term commitment. If no AI is in use yet, that is normal. Aim for 400 to 900 words, including the honest part: where a neat summary already hides uncertainty.
YOUR 90-DAY PLAN
Section 9 — Safety and environmental evidence: leading signals, escalation, provisional status, measurement basis and long-term obligations.
How this is marked — the rubric
| Names leading signals with a defined review and an escalation route that works without the model | 30 |
| Shows how provisional status and dissent are preserved through summaries | 20 |
| States the measurement basis of relied-on figures and a protected route for community reports | 25 |
| Keeps plan, asset scope, cost basis and funding evidence as separate owned records, with an honest example from your organisation | 25 |
| Total · pass mark 70 | 100 |
Submitting session 9
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: 9, 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
Leading evidenceModel-independent escalationProvisional statusEnvironmental measurement basisLong-term obligation records
Part Three · Intelligence in Regulatory PracticeSESSION 10
Host Communities and Accountable Development
Why This Matters
Community development joins financial accountability, project execution, consultation and the lived experience of beneficiaries, and oversight must keep reported activity apart from verified outcome. A dashboard can show money spent while a project remains unusable; a summary can describe a successful meeting while omitting a material objection. This session applies the week's discipline to accountable delivery, for a regulator supervising it and for a company funding or depending on it. Approved, paid, delivered, verified and operating are different milestones. By the end you will be able to report each separately, make consultation inspectable and route a financial or delivery flag to competent review without accusing anyone.
25-Minute Module
0:00 – 4:00
The legal anchor, stated exactly
- The PIA's host-community chapter addresses development trusts and related governance and reporting. For the applicable upstream settlor contribution, section 240(2) of the Petroleum Industry Act 2021 specifies 3% of actual operating expenditure of the preceding financial year in respect of relevant upstream operations.
- THE DOCTRINE
Section 240(2) is not a universal charge on every petroleum business. Applicability and the computation need qualified confirmation, and the statutory percentage does not supply the expenditure amount or settle a disputed base.
- Illustratively, with an assumed teaching input: if a qualified review accepted a relevant upstream operating expenditure of ₦1,000,000,000 for the preceding financial year, 3% of it is ₦30,000,000. Whether the rule applies, and what the base is, require authenticated company records and qualified review.
- If scope or an input is disputed, preserve that status and assign it to the appropriate specialist. The model should not choose a convenient value to complete the calculation.
When a contribution figure is presented, ask who confirmed applicability and who accepted the base.
4:00 – 9:00
The trust's evidence chain
- Link the relevant obligations to the trust's governance records, approved plans, financial information and project evidence. Keep the roles of settlor, trustees, management and advisory functions distinct.
- Worked process four connects statutory subjects to funding, delivery and service evidence. Its anchors are PIA sections 240 and 251 to 255, and the Nigeria Upstream Petroleum Host Communities Development Regulations 2022, including regulations 20 to 24, 30 and 35. Regulation 30 provides an upstream supervisory reporting anchor through quarterly-return subjects. Establish applicability and current text before implementation.
- A project record identifies the approved purpose, beneficiaries, location, budget, scope, contractor, milestones, delivery evidence and any change with its reason.
- THE DOCTRINE
No platform should give one user unrestricted power to approve plans, authorise spending, certify delivery and close every concern. It should reflect actual mandates and approved access.
- A photograph of construction is supporting evidence. It does not alone establish completion, serviceability or benefit to the intended community.
Ask which single role in your system could approve, pay and certify the same project. If one can, that is your first control gap.
9:00 – 15:00
Funding, spending, delivery and service
- The book's example is a water project. Review may need the actual equipment, power arrangements, accepted test results, commissioning evidence, maintenance and an accountable operating owner. If the pump is not commissioned, the project cannot be described as providing water merely because most money has been spent.
Reporting question Record to examine Responsibility to establish Was funding received? Calculation, transfer and receipt Finance and authorised trust roles Does spending support approved work? Budget, contract and payment evidence Financial and project reviewers Was the milestone achieved? Accepted technical or service evidence Competent reviewer Can the service operate? Operating and maintenance arrangements Accountable service owner Are concerns addressed? Issue, response and follow-up records Applicable engagement roles - This is an author-proposed review matrix; actual powers and prescribed returns remain governed by their controlling sources. A missing attachment is an evidence question, not automatic proof of improper spending.
- INSIDE THE REGULATOR · THE WORKED CASE
- Inside the regulator. A fictional supervisory desk reviews a quarterly return that reports a project at high payment progress. It asks for the milestone and service evidence behind the narrative, and records missing coverage, confirmed operation and confirmed interruption as separate states.
AT THE COMPANY · THE PARALLEL- At the company. A fictional settlor's community relations lead reports funding, spending, physical progress and service readiness as four statuses, each with its record and responsible role. Maintenance, power and replacement arrangements are examined, not assumed. An unrecorded period is not reported as successful service.
A payment percentage cannot establish that a service operates. Ask for the commissioning record.
15:00 – 20:00
Inspectable consultation, and review without accusation
- Record how issues were raised, who participated, what was proposed and how decisions addressed them. Attendance is not agreement. Keep material minority concerns visible, and provide a route for people who cannot use the portal or lack reliable connectivity.
- AI can transcribe and organise issues, but must not invent consent or simplify a contested discussion into unanimous support. Qualified reviewers check important translations and summaries against the source, and information that could expose a complainant to retaliation is protected.
- Analytical tools can flag duplicate invoices, unusual payment patterns or costs that differ from a comparator. Those flags are hypotheses for review; they do not prove corruption. The reviewer needs the contract, scope, location, procurement basis and payment evidence. Comparators must be relevant: a remote project may have a different cost basis.
- THE DOCTRINE
A model should not attribute sabotage or misconduct to a community based on proximity, online sentiment or a pattern of complaints. Any adverse determination requires the appropriate facts, lawful process and authorised judgement.
- Keep physical-security information apart from development-benefit analytics; combining them casually can create an opaque surveillance system.
When a flag appears, write down the competent reviewer's name before anyone writes down a conclusion.
20:00 – 25:00
Transparency with protection, and your wrap
- Publish useful aggregate information and accessible project status where lawful and appropriate, without exposing personal data, sensitive allegations or detailed security information. The NUPRC HostComply service is publicly described as supporting trust reporting, monitoring and data management. Its existence shows a digital administrative channel, not evidence that every trust is compliant or that any AI use proposed here is deployed.
- Evaluate the AI contribution through action: relevant gap identified, responsible task assigned, evidence supplied, authorised review and retained outcome. Claims about wider public-health improvement need their own evidence; they cannot be inferred from spending, photographs or sentiment analysis.
- For the board: verified service outcomes, overdue projects, unexplained financial differences, unresolved commitments, complaint response and participation-record quality. Ask whether the data represents the intended beneficiaries, rather than only those most able to submit a form.
- Before you close: trace one commitment behind your Session 1 decision and report its four statuses separately. That is Section 10 of your 90-Day Regulatory AI Plan, and it completes Week 2.
- Practice Exam 2 opens now: fifteen questions, closed book, on Sessions 1 to 10, closing Week 2. Sit it before Session 11, then write your assignment.
Community intelligence should improve accountability to people as well as accountability for money.
Key Terms
HCDT
Host Communities Development Trust, with applicability established under the relevant framework.
Service readiness
Demonstrated ability of a delivered project to operate against its established criteria, distinct from funding received, money spent or physical progress.
Inspectable consultation
A record of how issues were raised, who took part, what was proposed and how decisions addressed them, with material minority concerns kept visible.
Automated collective blame
Attributing sabotage or misconduct to a community from proximity, sentiment or complaint patterns; the book rules it out as a substitute for evidence.
Self-Check — Before You Submit
1. Which statement about section 240(2) of the Petroleum Industry Act 2021 matches the book?
- A. It imposes 3% of revenue on every petroleum business
- B. For the applicable upstream settlor contribution, it specifies 3% of actual operating expenditure of the preceding financial year in respect of relevant upstream operations
- C. It fixes the expenditure base that companies must use
- D. It applies to midstream and downstream licensees by default
B. For the applicable upstream settlor contribution, it specifies 3% of actual operating expenditure of the preceding financial year in respect of relevant upstream operations — The book states the rule exactly and warns that it is not a universal charge on every petroleum business. The percentage does not supply the expenditure amount or settle a disputed base, so the third option is also wrong.
2. An analytics tool flags a fictional trust's contractor payments as unusually high against a comparator. What should the case owner do?
- A. Publish the flag to show transparency
- B. Report the trust for corruption
- C. Route it to competent review with the contract, scope, location, procurement basis and payment evidence, checking the comparator is relevant
- D. Suspend all of the trust's projects
C. Route it to competent review with the contract, scope, location, procurement basis and payment evidence, checking the comparator is relevant — Payment flags are hypotheses for review, not proof of corruption, and a remote site may legitimately have a different cost basis. Publishing an untested allegation would expose people to harm the book warns against.
3. A fictional water project shows most of its budget paid and photographs of completed construction. The pump has not been commissioned. How should it be reported?
- A. As delivering water, because construction is complete
- B. As a failed project
- C. As funded only
- D. With funding, spending and physical progress reported separately, and service readiness not yet demonstrated
D. With funding, spending and physical progress reported separately, and service readiness not yet demonstrated — A payment percentage cannot establish that a service operates, and a photograph does not alone establish completion or serviceability. Calling it failed would be equally unsupported; the record states what its evidence establishes.
Today’s Assignment
Write Section 10 of your plan: accountable delivery for a commitment your organisation makes, funds, supervises or depends on.
- Trace the evidence chain for one commitment: for a regulatory institution, a supervisory or reporting question; for a company, a host-community or other development commitment linked to your Session 1 decision.
- Report funding, spending, physical progress and service readiness as separate statuses, with the record and responsible role for each.
- State how consultation and minority concerns are made inspectable in your organisation, including for people who cannot use a portal.
- Show how a financial or delivery flag would be routed to competent review without accusation, and what information is protected.
- The honest part: where your organisation currently reports spending or activity as if it were delivery. Any contribution figures must be labelled as assumed. If no AI is in use yet, that is normal. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 10 — Accountable delivery: funding, spending, delivery and service kept distinct, and consultation made inspectable.
How this is marked — the rubric
| Traces the evidence chain from obligation or commitment to service, with distinct roles | 25 |
| Reports funding, spending, physical progress and service readiness separately, each with record and responsible role | 30 |
| Makes consultation and minority concerns inspectable, with an offline route | 20 |
| Routes a flag to competent review without accusation or collective blame, with an honest example from your organisation | 25 |
| Total · pass mark 70 | 100 |
Submitting session 10
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: 10, 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 2 — Foundations and Intelligence in Regulatory Practice 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 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Exact legal anchorsDelivery versus expenditureInspectable consultationReview without accusationProtected transparency