Keep the qualification that changes the decision.

A defined quarterly plant account in 2023 contained 240 GWh of gross generation and 6 GWh consumed for station service, giving 234 GWh net generation. The proposed executive tile called the 240 GWh figure net output. The number came from the generating-terminal account, but its new label described a boundary it did not represent.

Elin Vardesund challenged the mismatch during reconciliation. The reporting sponsor wanted one prominent output measure and preferred the larger, already familiar figure. I wanted the account to preserve the physical distinction before the number travelled into a wider performance discussion. Calling the 6 GWh difference loss would also have been too casual. Station service described internal electricity use, not a demonstrated quantity of avoidable waste or a reduction the data team could promise to eliminate.

I chose to retain gross generation, station service and net generation as related but separately labelled quantities. EUR 2.1 million supported reconciliation and reporting work. We documented the measurement basis and the period so that a later reader could understand the arithmetic and its purpose. Plant operations and engineering retained responsibility for the underlying equipment and internal consumption; commercial and finance colleagues retained settlement and revenue interpretation. The data function did not acquire those responsibilities by making the account clearer.

We gave up a simpler headline that would have overstated what the generating-terminal total meant. The delivered result was a coherent boundary account, not additional electricity or a measured efficiency improvement. Nor did the 234 GWh figure establish how much was billed or ultimately delivered to end customers. The case mattered because an ordinary label had been doing more work than the source definition could support. I wanted the useful summary to remain connected to the physical quantities it described, even when that required showing more than one number.

A 2021 observation pilot followed one ordinary twenty-four-hour day through ninety-six quarter-hour intervals. Ninety-two intervals had measured energy observations and four were missing. The proposed report would fill the gaps with zero and present the resulting sum as the day's generation. That would have converted a lack of observation into a physical statement about output.

The operating-report sponsor preferred the uninterrupted daily series because it was easier to compare and export. I understood the need for a usable account, but zero was an energy value, not a neutral way to say that a reading had not arrived. A filled series could appear complete while concealing which portion came from measurement and which portion came from an assumption. The distinction would become harder to recover once only the daily total remained.

I authorised EUR 1.3 million of collection and reporting work to retain the measured intervals, missing-status flags and any explicitly estimated contribution separately. An estimate, where used for a defined purpose, needed its own method and label; it could not become a measured value through aggregation. We did not publish an invented complete-day total in this case. Plant and commercial owners retained decisions about operations and the suitability of an account for their particular uses.

The selected response gave up the appearance of a fully measured day. It supplied a more honest interval account and made the remaining uncertainty available to the reader. No generation was recovered by changing a flag, and no claim about lost output followed simply from the four gaps. My contribution was to prevent an apparently harmless filling convention from becoming an unsupported physical conclusion. The work also reminded me that the qualification has to survive aggregation: a daily figure should not erase the observation conditions of the intervals from which it was assembled.

In 2008 I approved an auxiliary-meter increment by subtracting a 980 kWh register reading from a later 20 kWh reading. The resulting minus 960 kWh entered the reporting account. I had applied a familiar difference calculation without examining the register's documented behaviour or the event recorded between the observations.

Mats Hellerby checked the maintenance record. The register wrapped at 1000 kWh, and one rollover was confirmed during the interval. On that evidence the increment was 40 kWh: twenty to the wrap point and twenty after it. My first inclination had been to classify the negative result as an anomaly for later investigation while preserving the rest of the report. I accepted that the explanation was already available and that the derived value needed correction before it was reused.

EUR 420000 supported review and correction. We retained the original readings, the documented register limit, the confirmed event and the derived increment together. The response did not create a universal rule to add 1000 kWh whenever a reading decreased. A reset, replacement, multiple rollovers or another source problem could require a different account. The single-wrap interpretation belonged to this documented interval, and its validity depended on that evidence rather than the convenience of a positive result.

The consequence was a corrected consumption account and a more explicit explanation of how it had been derived. It was not an increase in generated energy, an avoided cost calculation or a claim that every metering issue had been resolved. The mistake was mine to describe because I had approved the interpretation. I learned to ask what happened to the measuring instrument between observations before allowing a simple subtraction to stand for the physical process. The numbers had been copied accurately; the account I built from them had still been wrong.

A 2016 joint-venture wind asset had 240 MW nameplate capacity, with ownership divided 60%, 25% and 15%. Vardelune operated the facility and held the 60% share. The portfolio presentation moved between 240 MW and 144 MW without making clear that one described gross operated capacity and the other applied an ownership weighting.

The portfolio sponsor wanted one capacity number for every audience. That preference would simplify the presentation, but it could also make two legitimate questions appear to have one answer. The plant's nameplate capacity did not shrink when the report described the company's ownership share. Conversely, operating the whole facility did not make Vardelune the sole owner. Neither capacity figure was a quantity of electricity generated during a period.

I chose to preserve gross operated capacity and the ownership-weighted indicator as separate fields and views, with EUR 1.8 million in portfolio-data work. We retained the ownership basis beside the calculation rather than ask the reader to infer it from a short total. Finance and legal owners remained responsible for accounting treatment and rights; the data account did not settle consolidation, dispatch entitlement or the commercial distribution of output. The 144 MW indicator was arithmetic applied to a specified share, not a claim about those other matters.

We gave up the convenience of a single unqualified portfolio figure. The result was a clearer account of why the same physical asset could appear in two differently defined measures. There was no increase in installed capacity and no new generation outcome from the reporting change. I wanted the executive discussion to begin with the intended basis rather than spend its time reconciling apparently conflicting numbers whose definitions had been removed. That lesson remains relevant when describing Vardelune's current 4.6 GW gross operated capacity: the label is part of the figure, not optional decoration.

A 2025 forecast review combined eight hundred delivery hours into a single proposed day-ahead accuracy account. Five hundred predictions had been issued eighteen hours before their delivery interval, while three hundred were updates issued thirty minutes before it. Both groups were predictions made before delivery, but they represented different operational lead times and opportunities to incorporate newer information.

The analytical sponsor wanted to retain the pooled view because it offered a larger sample and one result to communicate. I thought the day-ahead label would misrepresent what was being assessed. A near-delivery update can be useful for its own purpose without demonstrating the quality of the earlier forecast. The problem was not that the shorter-lead predictions were invalid; it was that their performance would be used to describe a different forecasting task.

I supported EUR 2.9 million of evaluation work to retain issue time, delivery interval and lead-time grouping in the review. We separated the five hundred eighteen-hour cases from the three hundred thirty-minute updates rather than publish the pooled account as day-ahead evidence. The forecasting owners remained responsible for model assessment, and trading and operating colleagues retained decisions about use. The case provides no replacement accuracy score or demonstrated reduction in imbalance cost.

The cost was losing a convenient single result and accepting a more qualified discussion of the available observations. The output was a comparison basis that kept the intended use visible. I did not assume that the shorter-lead group must be more accurate or that either sample alone represented every condition the business would face. My role was to prevent different forecasting questions from becoming interchangeable because their outputs shared a unit. A power-data account becomes more useful when it tells the reader how much time remained to act on the prediction, not merely that the prediction came before the outcome.