15-Session Programme · 25 Minutes Per Day · Free · For leaders in Nigerian crude, petroleum-products and gas trading
WEEK 2 OF 3 · CRUDE, PRODUCTS, GAS AND PHYSICAL EXECUTION · SESSIONS 6–10 · WORK AT YOUR OWN PACE
Part Two · Evidence and IntelligenceSESSION 6
The Contract Copilot
Why This Matters
In Session 5 you asked what your decision needs to know about the future. Today you ask what it has already promised. Every pricing date, tolerance, notice period and payment condition behind your live decision sits in a contract, and a contract assistant can now read that contract faster than any analyst. Speed is not the risk. The risk is a fluent answer drawn from the wrong version, an exception left out, or a delivery term read as if it settled title. A contract copilot is useful when its omissions are measured as carefully as its answers. By the end you will know what it must extract, what it must never infer, and who accepts its work before a number moves.
25-Minute Module
0:00 – 5:00
Extract obligations, not summaries
- Start with what the assistant must pull out of the agreement: the pricing basis, quantity tolerance, delivery terms, payment terms, title provisions, documentation requirements, notice periods and claims conditions. For each field, keep the source page and passage.
- Separate three kinds of output: an explicit term the contract states, an interpretation someone has drawn from it, and missing information the contract does not supply. A summary that blends the three cannot be reviewed.
- Retrieval-augmented generation is the plain idea of letting a language model write only after a search has fetched the relevant evidence. The original research by Lewis and colleagues (2020) supports the method, not a guarantee of correct contract interpretation.
- THE DOCTRINE
A document-grounded answer can still omit an exception, retrieve the wrong version or cite a passage that does not support its conclusion.
- Apply it to your Session 1 decision: the terms that govern it live in a signed agreement, an amendment or correspondence. Name where.
Ask the team to show you the passage behind the number, not the paragraph about it.
5:00 – 10:00
Version and access before fluency
- Filter documents before retrieval by authorised user, entity, trade, accepted status and relevant period. A draft with a better filename should not outrank the accepted agreement.
- The reviewer should see the extracted field, its exact evidence location, any competing clauses and any unresolved qualification. A fluent paragraph with a confidence score is insufficient when it changes a pricing date or payment condition.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. A fictional product importer's operations team plans receipt against the quantity tolerance and delivery window the assistant extracted from the signed agreement. A later executed amendment changed the window, and retrieval did not bring it back. The receiving plan now rests on a superseded term.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The same omission reaches the pricing window, the exposure risk measures and the payment date treasury funds against. Four views each look consistent and all disagree with the contract. The explanation Session 3 demands is a version failure, not a market move.
- Session 4's point applies: an amendment's printed date can precede its receipt or acceptance. Use the evidence of acceptance.
When an answer is fluent, ask which version it read.
10:00 – 14:00
Delivery terms do not settle title
- ICC's Introduction to Incoterms 2020 distinguishes delivery, risk and costs and states that the rules do not determine transfer of title. It also excludes payment terms and several other sale-contract matters from their scope. The actual contract and applicable law determine the transaction treatment.
- So the assistant must extract the actual contract provisions rather than infer ownership from a three-letter delivery term. A contract can use an incorporated trade term and separately address title, quality acceptance and payment.
- THE DOCTRINE
Where the text conflicts, preserve the conflict for legal review. Do not invent an order of precedence unless the accepted agreement or applicable legal analysis establishes one.
- An inferred title is a guess entered as a fact. Contract interpretation remains with your accepted legal and commercial reviewers; this session is not legal advice.
When someone says the delivery term settles it, ask to see the title clause.
14:00 – 20:00
Measure omissions, and know the stack
- Build an expert-reviewed reference set covering clean scans, poor scans, amendments, tables, exceptions and missing clauses, plus the hard cases: negative clauses, alternative pricing windows, handwritten amendments, contradictory dates and instructions hidden in attachments.
- Score critical fields separately. An overall extraction accuracy of 98% can conceal systematic error in the 2% of fields governing claims deadlines. Ninety-nine correct descriptive fields cannot compensate for one wrong beneficiary or pricing basis.
Stage What it does What a leader asks Optical character recognition Extracts text and page coordinates from scanned documents Can the reviewer open the cited page? Schema-constrained extractor Proposes fields such as product, quantity basis, pricing window, due date and amendment reference Are critical fields scored separately? Retrieval Finds the relevant accepted evidence the user is permitted to see Did all material evidence appear among the retrieved passages? Language model Drafts an explanation from that evidence Are omitted obligations scored? Deterministic calculation service Computes the price or deadline under an approved specification Is the arithmetic reproducible? - For retrieval, compare keyword search, sentence embeddings that match meaning rather than words, and a hybrid. Sentence-BERT (Reimers and Gurevych, 2019) is a published embedding architecture; it supports the approach, not correct reading of your documents. Exact cargo references and clause numbers often reward keywords; paraphrased obligations can reward meaning. Chunk by clause and keep the amendment hierarchy and cross-references.
- A self-reported language-model confidence is not a calibrated probability of correctness. Selective prediction research (Geifman and El-Yaniv, 2017) treats abstention as a risk-versus-coverage trade-off; here that means an explicit review route for uncertainty. If an amendment is missing, the correct output can be an incomplete answer naming the missing evidence. A plausible but invented document identifier is a failure even if the sentence is correct.
Ask for the critical-field error before the overall accuracy figure.
20:00 – 25:00
The acceptance record, and your wrap
- Require a contract-to-trade acceptance record before extracted terms drive any calculation. Name the commercial, legal and operations reviewers appropriate to the terms. For each field store value, unit, evidence span, document version, extraction status and reviewer acceptance.
- THE DOCTRINE
A contract copilot may propose a commercial summary and a review queue. It should not acquire trading authority through a tool call, and it must not become an invisible party to the agreement.
- The workflow: compare extracted terms with the trade record, route discrepancies to the owner, obtain acceptance, then call the price engine. On the Autonomy Ladder this is drafting under review, not execution, and it can begin on one agreement type.
- Before you close: list the terms your Session 1 decision depends on and where each lives today. That is Section 6 of your 90-Day AI Trading Plan. Write your assignment now.
If nobody signs the acceptance record, the assistant has effectively signed it for you.
Key Terms
Contract-to-trade acceptance record
The reviewed values, evidence version and named commercial, legal and operations reviewers that must exist before extracted terms drive a calculation.
Retrieval-augmented generation
A language model that writes from evidence a search has retrieved. It supports grounding, not a guarantee of correct interpretation.
Critical field
A contract term whose error changes money, exposure or obligation, such as a pricing basis, beneficiary or claims deadline. Scored separately from overall accuracy.
Abstention
A deliberate refusal to answer when evidence is insufficient, routed to a reviewer with enough context to finish the task.
Self-Check — Before You Submit
1. A fictional crude exporter's contract copilot reports 98% overall extraction accuracy. The detail shows its errors cluster in claims deadlines. What should the head of commercial do?
- A. Approve it, because 98% is high
- B. Score critical fields separately and hold approval until claims deadlines meet their own standard
- C. Ask for a confidence score on every field
- D. Replace the poor scans in the reference set with clean ones
B. Score critical fields separately and hold approval until claims deadlines meet their own standard — An overall 98% can conceal systematic error in the 2% of fields governing claims deadlines. A per-field confidence is tempting, but a self-reported confidence is not a calibrated probability.
2. What does ICC's Introduction to Incoterms 2020 say about title?
- A. Title passes when risk passes
- B. The delivery term fixes when title passes
- C. The rules cover title and payment together
- D. The rules distinguish delivery, risk and costs and do not determine transfer of title
D. The rules distinguish delivery, risk and costs and do not determine transfer of title — The rules do not determine transfer of title and exclude payment terms and other sale-contract matters. The tempting error is to read ownership from a three-letter delivery term; the assistant must extract the actual title provisions.
3. In evaluation, the copilot cites a document identifier that does not exist, but the sentence it supports is correct. How is that scored?
- A. As a pass, because the sentence is right
- B. As a minor formatting issue
- C. As a failure
- D. As a pass if the stated confidence was high
C. As a failure — The reviewer must be able to open the cited page and see the supporting passage. Generating a plausible identifier is a failure even when the sentence happens to be correct, because the next one may not be.
Today’s Assignment
Write Section 6 of your plan: the contract terms your Session 1 decision depends on and the contract-to-trade acceptance record that governs them.
- List the terms your decision depends on in your company: pricing basis, quantity tolerance, delivery, payment, title, documentation, notice periods and claims conditions. Mark any that do not apply and say why.
- Say where each lives, which version is accepted and who may access it.
- Name the critical fields that must be scored separately, and the hard cases a reference set drawn from your own documents must include.
- Design the acceptance record: the fields stored, the commercial, legal and operations reviewers by role, and the point at which the price engine is called.
- State what the copilot may never do in your business.
- The honest part: where your company relies on someone's memory of a contract, or lets a delivery term stand in for title. Describe structure; no confidential terms, prices, counterparty names or bank details. If no AI is in use yet, that is normal; the first step may be a reviewed extraction template. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 6 — The contract terms our decision depends on and the contract-to-trade acceptance record.
How this is marked — the rubric
| Lists the decision's terms with location, accepted version and access rules for each | 25 |
| Names critical fields scored separately and hard cases drawn from the company's own documents | 25 |
| Designs an acceptance record with reviewers by role and the point the price engine is called | 30 |
| States what the copilot may never do and gives an honest example from the company | 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
Contract extraction disciplineVersion and access controlTitle versus delivery termsCritical-field evaluationAcceptance before calculation
Part Three · Crude and Petroleum ProductsSESSION 7
Crude Pricing, Differential Exposure and Cargo Quality
Why This Matters
Session 6 secured the terms. Today you follow the price those terms create and the quality that decides whether the cargo is worth buying. A crude price is rarely one number: it is a formula that keeps moving until its last pricing day, and a cheap cargo can be expensive to use. If your business trades gas or serves the trade from treasury, credit or legal, apply this to the nearest equivalent decision or to the counterpart you depend on. Price is a formula with a clock; quality is a measurement with an evidence chain. AI can sharpen both comparisons. It cannot fix a price with hindsight or release a cargo.
25-Minute Module
0:00 – 6:00
Price is a formula with a clock
- A crude price may be a benchmark average over an agreed window, plus or minus a differential, the agreed premium or discount to that benchmark. Record every pricing day, weight, calendar treatment and adjustment. A position can be physically fixed while part of its price remains unfixed.
Fictional purchase, assumed inputs Value Quantity and window 100,000 barrels; five pricing days, equally weighted Days fixed and open Three fixed, two unfixed Assumed move Each of the two open days rises $1 per barrel Change in the five-day average 2 of 5 days × $1 = $0.40 per barrel Change in purchase cost $0.40 × 100,000 = $40,000 - That assumes equal weights, unchanged quantity and no other linked exposure. The exposure depends on which observations remain open. A hedge with a different settlement window may leave timing basis, the gap between when the hedge prices and when the cargo prices. Show that mismatch explicitly.
- THE DOCTRINE
Use a controlled formula engine. A language model can explain the formula or identify missing terms, but the accepted calculation should be reproducible, with its input-price snapshot and provisional or final status retained.
Ask how much of the price is still open, and until when.
6:00 – 11:00
Compare on the same basis
- Netback is what a sale is worth back at your starting point after differential, freight, insurance, handling, quality adjustments, financing and other applicable costs. Compare at the same valuation time and quantity basis, keeping confirmed costs apart from estimates.
Assumed teaching prices, 500,000 barrels Destination A Destination B Price per barrel $80.00 $81.20 Transport $2.00 $3.10 Handling $0.40 $0.60 Netback before other costs $77.60 $77.50 Difference across the cargo $0.10 × 500,000 = $50,000 to A Reversed by any extra financing or delay cost above $50,000 - AI may estimate destination demand, congestion or differential behaviour; the executive should see the binding constraint and the assumption most likely to reverse the decision. Do not extrapolate a grade relationship merely because two crudes are both described as light; where history is sparse, use bounded scenarios and say so.
- THE DOCTRINE
Reconcile forecast and realised netback by price, differential, freight, delay, quality and financing. Distinguish an improved decision from a market move that would have benefited the baseline strategy as well.
A small netback advantage is one cost line away from reversing. Ask which line.
11:00 – 15:00
What to ask of a differential model and a fixing adviser
- The engine computes the contract price from accepted observations; a forecasting model estimates only observations not yet published, or a proposed future differential. It cannot replace fixed observations because a new forecast looks more attractive.
- A differential challenger can be a regularised regression or a gradient-boosted tree model on as-of inputs. Ask what the label is: transaction labels may contain negotiation effects and selective disclosure. Keep quoted indications, executable bids and completed deals separate.
- Where grades or routes have few observations, partial pooling learns across related groups while allowing each its own offset. An unseen grade or destination carries a scope warning and wider review.
- Price-fixing actions are precise: fix now, defer within the window, or fix an authorised fraction. The optimiser must use only information observable when each action is taken; a simulation that sees the full window will overstate value. Backtest against a permitted neutral policy on incremental realised contribution and its downside, not on whether the model guessed tomorrow's direction.
If the backtest knew how the window ended, so did its profit.
15:00 – 21:00
A cheap cargo can be expensive to use
- Crude value depends on the buyer's facilities, feedstock range, yields and constraints; product value on specification, contamination history, handling and permitted sale channels. The US Energy Information Administration explains that crude characteristics and refining processes affect outputs; that does not supply a Nigerian refinery's yield model.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. Record sample identity, sampling basis, laboratory method, result, certificate, batch and release authority. A quality prediction should not overwrite an accepted test result. A suspicious certificate triggers verification with the issuer, not a synthetic replacement generated to complete the pack.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. A model of refinery economics is not automatically a model of trading resale value; for a merchant trader, accepted buyer bids and contractual penalties may be the target. Predict yield, energy cost, quality penalty and uncertainty separately so the commercial owner can challenge each.
- A teaching blend: component A carries an assumed 10 units of a property, B carries 30. A 50:50 blend by mass gives 20 under a linear mixing assumption; against a limit of 18, B may be at most 0.40. Some properties mix nonlinearly, so this does not establish that a real fuel is acceptable.
- THE DOCTRINE
A fast surrogate does not certify compatibility, quality or safe processing. AI can suggest candidates for evaluation while the laboratory and technical authorities control acceptance.
- A surrogate is a fast model imitating a slower process or economics model over a stated range. Candidates the book names include regularised regression, gradient-boosted trees and a Gaussian-process surrogate for small, carefully defined datasets. A buyer's reported yield can reflect configuration and operating choices as well as crude quality.
Ask which number on the quality screen was measured and which was predicted.
21:00 – 25:00
Keep the optimiser honest, and your wrap
- An optimiser may find exactly the combinations where the surrogate is least trustworthy. Limit candidates to an accepted domain, flag distance from training evidence, send novel blends to physical review, and perturb assay inputs within measurement uncertainty.
- A high-margin candidate with almost no specification slack can be less useful than one that survives realistic uncertainty.
- Before authorising a cargo-selection model, ask which properties determine rejection, which are predicted and which must be measured. The first pilot should produce better candidate comparisons, not automatic quality release.
- Before you close: write the price formula behind your Session 1 decision, or its nearest equivalent, with its clock. That is Section 7 of your 90-Day AI Trading Plan. Write your assignment now.
The optimiser is honest about its objective. Make sure the objective is honest about the model.
Key Terms
Unfixed portion
The part of a formula price whose pricing observations have not yet occurred. Its size, not the whole cargo, sets the open price sensitivity.
Netback
The value of a sale back at the starting point after differential, freight, insurance, handling, quality, financing and other applicable costs, on the same valuation time and quantity basis.
Timing basis
The risk left when a hedge settles over a different window from the cargo's pricing window.
Domain of use
The documented range of inputs and operating conditions over which a surrogate has been tested. Outside it, results go to physical review.
Self-Check — Before You Submit
1. In the book's fictional five-day, equally weighted pricing window on 100,000 barrels, three days are fixed. If only the two open daily prices each rise $1, what happens to purchase cost?
- A. It rises $100,000
- B. It rises $40,000
- C. It rises $20,000
- D. It is unchanged, because the cargo is physically fixed
B. It rises $40,000 — Two of five days moving $1 raises the average $0.40, or $40,000 on 100,000 barrels. The tempting $100,000 treats the whole price as open.
2. A vendor's price-fixing adviser shows strong backtested gains. What is the first question to ask?
- A. Did the optimiser use only information observable when each fixing action was taken?
- B. How often did the model guess the next day's direction?
- C. Can the model fix the whole window at once?
- D. Was it trained on a neural network?
A. Did the optimiser use only information observable when each fixing action was taken? — A simulation that sees the full future window will overstate value. Direction-guessing is the tempting measure, but the book's measure is incremental realised contribution and its downside against a permitted neutral policy.
3. A fictional refinery offtaker's blend optimiser proposes a high-margin blend near the edge of its surrogate's training evidence, with almost no specification slack. What should the head of trading do?
- A. Approve it, since the surrogate predicts it is within limits
- B. Widen the specification limit to create slack
- C. Retrain the surrogate using the recommended blend
- D. Send it to physical review and compare it with a lower-margin candidate that survives realistic uncertainty
D. Send it to physical review and compare it with a lower-margin candidate that survives realistic uncertainty — Optimisers find where a surrogate is least trustworthy. Approving on the prediction is the trap: model feasibility and physical certification are distinct decisions.
Today’s Assignment
Write Section 7 of your plan: the price formula behind your Session 1 decision with its clock, the same-basis comparison and the quality evidence the decision needs.
- Write the price formula behind your decision in your company, or its nearest equivalent, with every pricing day, weight and adjustment, and the latest time each portion can change.
- Show the fixed and unfixed portions and the open sensitivity on numbers labelled assumed; no confidential prices.
- Compare the alternatives on the same valuation time and quantity basis, and name the assumption most likely to reverse the decision.
- Set out which quality properties determine rejection, which may be predicted and which must be measured, the evidence chain to release authority, and any model's domain of use.
- The honest part: where your company has credited a favourable market move to a decision, or treated a predicted quality as measured. If no AI is in use yet, that is normal; the first need may be a controlled formula engine. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 7 — Our price formula and its clock, the same-basis comparison and the quality evidence our decision needs.
How this is marked — the rubric
| Writes the formula with its clock, fixed and unfixed portions and the open sensitivity, numbers labelled assumed | 30 |
| Compares alternatives on the same valuation time and quantity basis and names the reversing assumption | 25 |
| Separates rejection properties into predicted and measured, with the evidence chain and release authority | 25 |
| States a domain of use for any model and gives an honest example from the company | 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
Price formula disciplineUnfixed exposureSame-basis netback comparisonQuality evidence chainSurrogate boundaries
Part Three · Crude and Petroleum ProductsSESSION 8
The Delivered Margin and Inventory Allocation
Why This Matters
Session 7 priced the cargo. Today you ask what it actually earns once it is delivered, funded and sold from stock that is genuinely available. A comfortable price spread can disappear in one movement of the sale price, a currency you have not yet bought can change the conclusion, and a positive stock balance can describe product you cannot sell. The useful economic unit is the actual accepted delivery, with a consistent quantity basis and a complete cost boundary. If your business does not distribute products, apply this to the nearest equivalent economic unit you use. Pop Quiz 2 opens at the end of the session.
25-Minute Module
0:00 – 6:00
The delivered basis determines the margin
- Quoted price differences do not establish realised contribution. Record purchase price, saleable quantity, transport, storage, handling, financing, credit cost and applicable charges.
Assumed teaching cargo Value Purchased 10,000 tonnes at $700 per tonne Freight and handling, financing, other variable costs $35 + $8 + $2 = $45 per purchased tonne Saleable quantity after an assumed 0.3% difference 9,970 tonnes Revenue at an assumed $765 per tonne $7,627,050 Modelled cost $7,450,000 Contribution before tax, overhead, credit loss and unmodelled contingencies $177,050 If the sale price falls $20 per tonne Revenue falls $199,400; contribution becomes negative $22,350 - A superficially comfortable $65 price spread was insufficient to absorb that movement. The 0.3% is treated only as a commercial sensitivity: the numbers are assumptions, not a market quotation or a normal loss allowance.
- THE DOCTRINE
Keep taxes and recoverable amounts in their accepted treatment. Do not insert a universal Nigerian tax or levy table into a model without a current transaction-specific applicability review.
Ask what the margin is on saleable tonnes, not purchased ones.
6:00 – 10:00
Funding currency changes the conclusion
- Suppose the same $7.45 million cost must be acquired at an assumed ₦1,500 per dollar: it becomes ₦11.175 billion. At an assumed ₦1,650 it becomes ₦12.2925 billion, an increase of ₦1.1175 billion before other effects. These are teaching assumptions, not current exchange rates.
- The exposure depends on what has already been funded or hedged and when naira receipts are obtained. Treasury should distinguish currency purchases, settlement dates, contractual pricing adjustments and remaining exposure.
- THE DOCTRINE
Do not translate the whole transaction at a single convenient closing rate and call the difference trading performance.
- Management review shows expected contribution, downside contribution, peak funding and unresolved conditions together, so a high-margin but cash-intensive cargo can be compared with a smaller executable alternative. The decision record keeps the declined option and the reason, allowing later review without hindsight rewriting.
Ask which part of the dollar cost is still unbought, and when the naira arrives.
10:00 – 15:00
A margin engine before a margin predictor
- At the customer and depot interface, AI can forecast demand, optimise allocation from confirmed saleable stock and rank collection actions. A sales forecast should not create stock or approved credit.
- Build the margin engine first: it recomputes sales value, saleable quantity, landed cost, financing and currency conversion from the trade state. A model may predict uncertain inputs; it should not learn an opaque shortcut around the accepted financial definition.
- When a depot runs dry, recorded sales stop at what was available. Observed sales during stockouts are a censored view of demand; treating zero sales as zero need biases replenishment. Use order history, lost-sales records and availability flags where reliable.
- Traders often lower prices when demand is already weak and raise them when it is strong, so a naive regression can infer that higher prices cause higher demand. Observational fit alone does not establish elasticity. Where suitable and lawful, a bounded randomised pilot can investigate response; until then, use approved response ranges.
When the demand curve slopes upward, ask who set the prices.
15:00 – 20:00
Available stock is a governed state
- Physical stock, owned stock, released stock and uncommitted stock can differ. Before allocation, reconcile title, batch quality, storage restrictions, existing commitments and movement capability.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. A fictional distributor's depot shows a high tank reading. Part of the product is on quality hold and part is reserved for an existing commitment. A positive ERP balance alone does not establish saleable availability.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The allocation must also respect each customer's accepted demand and credit capacity, firm supply obligations and funding. A plan that sells held stock or extends unsupported credit is a mathematical optimum with no commercial standing.
Assumed: 1,500 tonnes, two customers each wanting 1,000 Customer A at $30 per tonne Customer B at $20 per tonne Contribution Favour A 1,000 tonnes 500 tonnes $40,000 Equal split 750 tonnes 750 tonnes $37,500 - The apparent $2,500 improvement may disappear if the first allocation breaches a firm supply obligation or creates an unsupported credit exposure. Classify every constraint as mandatory, contractual, operational or discretionary. A discretionary service preference can be traded against contribution within authority.
- THE DOCTRINE
A safety restriction cannot become a soft preference because the solver otherwise fails.
- Storage value is net of storage, financing, loss exposure, quality deterioration and opportunity cost. A forecast price rise alone does not establish a feasible storage arbitrage.
Ask which stock you could sell this afternoon, and who would sign for it.
20:00 – 25:00
Predict, then optimise, and your wrap
- The book's design is to predict, then optimise. At each planning cut-off the forecast supplies demand distributions and arrival scenarios; the optimiser selects purchase, transfer and dispatch decisions under accepted constraints. A rolling-horizon design solves over a defined horizon, accepts only the immediate authorised decisions, observes new evidence and solves again, keeping past commitments fixed.
- Validate with a replay engine applying the same cut-offs, lead times and constraints to the model policy and the current policy, reporting stockouts, excess inventory, contribution, peak cash and overrides. A forecast with lower mean error can produce worse decisions if its upper tail is poorly calibrated.
- Reinforcement learning, which learns a policy from the consequences of actions, has a narrow place. Conservative Q-Learning and Decision Transformer are published offline approaches that do not establish suitability for energy trading. Ask whether the record contains the actions the policy proposes; keep it in advisory shadow mode with a deterministic feasibility layer and human authorisation.
- Before you close: write the delivered margin for your Session 1 decision on saleable quantity, and name the difficult day you would replay. That is Section 8 of your 90-Day AI Trading Plan.
- Pop Quiz 2 opens now: five questions, closed book, on Sessions 6 to 8. Sit it before Session 9, then write your assignment.
Replay the hardest day you had, with only what you knew that morning.
Key Terms
Saleable quantity
The quantity that can actually be sold and delivered on the accepted basis, which can differ from the quantity purchased.
Delivered margin
Contribution on the actual accepted delivery, with a consistent quantity basis and a complete cost boundary, before items explicitly excluded.
Censored demand
Recorded sales that stop at what was available. During a stockout they understate what customers wanted.
Constraint classes
Mandatory, contractual, operational or discretionary. A discretionary service preference can be traded against contribution within authority; a safety restriction cannot become a soft preference.
Self-Check — Before You Submit
1. In the book's assumed 10,000-tonne product cargo, why does contribution turn negative when the sale price falls $20 per tonne?
- A. Because the 0.3% quantity difference is a normal loss allowance
- B. Because the exchange rate moves from ₦1,500 to ₦1,650
- C. Because the $199,400 revenue fall on 9,970 saleable tonnes exceeds the $177,050 contribution
- D. Because financing cost rises with the price fall
C. Because the $199,400 revenue fall on 9,970 saleable tonnes exceeds the $177,050 contribution — 9,970 tonnes × $20 is $199,400, which exceeds $177,050, leaving negative $22,350. The 0.3% is a commercial sensitivity, not a normal loss allowance, and the currency example is a separate illustration.
2. A fictional distributor's optimiser proposes more volume to a higher-contribution customer by relaxing a tank-compatibility restriction that the operations team classifies as a safety condition. What should the commercial director do?
- A. Reject the plan: a safety restriction is mandatory and cannot be traded against contribution
- B. Accept it within authority, because contribution rises
- C. Give the restriction a penalty cost and rerun the solver
- D. Ask for a better demand forecast
A. Reject the plan: a safety restriction is mandatory and cannot be traded against contribution — A safety restriction cannot become a soft preference because the solver otherwise fails. The penalty-cost route is the tempting one, but an unauthorised activity is not made acceptable by assigning it a financial penalty in the optimiser.
3. A demand model trained on depot sales shows near-zero demand in weeks when the depot had no product. What is the right reading?
- A. Customer need was zero in those weeks
- B. Prices were too high in those weeks
- C. Those weeks should be quietly removed
- D. Sales during stockouts are a censored view of demand; use availability flags and lost-sales records where reliable
D. Sales during stockouts are a censored view of demand; use availability flags and lost-sales records where reliable — Treating zero sales during closure as zero need biases replenishment. Quietly removing the weeks hides the problem instead of disclosing demand that cannot be observed.
Today’s Assignment
Write Section 8 of your plan: the delivered margin of your Session 1 decision, its funding currency and the constraints that govern your available stock or capacity.
- Build the delivered margin for your decision in your company on a complete cost boundary and a saleable-quantity basis, or the nearest equivalent economic unit your business uses. Label every figure assumed, or give structure and size without confidential prices.
- Show the funding-currency exposure: what is already funded or hedged, what remains open and when receipts arrive.
- Present expected contribution, downside contribution, peak funding and unresolved conditions together, with the declined alternative and the reason.
- Classify the constraints on your available stock or capacity as mandatory, contractual, operational or discretionary, and say which one your current planning treats as softer than it is.
- Name the difficult day you would replay.
- The honest part: where your company reports a quoted spread or a closing-rate translation as trading performance. If no AI is in use yet, that is normal; the first need may be a margin engine, not a margin predictor. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 8 — The delivered margin, its funding currency and the constraints that govern available stock.
How this is marked — the rubric
| Builds the margin on saleable quantity and a complete cost boundary, figures labelled assumed | 30 |
| Shows the funding-currency exposure, what is fixed and when receipts arrive | 20 |
| Presents expected, downside, peak funding and unresolved conditions with the declined alternative | 20 |
| Classifies stock or capacity constraints into the four classes, names a replay day and gives an honest example | 30 |
| 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 — Contracts, Crude Pricing and the Delivered Margin 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
Delivered marginFunding-currency exposureCensored demandConstraint classificationPredict-then-optimise replay
Part Four · Gas and Physical ExecutionSESSION 9
Gas on an Energy Basis and LNG and LPG Portfolios
Why This Matters
Where a contract prices gas by energy, volume alone is insufficient, and the chain from nomination through delivery and billing to collected value must reconcile. LNG and LPG cargos add timing, quantity basis and destination rights that a spreadsheet can make look freer than they are. If your business trades crude or products, the lesson transfers: your quantity basis also needs evidence, and every option in your plan needs a right behind it. A model should not invent a right to redirect committed gas, or invent pipeline capacity. By the end you will know your accepted quantity basis, the statuses your records must keep, and which of your options are real.
25-Minute Module
0:00 – 5:00
Energy, mass and volume
- Three quantities answer three questions. Volume is the space the product occupies, which is why the book insists on stated reference conditions. Mass is how much product there is. Energy is the heat it can deliver. A contract may price on any of them, and moving between them needs accepted evidence.
- Pipeline natural gas, LNG, CNG and LPG should not be collapsed into one gas product code. LNG and CNG describe forms of natural-gas handling; LPG is a different commercial product family. Delivery, storage, measurement and contractual arrangements differ.
- Where the contract prices gas by energy, volume alone is insufficient. Preserve the reference conditions, accepted heating value, measurement method and agreed conversion.
- THE DOCTRINE
Do not substitute a generic heating value because the commercial model requires a populated cell.
Ask what unit the contract is paid in before you ask what the meter says.
5:00 – 10:00
A bounded energy calculation
Fictional accepted delivery record Accepted factor 0.036 Accepted factor 0.034 Stated volume 1,000,000 standard cubic metres 1,000,000 standard cubic metres Energy 36,000 MMBtu 34,000 MMBtu Gross value at an assumed $4 per MMBtu $144,000 $136,000 Difference $8,000 No change in stated volume - The factor, in MMBtu per standard cubic metre, is a teaching input, not a conversion standard for every gas stream. The real calculation requires the contract, composition or accepted calorific evidence and correct unit treatment.
- IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. A fictional gas aggregator's delivery point records volume reliably, but the accepted heating value for the month is still pending. The record should show it as pending, not fill the cell with last month's figure.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The invoice, receivable and contribution all depend on that pending value; at the book's assumed numbers, two factors produce an $8,000 difference on one delivery. An energy-content correction and a physical loss are different causes with different financial consequences, and the commercial review must say which it is.
When the volumes agree and the values do not, look at the factor first.
10:00 – 15:00
Nomination, capacity and order to cash
- An order for tomorrow differs from a confirmed nomination accepted by the transport and receiving parties. Represent supply rights, customer nominations, delivery intervals, capacity, quality and imbalance treatment, and keep those status distinctions intact in every forecast.
Status in the commercial review What it records Contracted energy What the agreement provides for Nominated energy What was requested and confirmed for delivery Delivered accepted energy What was delivered on the accepted measurement basis Billed energy What was invoiced Collected value What was paid - Explain differences by timing, measurement, contractual flexibility or unresolved evidence; approved rules govern imbalance, curtailment and priority. Choose one delivery point and one contract type, and reconcile a month of nomination, delivery, invoice and cash evidence before forecasting. The first useful model may be an exception detector exposing inconsistent bases.
- Ask of a nomination model: does it forecast both nominated demand and accepted delivered energy, using only evidence available at nomination time, on the contractual energy basis? The book names a hierarchical model, quantile boosting and a state-space comparator as candidates. Separate labels for delivery shortfall, customer demand and measurement adjustment stop a meter correction being learned as a demand shock.
- THE DOCTRINE
Evaluate a nomination policy with imbalance cost, unserved energy and peak funding alongside predictive error. A model that consistently underpredicts the largest customer can show a low pooled error while repeatedly failing the commercial obligation.
Ask for the error on your largest customer, not the average.
15:00 – 21:00
LNG and LPG: the real cargo rights
- A gas cargo portfolio needs product, loading and receiving windows, delivery point, accepted quantity basis, quality, transport, storage, destination rights and payment terms. Diversion or cancellation is possible only within the actual rights and constraints; a model cannot treat every destination as a free commercial choice.
- GIIGNL's Custody Transfer Handbook, sixth edition, identifies its 2021 best-practice basis for LNG custody transfer. Use the accepted applicable edition, contract and measurement procedure for real work. Citing that edition is a historical reference, not an assertion that no later edition exists.
- LNG economics often require an accepted energy determination; LPG trading may use mass with composition and quality conditions. A nominal vessel capacity does not establish accepted delivered quantity. Keep loaded, transported, received and invoiced states distinct.
- The book's LNG teaching cargo: 3 million MMBtu of accepted saleable energy at an assumed $9 per MMBtu, against $8.60 of allocated costs, gives $1.2 million contribution before excluded items. A separate $300,000 delay and funding sensitivity reduces it to $900,000. Neither the energy nor the prices describe a real cargo.
- A receiving outage may delay discharge and reduce downstream sales at once. Build joint scenarios rather than sampling every input independently. For LPG, forecast dispatch can create space in a planning scenario; confirmed capacity needs the actual receiving assessment.
Ask which destination change the agreement actually allows, and by when.
21:00 – 25:00
Option value tied to a right, and your wrap
- Represent each cargo as a package of product, accepted quantity basis, timing, location, vessel compatibility and contractual rights. A scenario-tree optimiser can compare permitted destinations and timing choices, but a destination change is available only if the agreement and logistics allow it at that decision stage.
- THE DOCTRINE
A perfect-foresight upper bound assumes the best decision after knowing the future. It is a diagnostic ceiling, not attainable value. An option value based on an unavailable destination is an analytical error, not a commercial opportunity.
- Compare the AI policy with a fixed-destination policy and the existing workflow on expected contribution, adverse outcome, peak cash and the frequency of infeasible choices. Boil-off and operational loss need accepted engineering and contractual treatment, not a generic percentage learned from unrelated voyages. If most proposed value depends on rights absent from the accepted agreements, the result is a commercial-development hypothesis.
- Before you close: list every optional action your Session 1 decision assumes, and the right behind each. That is Section 9 of your 90-Day AI Trading Plan. Write your assignment now.
Review the most valuable options first. That is where an absent right costs most.
Key Terms
Energy basis
Pricing on delivered energy, which needs stated reference conditions, an accepted heating value, a measurement method and an agreed conversion.
Confirmed nomination
A requested delivery accepted by the transport and receiving parties. It differs from an order and from delivered accepted energy.
Perfect-foresight upper bound
The value of the best decision made after knowing the future. A diagnostic ceiling, never attainable value.
Joint scenario
A scenario in which inputs sharing a cause, such as arrival, receiving readiness and demand, move together rather than independently.
Self-Check — Before You Submit
1. Using the book's fictional delivery of 1,000,000 standard cubic metres at an assumed $4 per MMBtu, what changes if the accepted factor is 0.034 rather than 0.036 MMBtu per standard cubic metre?
- A. Nothing, because the volume is unchanged
- B. Energy falls by 2,000 MMBtu and value by $8,000
- C. Value falls by $4,000
- D. Energy falls by 2%
B. Energy falls by 2,000 MMBtu and value by $8,000 — 36,000 MMBtu becomes 34,000, and $144,000 becomes $136,000. Unchanged volume is the trap: where the contract prices energy, volume alone is insufficient.
2. Why should a nomination model train separate labels for delivery shortfall, customer demand and measurement adjustment?
- A. So a meter correction is not learned as a demand shock
- B. Because separate labels always improve pooled accuracy
- C. Because pooled error is the main measure of a nomination policy
- D. Because heating values are always measured, never estimated
A. So a meter correction is not learned as a demand shock — Different causes need different labels. Pooled error is the tempting measure, but the book evaluates imbalance cost, unserved energy and peak funding, and warns that a low pooled error can hide repeated failure on the largest customer.
3. A portfolio model gives a fictional LNG cargo a large option value from diverting to a second destination. The sale agreement does not permit that destination. What should the commercial director conclude?
- A. Book the option value in the plan
- B. Use the diversion as the base case
- C. Halve the option value and proceed
- D. Treat it as an analytical error, at most a commercial-development hypothesis, not an executable improvement
D. Treat it as an analytical error, at most a commercial-development hypothesis, not an executable improvement — Diversion is possible only within the actual rights. Discounting the value is tempting but still counts an option the company does not hold.
Today’s Assignment
Write Section 9 of your plan: the accepted quantity basis of your Session 1 decision, its order-to-cash reconciliation and every option tied to a right.
- State the accepted quantity basis of your decision in your company (energy, mass or volume, with its reference conditions) and the evidence any conversion needs.
- Separate the statuses your records must keep: for gas, order, confirmed nomination, delivered accepted energy, billed and collected; for crude or products, the nearest equivalents.
- Reconcile one month or one cargo of that chain and explain the differences by timing, measurement, contractual flexibility or unresolved evidence, or say plainly why you cannot yet.
- List every optional action your plan assumes, with the contractual or operational right that permits it and its latest decision time. Mark any option with no right behind it.
- The honest part: where your company has filled a missing heating value, quality or quantity with a convenient figure, or counted an option it did not hold. Describe structure; no confidential terms or counterparty names. If no AI is in use yet, that is normal; the first useful tool may be an exception detector. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 9 — Our accepted quantity basis, the order-to-cash reconciliation and every option tied to a right.
How this is marked — the rubric
| States the accepted quantity basis and the evidence any conversion needs | 20 |
| Separates the record statuses from order or nomination to collected value | 25 |
| Reconciles one month or cargo with differences explained, or states why not | 25 |
| Ties every assumed option to a right and a latest decision time, with an honest example | 30 |
| 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
Energy-basis disciplineNomination status controlOrder-to-cash reconciliationOption rights reviewJoint scenario planning
Part Four · Gas and Physical ExecutionSESSION 10
Freight, Arrival, Demurrage and Custody Transfer
Why This Matters
Every trade in this programme eventually meets a voyage and a measurement. Vessels arrive late, notices are disputed, and the quantity received differs from the quantity loaded. Those moments decide whether money is paid out, recovered in a claim or written off without a reason. A variance is a question before it is a loss. Today you treat the voyage clock as commercial evidence, weigh a delay intervention on the evidence for its probabilities, and build a reconciled cut-off that counts a consequence once and accuses no one on the strength of a score. Practice Exam 2 closes the week.
25-Minute Module
0:00 – 5:00
The voyage clock is commercial evidence
- Two terms first. In plain words, laytime is the time allowance a particular agreement sets, and demurrage is the delay cost that agreement may attach when the allowance is exceeded. The reviewed agreement's own notices, exclusions and rates decide whether a period counts and what it costs, which is why the clock is evidence.
- Preserve charterparty and sale-contract obligations separately; their time provisions may differ. Extract notices, event statements, accepted times, exclusions and claims deadlines into a reviewed chronology. A chronological table is a stronger starting point than asking a language model to pronounce the final claim amount from scattered emails.
- THE DOCTRINE
Arrival prediction can improve coordination. It cannot decide whether a notice is valid or whether a period counts as laytime under a particular agreement.
- Show feed age and coverage in the arrival advisory. A gap in vessel-location data can arise for several reasons and should not alone determine commercial or compliance conclusions. Evaluate error at the horizons that matter: a moderately uncertain estimate that lets the receiving team prepare can beat an accurate one shortly before arrival.
A reviewed chronology beats a clever summary of scattered emails.
5:00 – 10:00
A delay decision on evidence
- Expected value weighs each possible cost by its probability. Here it asks whether paying to reduce the chance of a delay is worth more than it costs. Every input below is fictional: an avoidable delay costing $40,000 a day for three days under accepted applicable terms, and a $30,000 intervention that cuts the delay probability to 0.20.
Fictional delay case Baseline probability 0.60 Baseline probability 0.35 Expected delay cost without intervention 0.60 × $120,000 = $72,000 0.35 × $120,000 = $42,000 Expected delay cost with intervention 0.20 × $120,000 = $24,000 0.20 × $120,000 = $24,000 Expected avoided cost $48,000 $18,000 Expected net benefit after $30,000 $18,000 Negative $12,000 - The same intervention moves from worthwhile to loss-making when one probability changes. The recommendation depends on evidence supporting the probabilities, and should not present the first result as a guaranteed saving.
- Ask where the baseline probability came from and how it was checked. If nobody can answer, show the committee both columns.
When an intervention pays only at one probability, the probability is the decision.
10:00 – 14:00
Claims control and the arrival model
- AI can assemble events, compare terms and flag missing notices. A qualified commercial reviewer determines the claim position and required evidence.
- THE DOCTRINE
A claim submitted is not revenue received. Record claims raised, disputed, accepted and collected separately.
- An arrival model should predict a distribution over a defined event. Arrival at a point, berth access and discharge completion are different targets. The book names quantile boosting and a time-to-event model with time-varying operational evidence as candidates; voyage, weather, port, vessel and queue features must use their actual publication times.
- An unfinished voyage has no completed duration; its record is censored, telling you only that the event had not happened by a given time. If missing records are concentrated in disrupted voyages, the assumption that censoring is uninformative may fail and needs investigation.
- Keep the contractual laytime engine separate: it applies accepted notices, excluded periods and rates under the reviewed agreement. For claims assistance, measure missed material events, incorrect ordering and reviewer corrections. The value lies in a better decision or a supported claim, not an attractive map.
Ask whether the model predicts arrival, berth or discharge. They are three different targets.
14:00 – 20:00
A variance is a question before it is a loss
- Quantity differences can reflect measurement basis, timing, temperature, density, water, permitted adjustments or unresolved physical discrepancy. API's Petroleum Measurement 2024 catalogue covers gauging, sampling, temperature, density and quantity calculation; it does not provide the licensed calculation or establish a facility's compliance.
- Build a reconciled cut-off: opening stock plus accepted receipts less accepted dispatches and approved adjustments, compared with closing stock on the same basis, each adjustment linked to evidence and authority.
Fictional custody variance Tonnes Loaded, on a stated comparable basis 50,000 Received 49,850 Initial difference, 0.3% of loaded 150 Documented cut-off issue 40 Accepted correction 30 Residual 80, a $56,000 valuation sensitivity at an assumed $700 per tonne - IN THE PHYSICAL VIEW · THE WORKED CASE
- Physical view. The 80-tonne residual remains an open question for measurement and operations: basis, timing, temperature, density, water or a physical discrepancy. Investigation may establish that no physical loss occurred, and that is a valid operating result.
IN THE COMMERCIAL, RISK AND CASH VIEWS · THE PARALLEL- Commercial, risk and cash. The $56,000 is a valuation sensitivity, not automatically an accepted loss, recovery or claim. Separate physical recovery from corrected billing and accounting reclassification, so the financial consequence is counted once.
Ask what remains after the explained items, and who accepted each explanation.
20:00 – 25:00
Score the residual, never the person, and close Week 2
- Start with deterministic reconciliation; only then may an anomaly model score the residual. A transparent comparator standardises residuals within comparable measurement conditions. The book names an isolation-based detector and an autoencoder, which learns to reconstruct normal records and scores how poorly it reconstructs a new one, as challengers.
- THE DOCTRINE
An anomaly score is not a probability of theft, fraud or physical loss. Use scores to prioritise evidence review, preserve the original balance and prohibit automated accusation or write-off.
- A rare accepted adjustment is not misconduct merely because it is uncommon. Reviewing only alerts cannot reveal missed events, so sample lower-score cases through an accepted audit design, and measure precision (how many flags were real) and recall (how many real cases were flagged) on mature investigated cases.
- Before you close: list the last five material delays or variances that touched your Session 1 decision. That is Section 10 of your 90-Day AI Trading 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.
A flagged case is the start of an investigation, not its conclusion.
Key Terms
Laytime and demurrage
The time allowance a particular agreement sets, and the delay cost that agreement may attach when it is exceeded. Only the reviewed agreement's own terms decide which periods count and at what rate.
Reconciled cut-off
Opening stock plus accepted receipts less accepted dispatches and approved adjustments, compared with closing stock on the same basis, each adjustment linked to evidence and authority.
Censoring
A record that ends before the event occurs, such as an unfinished voyage. It shows only that the event had not happened by that time.
Anomaly score
A measure of how unusual a residual is. It prioritises evidence review and is never a probability of theft, fraud or physical loss.
Self-Check — Before You Submit
1. In the book's fictional delay case, the baseline delay probability is 0.35 instead of 0.60, and the intervention still reduces it to 0.20. What happens to the $30,000 intervention?
- A. Its expected net benefit stays $18,000
- B. Its expected avoided cost rises
- C. Expected avoided cost falls to $18,000 and the net benefit becomes negative $12,000
- D. It becomes worthwhile at any cost
C. Expected avoided cost falls to $18,000 and the net benefit becomes negative $12,000 — 0.15 × $120,000 is $18,000 avoided, less $30,000, giving negative $12,000. Keeping the $18,000 figure is the trap: the recommendation depends on evidence for the probabilities.
2. An anomaly model scores a fictional depot's monthly residual as highly unusual. What should the head of operations do?
- A. Prioritise the case for evidence review against the reconciled balance, without accusation or write-off
- B. Suspend the depot team pending investigation
- C. Write off the residual and close the month
- D. Report the residual as a confirmed loss
A. Prioritise the case for evidence review against the reconciled balance, without accusation or write-off — An anomaly score is not a probability of theft, fraud or physical loss, and investigation may show no physical loss occurred. Writing off or treating the score as proof skips the question the variance raises.
3. A vessel's position feed goes silent for a day before arrival. What does that establish on its own?
- A. A compliance concern requiring escalation as misconduct
- B. Nothing conclusive: a gap can arise for several reasons and should not alone determine commercial or compliance conclusions
- C. That laytime has started
- D. That the vessel will arrive late
B. Nothing conclusive: a gap can arise for several reasons and should not alone determine commercial or compliance conclusions — Feed gaps have several possible causes. Treating the gap as a finding is the tempting error; the advisory should show feed age and coverage and leave conclusions to evidence.
Today’s Assignment
Write Section 10 of your plan: the voyage clock, the delay decision and the reconciled cut-off behind your Session 1 decision.
- Review the last five material delays or variances that touched your decision in your company, and classify each avoidable cause: late information, unavailable resources, unresolved authority or an unavoidable condition.
- Set out the reviewed chronology (notices, events, accepted times, exclusions and claims deadlines) or the reconciled cut-off your decision needs, and who accepts each entry by role.
- Show, on inputs labelled fictional or assumed, how an expected-value delay or variance decision would depend on evidence for its probabilities, and where that evidence would come from.
- Separate physical recovery, corrected billing and accounting reclassification so the consequence is counted once, and record claims raised, disputed, accepted and collected separately.
- State how any anomaly score would be used, and the rule that prevents it becoming an accusation or a write-off.
- The honest part: where your company has booked a claim as revenue, or treated a variance as a loss before investigating it. Do not name counterparties, vessels, terminals or staff. If no AI is in use yet, that is normal; the first need may be deterministic reconciliation. Aim for 400 to 900 words.
YOUR 90-DAY PLAN
Section 10 — The voyage clock, the delay decision and the reconciled cut-off behind our decision.
How this is marked — the rubric
| Classifies five real delays or variances by avoidable cause | 25 |
| Sets out the chronology or reconciled cut-off with acceptance by role | 25 |
| Shows how the expected-value decision depends on evidenced probabilities, inputs labelled | 25 |
| Separates recovery, billing and reclassification, bars accusation and write-off, and gives an honest example | 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 — Crude, Products, Gas and Physical Execution opens now. Sit it before you start the next session — it is how you find out whether the doctrine landed, and the practice papers may be resat as often as you like.
Sit Practice Exam 2 →Closed book, no notes, and Coach Ritchie switches himself off while you sit it.Capabilities Covered
Voyage chronologyEvidence-based delay decisionsClaims controlReconciled cut-offsAnomaly review without accusation