From Reading to Evidence

Why this lesson matters

Most sites can produce numbers. Far fewer can defend them. The gap between the two is what this lesson is about, and it is becoming the operative constraint as energy, water and emissions figures move from internal interest to external obligation.

Six properties that make a reading evidential

1. Traceable identity. The reading resolves to one physical asset with a permanent key, a known location and a known measurement configuration. Free text descriptions do not qualify, because they cannot be shown to refer to a single instrument.

2. Verified time. A system generated capture timestamp with timezone, distinct from any later entry or synchronisation time.

3. Unit and basis. The unit, the multiplier, whether the value is cumulative or instantaneous, and for gas whether volume is corrected or uncorrected. Any conversion applied downstream is recorded with its factor and the source of that factor.

4. Source artefact. A photograph of the register or dial, an automated data record with device identity, or a system export with its origin. Something that exists independently of the person asserting the value.

5. Quality status. Verified, estimated, carried forward, unreadable, or superseded. An estimate that is labelled as an estimate is acceptable evidence. An estimate that is presented as a measurement is not.

6. Immutability with adjustment history. The record is append only. Corrections are recorded as adjustments carrying the original value, the revised value, the reason, the person and the time. Nothing is ever silently overwritten.

A reading with all six is evidence. A reading missing any of them is a number that may well be correct and cannot be demonstrated to be.

Accuracy and defensibility are different

This distinction catches people out. A reading can be perfectly accurate and entirely indefensible. An auditor is not assessing whether you are honest; they are assessing whether the record can be verified without relying on your assertion. That is why the source artefact carries so much weight relative to its cost.

Coverage and the explicit denominator

The most common weakness in a site energy or water figure is an unstated denominator.

coverage = readings received / readings expected

Expected comes from the asset's reading schedule and the reporting period. Report both numbers:

"Gas consumption for FY26 Q1, from 48 of 52 scheduled weekly readings across 4 assets. Coverage 92 percent. Four missing readings estimated by interpolation; method and affected periods noted."

That statement is defensible. "Gas consumption for FY26 Q1 was X" is not, because nobody can tell what proportion of the period was actually measured.

Handling gaps properly

Gaps are normal. Concealed gaps are the problem.

• Interpolation between two verified cumulative readings is legitimate for a cumulative register, because the total is bounded by the two indices. State the method and flag the derived values.
• Pro rata estimation from a comparable period is acceptable when clearly labelled and justified.
• Carrying forward a value is not estimation. It is fabrication that is indistinguishable from measurement, and it is the single most damaging habit in manual monitoring.
• Leaving the gap visible is always preferable to filling it invisibly.

Every estimated value should be individually flagged, so that a reader can compute what proportion of the reported total is measured and what proportion is derived.

The evidence chain

For any reported figure, you should be able to walk downwards without a break:

Reported figure
  -> reporting period and boundary
    -> activity category (gas, electricity, water, diesel)
      -> site
        -> asset
          -> reading events, with coverage stated
            -> source artefact for each reading
              -> adjustment history for each reading

If the chain breaks at any level, the figure above it inherits that weakness. In practice the breaks are almost always at the bottom two levels, because that is where the cheap effort was skipped.

Data quality as a measured thing

Do not treat quality as a subjective judgement. Measure it and report it alongside the figure:

• Reading coverage - received against scheduled, per asset and per period.
• Source evidence rate - the proportion of readings with an attached artefact.
• Exception rate - readings flagged out of range, implausible, or requiring review.
• Adjustment rate - the proportion of readings subsequently corrected, and the net effect of those corrections.
• Timeliness - readings captured within their scheduled window plus grace period.
• Instrument verification status - the proportion of readings taken from instruments within their calibration or verification interval.

These metrics have a second function beyond reporting: they tell you whether your monitoring process is degrading before the data becomes untrustworthy.

Retention and reproducibility

• Retain readings, source artefacts and adjustment history for at least the period your obligations require, and be aware that energy and emissions frameworks commonly expect several years.
• A report should be reproducible: running the same query over the same period should return the same figure, and if it does not, the adjustment history must explain the difference.
• Keep the factors and methods used in any conversion, with their version and source, alongside the report. A figure computed with a superseded factor is not wrong, provided you can show which factor was current at the time.

Where the effort actually goes

A useful rule of thumb from sites that have done this properly: making the data evidential costs far less than people expect, because five of the six properties are produced automatically by the capture process. The only one requiring genuine discipline is quality status, specifically the willingness to record an unreadable instrument as unreadable.

The expensive alternative is retrospective evidence assembly, which is what most organisations currently do at the end of a reporting period, at high cost, under time pressure, with a worse result.

Common mistakes at this stage

• Reporting a figure without its coverage denominator.
• Filling gaps invisibly, particularly by carrying values forward.
• Editing readings in place rather than recording adjustments.
• Storing source images outside the record, where they cannot be associated with a specific value.
• Treating quality as a judgement rather than as a set of measured metrics.
• Assembling evidence retrospectively, which is always more expensive and always produces a weaker result than capturing it at the instrument.