Where the physical account changed the decision

In March 2024, a storage proposal reached our technology review described as a twenty-megawatt, eighty-megawatt-hour system. The commercial presentation treated that pairing as a four-hour delivery promise. I asked the team to distinguish the nominal battery energy from the energy available at the agreed AC delivery boundary. That distinction changed the specification before it became a purchase commitment.

Our comparison used an explicitly illustrative year-ten condition. After the study's allowances for degradation, the operating reserve, conversion losses and auxiliaries, usable delivered energy was sixty-four megawatt-hours. This was a supplier-supported design assumption for evaluation at the specified temperature, not an observed operating result. Under that assumption, twenty megawatts lasted 3.2 hours; a four-hour interval implied sixteen megawatts. Neither calculation established availability across every temperature or proved performance throughout its operating life.

The development team preferred to preserve the larger power promise and address the energy shortfall during detailed design. Asset operations director Maren Solbrecht wanted a duty that the operating organisation could understand before taking responsibility for it. I recommended that we compare a sixteen-megawatt four-hour duty with a larger energy configuration, using the same delivery boundary and assumptions. This gave finance two defined alternatives rather than one attractive label with an unresolved addition behind it.

I authorised a €780,000, eleven-week qualification and specification programme. We retained the lower-power reference duty for procurement evaluation and postponed the commercial commitment while the alternatives were priced. The consequence was a six-week change to the tender calendar and a smaller headline power figure in the base case. We did not report a constructed asset, secured revenue or a saving from avoiding an addition that had never been ordered.

My contribution was to make the obligation testable. Specialists remained responsible for the detailed degradation model, electrical design and acceptance procedure. I required the board paper to state which number was nominal, which represented delivered energy and which conditions supported the proposed duty. If I repeated the work, I would settle that vocabulary when the commercial service was first proposed. Waiting until technology review allowed a convenient division of two nameplate values to acquire the authority of an engineering commitment.

In September 2024, we reviewed a candidate grid-forming inverter configuration for a future storage installation. The supplier's product description could not answer our site-specific question about its behaviour within the proposed electrical system. I sponsored a controlled hardware-in-the-loop investigation in which a physical inverter interacted with a real-time model of the selected network conditions. The laboratory arrangement was a way to examine interactions; it was not the completed plant.

The engineering team defined eight scenarios before the review. Seven met the project team's stated response criteria. One produced a discrepancy between the expected and measured settling behaviour that the team could not yet explain. We did not interpret seven satisfactory scenarios as an eighty-seven-and-a-half-per-cent probability of successful operation. The scenarios were selected investigations, not a random sample of every condition the asset might encounter.

The project sponsor wanted to proceed with the original qualification schedule and retain the unresolved case as a commissioning item. Maren Solbrecht argued that an unexplained interaction should be understood before it was transferred to an operating site. I supported a narrower continuation: retain the candidate, fund examination of the model and controller behaviour, and withhold the recommendation for wider application. Rejecting the entire technology would also have gone beyond what the evidence established.

The original twelve-week investigation cost €1.3 million. I approved a further €460,000 and six weeks for model verification, revised settings analysis and repetition of the affected scenarios. At the January 2025 review, the engineering group had resolved a modelling mismatch and repeated the original eight scenarios against the agreed criteria. That supported the next project-specific study stage. It did not certify compliance, confer connection approval or establish the configuration's suitability across our generation portfolio.

The delayed recommendation was uncomfortable because the laboratory programme had been presented as the final confirmation of an already attractive choice. I changed the internal review language so that an unexplained result could alter the plan without being described as a failed programme. My retrospective improvement is more concrete than asking colleagues to accept uncertainty: the funding request should have included an explicit allowance for investigating disagreement between the model and the hardware. That would have made the learning task visible before the first reassuring result arrived.

A solar-development review in February 2025 compared two array sizes behind the same one-hundred-megawatt AC inverter capacity. The alternatives were one hundred and ten and one hundred and thirty-five megawatts of DC module capacity. The larger array increased the amount of energy available during many lower-irradiance periods, while also increasing the periods in which the inverter's power limit constrained output. A higher clipping figure did not by itself tell us which design was preferable.

I asked for one reconciled annual simulation rather than a chart that showed only the extra module capacity. With the same weather file, availability assumptions and non-clipping loss methods, the larger array added a modelled 35.6 gigawatt-hours before the additional clipping loss. The extra clipping was 11.2 gigawatt-hours, leaving 24.4 gigawatt-hours of incremental modelled AC energy at the inverter output. Grid curtailment and downstream losses were outside that stated boundary and had to be assessed separately.

The procurement team preferred the smaller array because the module order was lower and its clipping chart looked cleaner. Development colleagues preferred the larger array but initially quoted the pre-clipping increment. I rejected both shortcuts. I recommended carrying the larger configuration into detailed commercial and connection assessment, with the extra module cost and the complete energy account shown together. I did not recommend construction solely because the energy increment was positive.

The eight-week comparison cost €620,000 within my technology-development authority. Its completed result was a documented design basis and a shortlist decision, not an operating-generation result. Finance still needed to examine the value of output in the relevant hours, while the project team retained responsibility for land, detailed electrical limits and connection conditions. The model did not establish an investment return or a promise that the eventual plant would export the simulated energy.

The most useful change was in the decision discussion. Colleagues could see why some clipping might accompany a stronger production case without treating clipping as free. They could also challenge whether the incremental energy justified the added equipment. Looking back, I would have required a common output boundary in the first request for alternatives. The initial slides made reasonable people disagree partly because one was discussing modules, another inverter energy and another saleable output. A shared comparison brought those separate questions back into a sequence that could be examined.

In June 2025, an industrial heat-recovery opportunity arrived with an appealing coefficient of performance. The proposal involved a twelve-megawatt useful-heat duty. At the reference source and delivery temperatures, the engineering study estimated four megawatts of electrical input for the agreed system boundary, giving a heating coefficient of performance of three. The customer's required delivery temperature and the seasonal condition of the source stream made that one operating point an incomplete description of the service.

I commissioned a comparison at the customer's warmer delivery condition. At that point the same twelve-megawatt useful-heat duty required an estimated five megawatts of electricity within the same boundary, giving a coefficient of 2.4. Both values included the defined compressor and auxiliary loads used in our evaluation. They were engineering estimates at stated conditions, not measured seasonal averages. The lower ratio did not make the technology unsuitable; it changed the electricity requirement we had to take into the service proposal.

The commercial team wanted to retain the more attractive reference figure in its opening offer. Maren Solbrecht asked whether the equipment would meet the actual temperature duty without relying on an unstated operating change. I backed the customer-duty comparison and required a separate account of the source-stream availability, electrical capacity and backup arrangements. I was not willing to let a useful technology depend on a customer interpreting a reference point as an all-year promise.

Our fourteen-week, €940,000 study produced a revised technical offer with a performance map and explicitly bounded operating conditions. It also moved part of the proposed equipment decision into a later verification stage. The resulting proposal carried a larger electrical requirement than the original reference slide, and we accepted the risk that it would be less commercially attractive. No equipment was represented as installed through this study, and we claimed neither annual carbon savings nor a reduction in the customer's energy bill.

My role was to connect the heat duty, the technology assessment and the terms that colleagues were preparing to discuss. Specialist engineers remained responsible for the thermal design and verification. In retrospect, the sales handover should have required the customer's temperature profile before anyone circulated the reference coefficient. That small change would have made the difficult condition part of the question from the start, rather than something engineering appeared to introduce after the opportunity had gained support.

In February 2026, I approved an investigation of wake steering at one of our onshore wind sites. The proposal was to alter how selected turbines faced the incoming wind so that the combined output of the relevant group might improve. A loss at an upstream turbine could be compatible with a gain downstream. That made a single machine's output an inadequate basis for deciding whether the control approach deserved further work.

The authorised first phase comprised eighty planned observation blocks across two selected wind-direction sectors. Engineers set the eligibility rules before examining the result, including operating availability and the wind conditions needed for a usable comparison. Sixty-two blocks met those rules. Alternating the control condition supported comparison, but it did not eliminate every atmospheric difference or turn two sectors into a complete annual wind distribution. The excluded eighteen blocks remained in the trial account with their reasons.

By the July review, one sector supported further investigation of a group-level improvement; the other remained inconclusive within the team's stated uncertainty. I did not publish a percentage uplift or multiply the encouraging sector's result across a year. The commercial sponsor wanted an annual production allowance for the next portfolio plan. Asset operations wanted a decision that also considered turbine loads, equipment limitations and the supplier's operating permissions. I supported extending the investigation rather than inserting an unsupported production assumption.

The initial five-month phase cost €870,000. I authorised a further €390,000 for a four-month extension covering additional conditions and specialist review of the load and operating evidence. At September 2026 that extension was still under way. The completed first-phase result was therefore an improved understanding of where the control might merit evaluation, accompanied by a bounded next decision. It was not a fleet rollout, a warranty conclusion or evidence that every turbine could be operated in the proposed way.

This case tested my willingness to protect a promising experiment from the demands of an annual planning cycle. I had to explain why continuing to spend money was reasonable while declining to book a production gain. My next-time change would be to agree the permitted planning use of the trial results before the measurement campaign starts. Doing that would give the project sponsor a clearer expectation of what an encouraging observation can support, and what remains a separate engineering and commercial judgement.