In short: An LNG delivery programme is an inventory problem at both ends, with the binding constraints sitting in liquefaction and regasification tanks rather than in the ships everyone argues about. A slipped loading slot converts directly into days of tank at the liquefaction plant, and the arithmetic surprises people who have never run it. A diversion right on a delivered cargo is an option with exercise economics that can be computed rather than guessed at. Regasification capacity is a booking rather than a production rate, so slot rules at the receiving end shape the programme more than physical throughput does.
The annual delivery programme was agreed in October. Sixty-four cargoes, laycans allocated by buyer, vessels assigned, everyone signed. By the second week of January three things have happened: a regasification terminal has moved its planned maintenance by nine days, a vessel came out of dry dock late, and the buyer with the widest tolerance band has quietly indicated they will take the low end. Two of those were foreseeable and none of them were in the plan.
What follows is usually a fortnight of bilateral phone calls, and the result is a revised programme that nobody has checked against the liquefaction tank curve. The reason it works out more often than it should is that the plant has more storage than the schedulers give it credit for, right up until the month when it does not.
The programme is an inventory problem at both ends
The habit is to treat cargo scheduling as a shipping question, because ships are what move and ships are what get argued about. The binding constraints sit in tanks.
At the liquefaction end, the trains produce at a rate you can trim but cannot stop cheaply, and the product goes into storage that fills on a clock. At the regasification end, the tanks drain into a send-out obligation that is set by a downstream gas market, a power station, or a distribution network with its own seasonality. The ships are the mechanism connecting one inventory curve to another, and both curves have hard ceilings and floors.
This is the structure the maritime operations research literature calls an inventory routing problem, and the LNG version has been treated directly. Rakke and colleagues published a rolling horizon heuristic specifically for constructing an annual delivery programme in Transportation Research Part C in 2011, and the broader family is surveyed by Christiansen, Fagerholt, Nygreen and Ronen in the European Journal of Operational Research in 2013. The useful thing about that framing for a commercial scheduler is that it puts the tanks in the objective rather than in the constraints nobody looks at.
Stated properly, the programme has to keep the loading tank between its heel and its ullage limit for every day of the year, keep every discharge terminal above its own floor, respect vessel availability and speed, respect each buyer's contractual quantity and its tolerance, and do all of it while the arrival of information runs backwards to the decision.
What a slipped slot costs, in days of tank
Put arithmetic on the liquefaction end, because the number surprises people who have never run it.
Take a single train at five million tonnes a year. That is 13,700 tonnes a day. A cargo of 174,000 cubic metres at a density of 0.45 tonnes per cubic metre is about 78,000 tonnes, so the plant fills a ship every 5.7 days and lifts around 64 cargoes a year.
Now the storage. Two tanks at 165,000 cubic metres each hold 330,000 cubic metres, roughly 148,500 tonnes, which reads as 10.8 days of production. Take out the heel that has to stay in each tank to keep it cold, take out the ullage you need at the top for a safe filling operation, and the genuinely usable band is closer to 65 percent of that, so about seven days.
Seven days is the entire buffer between a missed loading and a rate cut on the train. One cargo slipping by a week, with no substitute vessel found, is not an inconvenience for the commercial team; it is a production decision. And because the tank position going into that week depends on the previous six loadings, the plant can be at four days of headroom or at one day of headroom on the same nominal programme, purely as a result of how the earlier laycans clustered.
Almost no annual programme is reviewed against a plotted tank curve. Building one is a morning's work in a spreadsheet: cumulative production minus cumulative liftings, day by day, with the heel and ullage lines drawn on. The weeks where the line runs close to a limit are the weeks worth defending contractually, and they are rarely the weeks the commercial team thought were tight.
Destination flexibility is an option with a running cost
A cargo sold on a delivered basis with a diversion right is an option, and options have exercise economics that people either compute or guess at.
Suppose an inter-basin spread opens at 2.50 dollars per MMBtu against the contractual destination. A 78,000 tonne cargo at 52 MMBtu per tonne is about 4.1 trillion Btu, so the gross prize is roughly 10.2 million dollars. Against that, the diversion adds twenty days of round voyage. Take a combined charter and fuel cost of 110 thousand dollars a day and substitute your own number, which gives 2.2 million. Add half a million for canal and port differences.
Then the term everyone forgets. At a guaranteed boil-off rate of 0.1 percent a day, a 174,000 cubic metre cargo loses about 78 tonnes daily. Twenty extra days is 1,560 tonnes, roughly 81,000 MMBtu, and at 11 dollars that is close to 900 thousand dollars of cargo that has evaporated. On a vessel burning boil-off as fuel some of that displaces bunkers and the net cost is lower; on a reliquefaction-fitted ship you pay for the power instead. Either way the term belongs in the calculation, and in most of the diversion notes I have seen it does not appear at all.
Total incremental cost lands near 3.6 million, which is 0.88 dollars per MMBtu on the cargo. That is the threshold. A spread below roughly 0.9 dollars does not pay to chase, however good it looks on a screen, and a spread of 2.50 is worth about 6.6 million net rather than the 10.2 that gets quoted into the meeting.
Work that threshold out once per trade route and put it on a card. It converts a recurring argument into a lookup.
Slots at the regasification end behave differently
Liquefaction capacity is a production rate. Regasification capacity is a booking, and bookings have their own rules that shape the delivery programme more than the physical throughput does.
Terminal capacity is usually sold as a bundle of unloading slots plus storage plus send-out, allocated in an annual process, and it commonly carries use-it-or-lose-it provisions. A slot you hold and cannot fill is a sunk cost that also removes a berth day from the system. A slot you need and do not hold has to be acquired on a secondary market where the price reflects how tight the month is, which is exactly the month you will need it.
Three practical consequences for the programme.
Slot inventory is a planning object. Hold the slot calendar as data alongside the vessel and tank data, because a delivery programme that is feasible on shipping and infeasible on slots looks fine until the nomination deadline.
Slot swaps are cheaper than vessel swaps and get considered later. Two buyers whose flexibility runs in opposite directions in the same month can often solve a problem between them at no cost to either. That trade is invisible unless somebody is looking at the whole book rather than at one contract.
The nomination deadline sets the real horizon. Whatever the contract says about the annual programme, the decision hardens at the point where the terminal's notice period runs out, and the plan should be re-solved at that point rather than defended from October.
Queueing and berth availability at the receiving terminal is its own problem with its own arithmetic, covered in N18.
Contract structure decides how much of the plan is yours
The programme is constrained by what was signed years earlier, and the terms that matter for scheduling are a short list.
The annual contract quantity with its upward and downward tolerance sets the volume band. A ten percent downward tolerance on a two million tonne contract is 200 thousand tonnes of swing, or about two and a half cargoes, sitting inside a number the plan treats as agreed. Across a book of eight term contracts, that is twenty cargoes of uncertainty, which is a third of a train's annual programme and considerably larger than any forecasting improvement available elsewhere.
Take-or-pay protects revenue and does nothing for the tank curve. A buyer who pays for gas they did not lift has met their obligation, and the plant still has to place the molecules or cut rates.
Destination clauses, diversion rights and profit-sharing formulas decide who captures the arbitrage computed above, and the split changes the threshold. If the seller keeps half the upside, the spread has to clear roughly 1.8 dollars per MMBtu on the numbers above rather than 0.9, because the costs are not shared in the same proportion as the gain. Check whether your profit-sharing mechanism nets the voyage costs before the split or after it, because that clause is worth more than most of the ones that get negotiated hardest.
Scheduling flexibility clauses, meaning how much a laycan can move and with how much notice, are the terms the programme actually consumes, and they are usually drafted by people who will never build a programme.
Building and rebuilding it
Two things make a delivery programme durable.
The first is solving it as a whole rather than sequentially. Assigning cargoes buyer by buyer in order of contract seniority produces a feasible answer and leaves value on the table, because the last buyer inherits every awkward slot. A mixed integer formulation over the year with tank levels as state variables handles the coupling directly, and the rolling horizon approach in the Rakke paper exists because solving twelve months at full fidelity in one shot is hard.
The second is treating the programme as a decision that gets remade. Information arrives continuously: vessel positions, maintenance confirmations, buyer indications, market spreads. A programme fixed in October and defended until March is running on the least information anyone will have all year. Re-solve monthly with the near term frozen at whatever the contractual notice periods actually require, and let the far end of the horizon stay soft.
What that produces is a rolling artefact where the first six weeks are commitments, the next three months are intentions, and the back half of the year is capacity that has been checked for feasibility rather than allocated. Most disputes about the programme are really disputes about which of those three categories a given cargo is in.
The limit
The arithmetic above assumes you know the distributions. In practice the two inputs that drive everything, unplanned plant availability and voyage time variability, are estimated from a small number of observations, and a five year history of a plant that has had one major outage tells you very little about the frequency of major outages.
That argues for using the model to compare programmes rather than to predict outcomes. A schedule with three days of tank headroom in its tightest week and one with one day of headroom can be compared honestly even when the absolute probability of hitting the limit is not knowable. The ranking is more reliable than the number.
There is also a data problem that has to be solved before any of this runs. The tank curve needs daily production and daily lifting data reconciled to the same basis, and those two series usually live in a plant historian and a commercial system that disagree about what a day is and about whether a cargo counts on loading completion or on bill of lading date. Reconciling them is a week of unglamorous work and it is the prerequisite for everything above.
Plot last year's actual liquefaction tank level against the programme you agreed for last year, and count the days where the realised level sat outside the band the programme implied.