In short: Drilling rig scheduling is an asymmetric travelling salesman problem, because the cost of reaching the next well depends on which well the rig just left. Building the schedule by walking down the subsurface priority list makes the sequence a consequence of the priority order, which usually leaves avoidable rig moves inside it. Each avoidable move is roughly a week of spread rate, so move cost dominates the objective rather than sitting alongside it. Well durations are skewed, closer to lognormal than to normal, because stuck pipe, lost circulation and equipment failure all have long tails.
The programme has eighteen wells, three rigs, two of them contracted through the year and one on a short-term slot. The schedule was built in a spreadsheet by a superintendent who knows every well and every rig, and it is a good schedule. It was also built by walking down the priority list from the subsurface team and assigning the next available rig, which means the sequence is a consequence of the priority order rather than a decision anyone made.
Somewhere in that sequence are three or four rig moves that could have been avoided by reordering wells whose relative priority nobody would defend to within a month. Each of those moves is a week of spread rate.
What the scheduling problem actually is
A single rig working through a set of wells, where the cost of getting from one well to the next depends on which well you just left, is a travelling salesman problem. It is asymmetric, because a move from A to B need not cost the same as B to A once tow direction, weather exposure and rig-up requirements are considered. It has time windows, because permits expire, partner funding arrives on dates, weather seasons close, and some wells cannot be drilled until a facility is available.
With several rigs of differing capability it becomes a vehicle routing problem with a heterogeneous fleet, and once you add the shared resources that several rigs draw on at once, meaning specialist crews, completion equipment, marine support and the long lead materials covered in CC6, it takes the shape of a resource-constrained project scheduling problem. Iyer, Grossmann, Vasantharajan and Cullick formulated the offshore version, covering infrastructure investment together with drilling and production scheduling, in Industrial and Engineering Chemistry Research in 1998, and the structure they described has not changed.
Naming it properly matters for one practical reason. These problems are NP-hard, meaning the number of possible sequences explodes with programme size, so no amount of experience allows anyone to see the best answer by inspection. Eight wells on one rig admit 40,320 orderings. Twenty wells admit about 2.4 times ten to the eighteen. The superintendent's schedule is a good local answer found by a sensible heuristic, and the question is how far it sits from a better one, which is a measurable quantity rather than a matter of respect for experience.
The move cost is the whole game
In manufacturing this structure is called sequence-dependent setup, and the treatment is the same. What varies is the magnitude, because a rig move is one of the most expensive setups in industry.
An onshore move between pads in the same field is measured in days: rig down, transport, rig up, and back to making hole. A jack-up moving between locations in the same field is a similar order. A move between fields adds a tow, a marine spread, weather standby and often a regulatory notification. A move between basins adds a mobilisation and demobilisation charge negotiated in the contract, and possibly a shipyard visit for class or flag requirements.
Two properties make this worse than an ordinary setup cost. The move consumes calendar in a programme that is usually calendar-constrained by lease terms or partner commitments, so a day lost to a tow is a day removed from the drilling season rather than a day added at the end. And the move is exposed to weather, so its duration is itself a random variable with a long right tail in the seasons when most programmes run hard.
Eight wells, worked
Numbers make this concrete faster than the argument does.
Take eight wells across three fields, with three wells in field A, three in field B and two in field C. Moves within a field take 2 days. Moves between fields take 9 days. The rig spread rate, meaning the rig plus all the services running against it, is 180,000 a day.
Clustered sequence: drill all of A, move to B, drill all of B, move to C, drill both. That is five within-field moves at 2 days, which is 10 days, plus two between-field moves at 9 days, which is 18. Total move time 28 days, costing 5.04 million.
Priority-ordered sequence: the subsurface ranking happens to alternate across fields, so the rig ends up making six between-field moves and one within-field move. That is 54 plus 2, or 56 days, costing 10.08 million.
The difference is 28 days and roughly 5 million on a programme of eight wells, produced entirely by the order of a list. Those 28 days also come out of the drilling season, which on a weather-limited programme can decide whether the eighth well gets drilled at all.
Nobody defends the alternating sequence when it is put that way. It happens because the priority list arrives ranked by expected value per well, the scheduler works down it, and the move cost never enters the ranking. The fix is to rank by expected value per well net of the sequence cost that well's position imposes, which requires solving the sequence and the ranking together.
The constraints that make the tidy answer infeasible
Pure clustering is rarely available, and the constraints that break it are worth listing because each one has a different remedy.
Rig capability. Water depth, hook load, mud pump capacity, blowout preventer rating and deck space all restrict which rigs can drill which wells. This shrinks the assignment space and can force a move that a capability upgrade would avoid.
Time windows. Lease expiry, permit validity, partner funding dates, seasonal weather restrictions and environmental exclusion periods around spawning or migration all impose windows. A well with a hard window drags the sequence around it.
Facility readiness. A development well is worthless until the tie-in and the host facility can take it, so drilling ahead of facility readiness converts capital into a shut-in well.
Shared scarce resources. Completion crews, coiled tubing units, specialist tools and marine support serve multiple rigs. Two rigs both needing the same completion spread in the same fortnight is a resource conflict the rig-level schedule does not see.
Information dependencies. An appraisal well's result determines whether and where the next well is drilled. Sequencing an appraisal early has an option value that a pure cost minimisation ignores entirely, and it is often the single largest term in the objective.
That last one deserves emphasis because it runs against the clustering logic. Drilling the appraisal well first may require an extra move and still be correct, because it resolves uncertainty that determines the remaining programme. The way to handle it is to value the sequence under both possible appraisal outcomes and take the expectation, rather than to schedule a deterministic programme and treat the appraisal as just another well.
Duration uncertainty and where the buffer should sit
Well durations are uncertain and the uncertainty is skewed. A well can go faster than planned by a little and slower by a great deal, because stuck pipe, lost circulation, wellbore instability and equipment failure all have long tails. The distribution is closer to lognormal than to normal.
Most programmes handle this by padding each well individually. The superintendent adds three days to every well's plan, and everyone knows the pad is there, which means it gets consumed by the first thing that goes wrong on that well and never protects the well after it.
The statistical argument for pooling is simple and it is worth doing in front of the team. Eight wells each padded by 3 days carries 24 days of protection. If the well durations are independent, the variance of the total is the sum of the variances, so the standard deviation of the total is the individual standard deviation multiplied by the square root of eight, which is 2.83. A single shared buffer at the end of the programme, sized to give the same protection to the programme as a whole, is about 3 times 2.83, or 8.5 days.
Fifteen and a half days of calendar recovered, which at the spread rate above is 2.8 million, from moving the same protection to a different place. That result is the reason for the shared buffer, and the reason it is hard to implement is organisational: an individual buffer belongs to the person who will be blamed, and a shared one does not.
Two cautions on the arithmetic. Independence is an assumption, and it fails when a common cause such as weather or a fleet-wide equipment problem hits several wells at once, which reduces the pooling benefit. And the skew means a buffer sized on standard deviations understates the tail, so the pooled figure should be set from the quantiles of a simulated total rather than from a multiple of a standard deviation.
What the schedule should tell the contracting team
Rig contracts get negotiated on day rate, and the schedule determines what the day rate is actually buying.
The decision worth modelling is whether to hold a rig through a gap in the programme or release it and re-mobilise later. Holding costs the day rate, possibly at a reduced standby rate, for the gap. Releasing costs a demobilisation charge, a mobilisation charge on the next contract, the risk that no suitable rig is available when you need one, and whatever the market rate has moved to in the interim. A 60 day gap at a 40 percent standby rate on a 180,000 spread is roughly 4.3 million if the associated services are also stood down, against mobilisation and demobilisation charges that can be a similar order. The comparison is close enough that it deserves a calculation rather than a convention.
Term length interacts with the sequence in a way that is easy to miss. A longer contract lets you cluster wells across a wider set of fields because you have the calendar to make the moves worthwhile, and a short slot forces you to take whatever is nearest. Run the sequence optimisation at two contract lengths and the difference in total programme cost is the value of the longer term, which is a better negotiating input than a market rate comparison.
Options rather than commitments do the same work here as anywhere. A priced option to extend, exercised against the actual state of the programme six months in, costs a premium and removes the need to guess.
The limit
Optimising a sequence assumes you know the durations, the move times and the well list, and all three are estimates that move. A schedule optimised to the day against inputs with a fortnight of uncertainty is precise about the wrong thing. What the optimisation genuinely delivers is the structure of the answer: which wells cluster, which moves are unavoidable, which time windows bind. Those conclusions are stable across a wide range of input assumptions, and the exact dates are not.
So run it as a comparison rather than as a plan generator. Solve the sequence under three duration scenarios and look at what stays the same. The invariant part is what you commit to, and the rest is left to be decided when the information arrives.
The other limit is that a rig schedule optimised in isolation can be infeasible on materials, on crews or on facilities. A sequence that saves four rig moves and requires two wellhead assemblies to arrive a quarter earlier than their lead time allows is not an improvement. The check has to run against the material plan, which is why the two need to share the same schedule object rather than exchanging spreadsheets.
Take your current programme, count the between-field moves in the planned sequence, and work out how few there could be if the wells within each field were drilled together.