Evidence of performance

A performance claim specifies asset, useful service, period, boundary, unit and comparison method. Record changes in comfort, product quality, output and reliability separately. Lower consumption following a production shutdown requires a different explanation from improved efficiency at unchanged output.

Every figure has an origin: measurement, calculation, assumption or external estimate. Preserve these statuses. Expand “20% savings” into its numerator, denominator, observation window and adjustments. This makes the claim’s meaning testable before any recalculation.

Equipment-level measurement helps connect an intervention to its effect but requires accounting for interactions with the rest of the asset. A whole-building meter covers the overall result while responding to other load changes. A calibrated model may help when an appropriate baseline is unavailable, provided inputs and validation are explicit.

Method choice depends on effect size, available data and verification purpose. IPMVP provides a framework of measurement and verification options; record the edition and selected option. M-01 illustrates adjustment of baseline consumption to reporting-period conditions.

Align energy, weather and operating data to identical intervals and a common time zone. Check gaps, duplicates, meter replacement, multiplier changes and impossible values. Preserve raw files and a processing log. Data corrections must be reproducible.

Degree-days require a consistent base temperature and calculation method. Operating days need a stable definition. If within-day operation varies substantially, hours or output may be more suitable predictors. Select predictors for physical relevance and evaluate them on data excluded from model fitting.

M-01 coefficients are supplied by the exercise. Six monthly intercepts give 6 × 400 = 2,400 kWh. Weather contributes 12 × 1,200 = 14,400 kWh. Operation adds 80 × 120 = 9,600 kWh. The adjusted baseline is 26,400 kWh.

A real project estimates and validates coefficients before applying them to the reporting period. Check signs, units and applicability. Extrapolation to temperatures or output outside the observed range needs separate justification. Post-intervention data must not silently refit the original baseline.

M-01 savings equal 26,400 − 21,000 = 5,400 kWh, or 20.45% of the adjusted baseline. Dividing by actual consumption would produce a different percentage with a different meaning. State the denominator alongside the result.

For monetary valuation, apply the agreed tariff to the corresponding intervals. With time-varying prices, multiplying annual energy by an average rate can misstate value. Record capacity services, technical costs and deferred replacement separately. Keep energy reduction and cost change visible as distinct measures.

Performance estimates contain measurement, baseline-model and processing uncertainty. Combining them depends on correlations and the evaluation method. The exercise supplies aggregate standard uncertainty of 1,200 kWh. Under the assumed normal approximation, 5,400 ± 1.96 × 1,200 gives an interval from 3,048 to 7,752 kWh.

The 1.96 multiplier belongs to this approximation. Small samples, dependent residuals or another model require a suitable method. The lower endpoint can inform an explicitly agreed conservative teaching settlement scenario. It is not a universal payment or certification rule.

For every claimed effect, propose another possible cause: warmer weather, shorter hours, another equipment repair, a sensor change or a different product. Identify observations that distinguish explanations. A control asset is useful when conditions are comparable and interactions are understood.

A strong baseline fit does not establish causation. Examine the intervention, timing and concurrent changes. Where required data is missing, narrow the conclusion to the observed change with stated uncertainty. Plan the next experiment to address a specific evidence gap.

The package contains raw data, a variable dictionary, processing description, versioned model, result calculation, uncertainty, change log and claim limitations. Separate calibration and validation data. Record preparation date and the owner of each source.

A reviewer should reproduce the result without oral instructions from its author. Correct errors through a new version with a change history. The next season adds another observation period. Real-asset materials can become a teaching case after authorized de-identification and checking reuse rights.

Prepare a protocol for a real intervention: claim, boundary, baseline and reporting periods, data dictionary, raw series, reproducible processing, model validation and uncertainty evaluation. Add at least three alternative explanations for the observed change.

Another agent must recalculate the result and identify claim limitations. Practical defence requires an implemented project’s data and verification that the useful service was preserved. M-01 supports method practice and alone does not establish real-asset competence.