Evidence checked October 31, 2026. Research models are distinguished from operating results. All numerical examples are hypothetical unless explicitly attributed.
1. The forecast has to survive the trip offshore
A research preprint submitted on October 20, 2026 proposes bringing grid and market information into offshore wind maintenance scheduling. Its idea is that periods of curtailment or low electricity value may offer opportunities to service equipment. The authors present an optimisation approach, not proof that a particular operating fleet has already achieved its modelled benefits. The distinction is important, but so is the question the paper raises: maintenance timing is part of the electricity system's performance. October 2026 maintenance-scheduling preprint
An offshore wind farm is often introduced through its turbine count, nameplate capacity and expected annual generation. Those figures describe a project at a distance. The operating reality includes a vessel waiting for a safe transfer window, a replacement component waiting at a port, a technician interpreting a warning and a control system deciding how much production to sacrifice to protect an asset.
The wind resource remains fundamental. Without enough usable wind, excellent maintenance cannot create the expected energy. But a strong resource cannot compensate automatically for equipment that is unavailable or an export system that cannot carry the output. The annual result emerges from their interaction.
This makes offshore wind reliability a more specific subject than the familiar argument about whether wind varies. Weather-driven variation, equipment failure, planned maintenance, wake losses and grid curtailment have different causes. Treating every shortfall as the same kind of unreliability prevents useful diagnosis.
The article follows the chain from forecast resource to delivered electricity. It asks how much energy the turbines could capture, what reduces that capture, which losses can be changed and what evidence is needed to demonstrate improvement. It does not prescribe vessel operating limits or maintenance procedures; those belong to qualified teams using site-specific safety systems.
The central point is practical. A credible forecast needs an equally credible operating plan. That plan should explain how the project will detect problems, reach the equipment, obtain parts, complete repairs and measure the result. The maintenance vessel is not an afterthought to the energy model. It is one of the ways the model becomes reality.
2. Capacity factor is a result, not a diagnosis
Capacity factor compares actual energy production over a period with the energy that would have been produced at continuous nameplate output. It is useful because it connects capacity to energy. It is limited because many different processes can produce the same number.
The UK government's Energy Trends renewables collection publishes capacity and generation statistics and was updated again on October 29, 2026. Such official series are useful starting points for observing sector output. They do not, by themselves, identify the maintenance history or technical cause of a particular project's performance. DESNZ renewables statistics
Consider a hypothetical 1,000-megawatt offshore wind farm producing 4,380,000 megawatt-hours in a non-leap year. Its capacity factor is 4,380,000 divided by 1,000 times 8,760, or 50 percent. That calculation does not mean the turbines were broken for half the year. They may have been mechanically available for most of it while operating at varying output as wind conditions changed.
Now imagine another farm with the same capacity factor. It could have a stronger wind resource but more downtime. A third could have excellent availability but greater wake losses or curtailment. Equal annual output does not imply equal operational quality, and different output does not automatically imply different maintenance competence.
The denominator also needs attention. A farm that adds turbines during the year should not be analysed as though its final capacity had operated for the full period. Commissioning and partial operation can distort a simple calculation. Comparisons should state how capacity changes and the start of commercial operation are treated.
For a performance investigation, the useful sequence is to estimate the energy available under the actual wind conditions, then account for losses and constraints using consistent definitions. The comparison with a long-term forecast is a separate exercise because the year's weather may differ from the forecast's assumed climate.
A capacity factor is therefore an excellent question starter and a poor question stopper. It tells readers what the farm delivered relative to its maximum theoretical output. Explaining why requires weather information, operating data and a loss-accounting method that does not mix unrelated causes into a single judgement.
3. Availability needs a clock and an energy measure
A turbine can be available during many low-wind hours and unavailable during a few high-production hours. Its time-based availability may look strong while the energy lost is significant. Conversely, maintenance during calm weather may reduce available hours with little effect on annual generation.
Suppose a hypothetical turbine is unavailable for ten hours. If it would otherwise have produced one megawatt during each hour, the lost energy is ten megawatt-hours. If the same outage occurs when it would have produced twelve megawatts, the loss is 120 megawatt-hours. The duration is identical; the energy consequence differs by a factor of twelve.
This distinction creates a need for both time-based and production-based measures. The first describes how often equipment can operate. The second estimates how much output is lost because it cannot. Neither should replace the other. A turbine with frequent short faults may deserve attention even if those faults happen to coincide with low wind during one reporting period.
Definitions also determine which events count. Is a turbine considered available when it is technically ready but the grid connection is constrained? How are planned inspections treated? What happens when a component is derated rather than completely stopped? Different reporting rules can produce different availability figures from the same physical history.
ORE Catapult's SPARTA programme addresses part of this problem through anonymised performance and maintenance data, agreed metrics and consistent reporting. The programme's existence supports the value of benchmarking; it does not make every public availability claim automatically comparable with its internal data. ORE Catapult SPARTA programme
A useful public report should define its availability metric and show the principal exclusions. Where possible, it should report both time and energy consequences. It should also distinguish the turbine from the whole farm, because shared infrastructure can prevent delivery even when individual machines are healthy.
The analytical goal is to avoid rewarding a reporting convention instead of operational performance. If one operator excludes certain events and another includes them, a ranking can measure accounting choices rather than reliability. Clear definitions allow genuine improvements in maintenance and equipment design to emerge from the data.
4. The wake is part of the resource
A turbine extracts energy from moving air and changes the flow encountered by equipment downstream. Within a wind farm, and potentially between farms, those wake interactions affect the energy available to other turbines. A forecast that treats every machine as though it sees undisturbed wind will overstate what the array can capture.
A 2026 Wind Energy Science paper reports airborne measurements used to investigate offshore wind-farm wakes and discusses the benefits and limitations of that measurement approach. It is valuable because it examines the atmosphere directly rather than treating a wake model as self-validating. Its observations still have a particular sampling pattern and set of conditions; they are not a universal loss percentage for every offshore project. 2026 airborne wake-measurement study
The practical complexity is that wake effects vary with wind direction, atmospheric conditions, turbine operation and array layout. A farm can experience different losses under different weather regimes. A single annual percentage may be useful for summarising a forecast, but it conceals the conditions that create the loss.
For a hypothetical example, suppose two wind directions occur equally often but do not contribute equal energy. One direction may align turbines in a way that creates stronger interaction, while another exposes more turbines to relatively undisturbed flow. Averaging directional losses by hours alone can misrepresent their contribution to annual energy if the wind speeds differ.
The model should therefore use the relevant joint distribution of conditions rather than a simple average assembled from unrelated measurements. The same principle applies when comparing measured output with a prediction. An apparent improvement may result from a more favourable mix of wind directions rather than a better control strategy.
Wake analysis also introduces a boundary beyond the individual project. A neighbouring development can change the flow conditions a farm experiences. Assessments should state whether external wakes are included, what future buildout is assumed and how uncertain those assumptions are.
The conclusion is not that wakes make forecasts impossible. It is that wake losses are a physical part of the energy calculation. They should be modelled, measured where feasible and updated as the surrounding development and operating conditions change, rather than hidden inside an unexplained adjustment factor.
5. More energy is not always the only control objective
An operating strategy can affect both energy capture and equipment loading. The best decision for one turbine at one moment may differ from the best decision for the entire farm over its life. That creates a trade-off that should be described rather than compressed into a claim of higher efficiency.
Consider a hypothetical control change that slightly reduces one turbine's output but improves conditions for several downstream machines. The farm's total generation might rise even though the first turbine produces less. Alternatively, a strategy might sacrifice a small amount of near-term energy to reduce damaging loads and improve long-term availability. The sign and size of these effects require evidence for the actual design and conditions.
A convincing trial would define the outcome before evaluating the data. Is success measured as gross generation, net revenue, component loads, outage reduction or lifetime value? A strategy can improve one measure while worsening another. Selecting the most favourable metric after the experiment risks overstating the result.
The comparison group matters too. Wind conditions change over time, so a simple before-and-after output comparison may confuse the control effect with weather. A useful assessment needs an appropriate reference, a method for normalising conditions and a clear treatment of uncertainty.
Equipment condition complicates the test. A new control strategy deployed during an unrelated repair programme may appear responsible for improvements actually caused by restored hardware. Conversely, a beneficial strategy may look ineffective if a major outage occurs during its trial. The analysis should identify such events rather than allowing them to disappear into annual averages.
This article's proposed standard is modest: report the intervention, the comparison method, the measured outcomes and the limits of transferability. A successful trial under one atmospheric regime should not automatically become a guaranteed annual benefit at another site.
The broader operational lesson is that offshore wind performance is a system problem. Maximising the output of every turbine independently does not necessarily maximise the value of the farm. The evidence needs to show how the chosen strategy balances production, loads, maintenance and the obligations of the project as a whole.
6. A repair begins before the technician arrives
The physical repair may be the shortest part of an outage. Before work begins, the operator may need to detect the problem, diagnose it, obtain a component, assign qualified personnel, arrange a suitable vessel and wait for conditions that permit safe access. Each step can add time even when the final repair is technically straightforward.
The May 2024 US offshore wind operations and maintenance roadmap treats maintenance as a broad research and development challenge involving monitoring, repair and logistics. Its date is important: it is a roadmap of needs and opportunities, not a 2026 report proving that those needs have all been solved. US offshore wind operations and maintenance roadmap
For analytical purposes, break an outage into detection delay, diagnosis time, mobilisation time, weather waiting, repair work and return-to-service verification. This decomposition is a proposed way to inspect performance, not a universal industry accounting rule. It reveals which interventions could actually shorten the outage.
If most lost time comes from waiting for a specialised vessel, improving fault detection by an hour may have limited effect on total downtime. If diagnosis repeatedly requires multiple visits, better remote information may have much greater value. If spare parts are unavailable, adding technicians alone cannot restore the turbine.
The sequence can also overlap. A component may be ordered while diagnosis continues, accepting the risk that the initial interpretation is wrong. A vessel may be reserved before the weather forecast is sufficiently certain. Those choices involve costs and probabilities, not simply a preference for faster action.
A useful maintenance report should distinguish repair duration from total outage duration. It should also show whether repeated attempts were needed and whether the same fault returned. A quick repair that does not resolve the underlying problem may produce more downtime over the following months than a longer, more complete intervention.
The public discussion often treats maintenance as an operating expense that can be reduced without changing anything else. In reality, reducing expenditure can either reflect better methods or defer necessary work. The performance record needs to show which occurred. Cost savings are persuasive when they are accompanied by sustained condition and delivery, not when they merely postpone a bill.
7. Weather windows have several dimensions
A usable weather window is not simply a period with low wind. Offshore work may depend on waves, wind, visibility, currents, vessel capability, transfer arrangements and the duration of the task. Conditions must support the operation as a whole, including travel and safe return, rather than just the moment a technician steps onto a structure.
India's National Institute of Wind Energy publishes a metocean study assessing weather windows and weather-related downtime for installation and maintenance at the proposed FOWPI location in the Gulf of Khambhat. Its site-specific character is the important lesson: access assumptions should come from the relevant conditions and activities, not from a generic offshore average. NIWE FOWPI weather-window study
A hypothetical four-hour task may require more than four hours of acceptable conditions once travel, preparation and contingency are included. If the forecast window is too short, starting work can create unacceptable safety and completion risks. This article does not specify operational thresholds; those must be established by qualified teams for the equipment and activity.
Weather persistence matters as well as the fraction of favourable hours. Ten isolated favourable hours are not equivalent to one continuous ten-hour window. A monthly statistic showing that conditions are acceptable half the time can therefore be insufficient for a task requiring a long uninterrupted period.
Forecast uncertainty adds another layer. A model may identify a likely opening, but the decision must account for the consequences if conditions deteriorate earlier than expected. A plan that assumes perfect knowledge of future weather will understate real waiting and mobilisation costs.
An ECN research report on offshore accessibility identifies the complexity of vessel access and its consequences for maintenance downtime. The underlying publication is from 2016, so it should be used for the mechanism rather than as a current fleet-performance estimate. ECN offshore accessibility research
The practical question is whether the project's maintenance model uses realistic sequences of weather and task requirements. A percentage of accessible time is a useful summary, but it cannot replace the operational question: can the necessary people and equipment complete this particular job safely within a usable window?
8. Vessel strategy is an engineering and scheduling choice
Different maintenance tasks require different capabilities. Routine inspections, minor repairs and major component replacements do not necessarily use the same vessel or equipment. A project that has access to one type of service does not automatically have access to all of them.
For a hypothetical farm, compare two arrangements. One keeps a vessel under a long-term contract, paying for availability even when demand is low. The other charters capacity when needed, reducing fixed commitments but risking delay during busy periods. Neither is universally superior. The appropriate choice depends on expected work, distance, weather, vessel market conditions and the cost of lost output.
A maintenance forecast should state those assumptions. If it assumes immediate vessel availability, readers should ask what contractual arrangement supports that assumption. If the project shares a vessel with other farms, the model should account for competing demand, especially when similar weather creates simultaneous access opportunities.
Ports and shore facilities also matter. Transit time, storage, lifting capability and the ability to prepare components can change how much useful work fits into a weather window. An efficient offshore repair can still be delayed by a shore-side bottleneck.
The size of components affects logistics. A replacement part may be technically available but difficult to transport, handle or install with the equipment already contracted. The maintenance plan should connect the turbine design to actual service capability, including plausible major interventions over the project life.
One useful measure is the portion of downtime attributable to logistical waiting rather than active repair. If that share is large, the project may gain more from improved vessel access, inventory or coordination than from marginally faster repair techniques. If it is small, attention may belong elsewhere.
The broader lesson is that reliability is partly designed into the service system. A turbine is not maintainable in practice merely because a manual describes how to replace a component. The operator needs a feasible route from diagnosis to completed work under the conditions at the site. That route should be tested in the project model before the first major failure forces the issue.
9. Corrosion evidence must respect the exposure
Offshore structures operate in an environment where material condition changes over time. Corrosion, protective systems and biological growth need to be evaluated through relevant exposure and inspection evidence. A laboratory result can help explain a mechanism, but it is not automatically a complete prediction of an installed structure's service life.
BOEM reports field observations from two turbines at the Coastal Virginia Offshore Wind pilot project. The study included corrosion coupons exposed for two years and observations of biofouling. Treated and untreated samples behaved differently, and pitting was observed around some biological attachment sites. The study's limited sample and exposure period should remain visible; it does not establish a universal failure rate for offshore wind foundations. BOEM field observations at the CVOW pilot
The analytical lesson is that material performance depends on conditions and protection, not simply on the word offshore. A comparison should identify the material, treatment, exposure zone, inspection method and duration. Without those details, a dramatic photograph or a reassuring test result may be unrepresentative.
Inspection findings also require interpretation. A detected defect does not necessarily imply imminent failure, while an absence of visible damage does not prove that every relevant degradation process is absent. The inspection method has a detection capability and a coverage limit. Those limits belong in the condition assessment.
A hypothetical coating test might show good performance on intact samples but reveal vulnerability at damaged edges or attachments. The maintenance implication is not merely to select the coating; it is to understand how installation, handling and inspection preserve the protection system over time.
Long-term plans should account for the cost and feasibility of access to the surfaces that need attention. A protective strategy that performs well only with frequent interventions may be less attractive at a difficult site than a more durable alternative with a higher initial cost. The comparison should include the whole maintenance pathway.
Corrosion management is therefore part of energy reliability even though it may not appear in a power curve. Structural condition can influence inspection intervals, operating restrictions, repairs and life-extension decisions. The evidence needs to connect material observations to those decisions without extrapolating beyond what the tests and inspections can support.
10. New materials need a field-performance bridge
Research into more durable materials can improve offshore economics, but the path from a promising material to lower lifetime cost includes several tests. Manufacturing consistency, installation, compatibility with other components, inspection and repairability all influence whether the laboratory benefit survives in service.
EDF's February 2025 account of the MAREWIND project describes a four-year materials programme that finished in early 2025, addressing durability challenges in offshore wind. The project is evidence of organised research and development. Its summary should not be read as proof that every resulting material has already delivered decades of commercial service. EDF account of MAREWIND
A useful performance claim states the comparison. Was the new material tested against an existing product under the same conditions? What failure mechanism was measured? Did the experiment reproduce the relevant temperature, moisture, mechanical loading and exposure? How does the test duration relate to the proposed service life?
Accelerated testing can be valuable, but acceleration requires a defensible relationship between the test and real conditions. Increasing stress may reveal a failure mechanism that is not dominant in normal service, or omit interactions that develop slowly. A forecast based on such testing should state how the extrapolation is justified.
Manufacturing scale can introduce another difference. A material made carefully in a small research batch may behave differently when produced in large volumes with ordinary process variation. Quality control and acceptance criteria help determine whether the intended properties are consistently achieved.
Repairability matters because offshore access is expensive and constrained. A component that lasts longer but requires an unusually difficult repair may still be attractive, but the trade-off should be evaluated. A product that simplifies inspection or local repair can create value even without a dramatic increase in nominal strength.
The strongest evidence bridge combines laboratory understanding, representative field exposure and operating records. Each stage answers a different question. Research should be credited for the stage it has reached, while claims about lifetime cost should include the uncertainty remaining between that stage and a mature maintenance history.
11. Predictive maintenance is a decision system
A sensor can detect a change before a component fails, but detection alone does not restore the asset. The value of predictive maintenance depends on whether the signal leads to a better decision than the operator would otherwise have made. That requires a connection between measurement, diagnosis, planning and action.
Consider a hypothetical warning system that correctly identifies many developing faults but also produces frequent false alarms. If each alarm triggers a costly offshore visit, the system may consume resources without improving availability enough to justify them. If operators become accustomed to false alarms, they may discount a genuinely important warning.
The opposite failure is also possible. A highly selective system may produce few false alarms while missing problems until repair options are limited. Evaluating it requires more than a headline accuracy figure. The relevant measures include missed events, warning lead time, false alarms and the consequences of the actions taken.
The base rate matters. If serious faults are rare, even a seemingly accurate classifier can generate many false alerts relative to true ones. A useful evaluation should report the number of actual events and the operating exposure, not just a percentage calculated from a selected dataset.
A model trained on one turbine design or operating environment may not transfer directly to another. Sensor changes, maintenance history and control software can alter the patterns it sees. The deployment plan should include monitoring for such changes and a method for reviewing the model's continuing usefulness.
The operational test is whether earlier information changes an achievable plan. If a warning arrives before a convenient weather window and a part can be obtained, it may prevent a prolonged outage. If no suitable vessel or component is available, the same warning may have less immediate value, though it can still support safer planning.
A credible predictive-maintenance claim therefore reports outcomes at the decision level: avoided failures where supportable, improved scheduling, reduced unnecessary visits or better condition knowledge. It should not equate the existence of a dashboard with demonstrated reliability improvement. The system succeeds when the information leads to useful action under real offshore constraints.
12. Curtailment should not become a maintenance alibi
Grid curtailment and equipment downtime can overlap. If a turbine is serviced while the farm is unable to export, the energy cost of that maintenance may be lower than at another time. But the overlap must be accounted for carefully, or the same lost energy can be counted twice or disappear from both categories.
The October 2026 preprint introduced at the start proposes using grid and market signals in maintenance planning. Its contribution is a modelled scheduling framework, with theoretical results under stated assumptions. Treating that as a proven fleet-wide saving would outrun the evidence. The useful next question is how such a method performs with imperfect forecasts, actual access constraints and real maintenance tasks.
A hypothetical example illustrates the accounting. Suppose a farm would otherwise generate 100 megawatt-hours during an interval, but the grid allows only 60. A maintenance intervention affecting 20 megawatt-hours of potential production might cause no additional export loss if the remaining turbines can still supply the permitted 60. Under another operating arrangement, it might reduce delivery. The result depends on the constraint and the distribution of available generation.
The operator should therefore distinguish technical availability from incremental energy loss. A turbine under maintenance is not technically available merely because curtailment masks the energy effect. At the same time, charging the full unconstrained production loss to maintenance can overstate its economic consequence during that interval.
Forecast uncertainty limits opportunistic scheduling. A predicted curtailment period may not occur, or a low-price interval may end before the work is complete. The plan needs to account for that uncertainty rather than assuming the grid provides a perfectly timed free maintenance window.
Safety and equipment requirements remain primary constraints. Market conditions cannot justify extending work beyond safe limits or deferring necessary intervention indefinitely. The optimisation problem should reflect those boundaries explicitly.
The broader point is that good maintenance planning can align with system conditions, but the performance claim must preserve the underlying categories. Grid restrictions, equipment condition and scheduling choices should remain separately visible so that an apparent improvement can be traced to its actual cause.
13. Reliability data need an exposure denominator
A count of failures is difficult to interpret without knowing how many machines operated and for how long. A larger fleet may report more failures while having a lower failure rate. A young fleet may have different problems from an older one. A change in reporting practice may alter the count without changing physical reliability.
For a hypothetical comparison, ten failures across 100 turbine-years of exposure imply a different observed rate from ten failures across 1,000 turbine-years. The simple rates are 0.1 and 0.01 failures per turbine-year respectively. These figures still require context about component type, severity, operating conditions and repeated failures on the same equipment.
A turbine-year is itself only a starting denominator. Larger turbines, different duty cycles and different designs may not be directly comparable. Component-hours or another exposure measure may be more appropriate for a particular analysis. The report should explain why its denominator matches the mechanism being studied.
Survivorship can also bias a dataset. If poorly performing equipment is removed from the sample, the remaining fleet can appear to improve even without better maintenance. If newer machines enter the fleet, the age distribution changes. A fair trend analysis should account for those shifts.
Censored observations require care. A component that has not failed by the end of the study is not proven to last forever; its observed life is simply incomplete. Conversely, a component replaced preventively may never produce a failure event even though its condition influenced the decision. The maintenance history helps interpret both cases.
Public data may not contain enough detail to calculate robust design-specific rates. The honest response is to state that limitation, not to infer a precise failure probability from a few highly visible incidents. Individual events can reveal mechanisms and operational consequences without representing the entire fleet.
The most useful benchmarking programmes preserve consistent definitions, exposure and severity categories. They make it possible to distinguish fewer failures, shorter outages and lower repair costs. Those are related achievements, but they are not interchangeable measures of reliability.
14. A forecast should publish its loss chain
A transparent energy forecast would show how the project moves from the wind resource to electricity at the contractual delivery point. It would identify wake assumptions, turbine performance, electrical losses, availability, environmental constraints and curtailment treatment. The exact categories can vary, but their boundaries should prevent omissions and double counting.
Multiplicative losses deserve particular care. In a hypothetical calculation, applying a 10 percent loss and then a separate 5 percent loss to the remaining energy leaves 85.5 percent of the starting value: 0.9 multiplied by 0.95. Simply subtracting both percentages leaves 85 percent. The numerical difference is small in this example, but the underlying accounting convention should be explicit.
More importantly, the losses may not be independent. Bad weather can increase the resource while restricting access to a failed turbine. A grid constraint may coincide with high production. A maintenance plan can shift some outages into low-value periods. A model that treats every loss as an unrelated annual percentage may miss these relationships.
The forecast should distinguish uncertainty in the resource from uncertainty in the operating model. Additional wind measurements may reduce one type of uncertainty without improving knowledge of component reliability. Better maintenance records may improve availability estimates without resolving external wake assumptions.
A useful sensitivity analysis changes the assumptions that matter most rather than varying every input by the same arbitrary percentage. If vessel waiting dominates the downside, examine access and charter scenarios. If wake uncertainty dominates, examine the relevant atmospheric and neighbouring-development assumptions. The results should guide further evidence collection.
Once the farm operates, the same loss chain provides a structure for explaining deviations. Did the wind differ from expectations? Were wake losses larger? Did equipment downtime exceed the plan? Was export constrained? The answer should be supported by measurements and a consistent attribution method.
A forecast becomes more valuable when it is designed to be checked. Its purpose is not merely to produce a bankable-looking annual number before construction. It should establish a baseline against which the operator and public can understand what the project actually delivers and why.
15. The practical test is repeated delivery
Offshore wind reliability will improve through many specific gains rather than a single universal breakthrough. Better component design can reduce failures. Better monitoring can improve diagnosis. Better vessel and spare-part arrangements can shorten waiting. Better forecasting and scheduling can reduce the energy consequence of necessary work. Each improvement should be measured at the point where it changes performance.
The October 2026 maintenance research is interesting because it connects operational decisions with grid conditions. The 2026 wake measurements are useful because they test assumptions about the resource reaching turbines. Field corrosion observations matter because they connect material behaviour with real marine exposure. These are complementary pieces of evidence, not interchangeable proof that an entire project's forecast is correct.
For electricity buyers, the strongest account will combine them in an operating history. It will show output against weather-adjusted expectations, define availability, explain major outages and distinguish equipment losses from grid constraints. It will connect maintenance expenditure to condition and delivery rather than treating lower spending as an automatic success.
For communities and policymakers, the same discipline helps separate installed capacity from dependable infrastructure. A project can add substantial clean energy while still requiring complementary resources and network arrangements. Recognising those needs does not erase its contribution; it describes the system in which that contribution becomes useful.
The next generation of offshore projects should be judged by how well their forecasts survive contact with the sea. That includes the ordinary details: whether a replacement part fits, whether the vessel is available, whether the weather window lasts, whether the inspection finds the problem early enough and whether the repaired machine stays in service.
Those details are not less important because they make less dramatic headlines than a giant turbine. They determine how much of the promised annual energy reaches the meter. The most credible progress is repeated delivery supported by a maintenance system that can explain its losses, learn from them and reduce the ones within its control.