The baseline is a model of normal operation

A baseline is a representation of how the relevant system would operate without the proposed measure. It is not merely the oldest bill available or the largest consumption total that can be found. A bill records energy delivered within a billing boundary during particular conditions. The baseline needs to explain the conditions that make that record relevant to the decision. Production, operating hours, product mix, weather, maintenance and shutdowns can change consumption without any technical efficiency improvement. A defensible baseline makes those influences visible before assigning an expected saving.

The first task is to define normal operation for the decision being considered. Normal does not mean perfectly stable or free from unusual events. It means that the operating conditions represented by the baseline correspond to the service and horizon used in the appraisal. A facility planning an expansion should not treat an old low-output year as the counterfactual for a larger future operation. A facility reducing production should not assign the resulting lower energy use entirely to new equipment. The evidence record should explain the reference operation and the reason it represents the relevant alternative.

Equipment boundaries help turn a facility total into a testable mechanism. A measure targeting compressed air should identify the compressors, treatment equipment, distribution losses and users included in its saving claim. If a meter captures other loads, the analyst needs a way to distinguish the targeted service from those loads. Conversely, an equipment meter can omit a supporting system whose demand rises after the intervention. The boundary should follow the causal mechanism rather than stop wherever metering happens to be convenient. The analyst may need an explicitly incomplete record while the missing measurement is commissioned.

Rated power alone is a poor substitute for operating evidence. Equipment can spend much of its available time idle, run at variable load or operate below its design condition. Multiplying a nameplate by an assumed duty can produce a plausible annual total while concealing the very behaviour a control measure is intended to change. The baseline record should distinguish measured consumption, measured operating state and inferred demand. An inference can be useful when its basis is transparent. It should not acquire the status of measurement because it appears in the same worksheet.

Production is especially important when the saving mechanism concerns intensity. A lower energy requirement per useful output can coexist with higher total energy use if output increases. A lower total can coexist with worse intensity if the facility runs at inefficient low load. Neither finding is contradictory. They answer different questions. The baseline should retain the service denominator and the absolute carrier balance where both matter. An investment committee may value increased production, but the resulting demand must remain in the emissions and commercial assessment rather than disappear behind an intensity improvement.

A useful evidence record distinguishes provenance from confidence. Provenance states where a reading, contract term or engineering estimate came from. Confidence states the reason it is adequate for the intended decision. A meter record can have clear provenance but weak confidence because its calibration, completeness or boundary is unresolved. A supplier estimate can be clearly attributable but unsuitable for a commitment because it assumes a duty the operator cannot maintain. Keeping these fields separate prevents a document reference from being mistaken for assurance.

The record should also preserve exclusions. A planned shutdown, temporary product trial or meter failure may warrant excluding a period from a normal-operation baseline. That choice needs an explanation of the event and its consequence. Excluding only high-consumption periods can understate normal demand; excluding only low-consumption periods can overstate the saving opportunity. The issue is the coherence of the operating representation. An exclusion should be justified by the reference operation and applied consistently, rather than selected because it improves the investment result.

Where adjustments are necessary, the adjustment rule should be declared before evaluating savings. Routine changes might be associated with production or ambient conditions within an established operating range. A structural change, such as adding a process line, can require a separate treatment. The analyst should avoid allowing an adjustment to absorb every disappointing outcome. If the rule is chosen after the performance result is known, it can become an explanation that cannot be falsified. A baseline plan should state how change will be detected, who authorises the adjustment and which evidence supports it.

Uncertainty should be attached to the relevant baseline term rather than expressed as a general feeling about the project. Missing production records affect the ability to distinguish intensity from output changes. Uncertain meter coverage affects the relation between the reported total and the targeted load. An unresolved operating plan affects the future reference case. Those uncertainties call for different actions. Adding a contingency to capital does not repair an ambiguous load boundary. Reducing the saving assumption does not establish whether the same service is delivered. The evidence gap determines the next investigation.

The annual electricity engine uses a supplied baseline demand rather than estimating one from raw operating records. This is an important interface choice. It allows the numerical study to focus on how measures transform an explicit annual demand, but it places the burden of baseline selection outside the computation. A user must not infer that the calculator validated meter quality, production comparability or normal conditions. The completed illustrative record below makes that limitation visible instead of leaving the supplied demand as an unexplained number at the top of a form.

A baseline can be reproducible without being representative. Another analyst might follow the same extraction rules and obtain the same annual total from a retained dataset. That is useful for detecting arithmetic or selection errors. It does not prove that the chosen period corresponds to the relevant future operation. Reproducibility and representativeness should therefore be reviewed separately. The former concerns whether the record can be reconstructed; the latter concerns whether the reconstructed record answers the intended decision question. A retained dataset helps both reviews but does not settle either by itself.

The baseline should contain an account of deterioration and replacement. If existing equipment is near the end of its usable life, continuing present operation indefinitely may be an inappropriate alternative. The comparison may need to distinguish ordinary replacement from the incremental transition feature. Conversely, assigning improvement from an already committed maintenance programme to a new proposal can double count its contribution. The operator should identify which actions are discretionary, which are already planned and which are necessary to maintain service. This turns the baseline into a decision alternative rather than a frozen photograph.

A baseline does not have to contain every possible future condition. It should contain enough information to test conditions that can materially change the measure's effect. An annual estimate might be adequate to reject an obviously uneconomic concept before expensive instrumentation. A load-shifting proposal may require temporal records from the outset because its mechanism depends on timing. A process conversion may require quality and temperature evidence because total energy alone cannot establish comparable service. Proportionality concerns the causal question, not a universal amount of data to collect for every proposal.

The evidence record can support useful disagreement. Engineering may regard a supplier's duty assumption as technically plausible, while operations knows that production constraints prevent that duty. Finance may prefer a conservative baseline, while the measurement team warns that an arbitrary reduction undermines the adjustment method. The record should capture the disputed proposition and the evidence that could resolve it. Averaging opinions does not necessarily produce a defensible operating case. Where an unresolved alternative is plausible, retain separate cases and show how the commitment depends on the choice.

After commissioning, the baseline becomes part of a continuing comparison. The organisation should retain the original service perimeter, adjustment rules and reasons for exclusions, alongside any authorised revision. Otherwise a later analyst can report a change in consumption without knowing whether it represents savings, a production change or a new boundary. A transparent baseline history supports learning when a measure underperforms. It also protects a valid performance result from being dismissed merely because operating conditions have changed. The record gives both success and failure an interpretable reference.