1. What exactly is being counted?
Write down the unit and definition before making a comparison. A customer account, a connection, a household and a person need not be the same unit. Capacity and energy generated are different quantities. A change in definition can alter a result even when the underlying situation is unchanged.
The practical test is whether another analyst could identify the same denominator and reproduce the calculation from the documentation.
2. When was it observed?
Separate the observation period from the release date and the day the file was downloaded. A recently published report can describe earlier observations. A series can also mix periods. Record those differences before labelling a comparison “current”.
For a tariff, the relevant date is the period when the schedule applies—not simply the date on which it was found. ERA’s tariff catalogue identifies schedules by supplier and applicable period, illustrating why those details belong with the input.1
3. Whose experience does it represent?
Identify the geography, population and selection process. Supplier records may describe that supplier’s customers. A household survey may aim to represent a wider population, subject to its sampling and response limitations. Neither should quietly inherit the scope of the other.
Ask who is absent. The people who left a service, could not be reached or never became customers may be central to the decision.
4. How was the value produced?
Keep direct observations, administrative records, self-reports, estimates and model outputs distinguishable. Document adjustments, assumptions and missing information. A value generated by a model is not necessarily unsuitable; it simply needs to be interpreted as a model output.
When combining sources, examine whether their methods are sufficiently compatible. A unified table does not, by itself, create comparability.
5. What are you permitted to do with it?
Record the applicable licence or access terms. Permission to read a file is not automatically permission to republish it, distribute individual records or incorporate it into a commercial product.
For example, ENERGYDATA.INFO describes many resources as CC BY 4.0 while identifying exceptions. The practical step is to check the selected resource, not assume that a portal-wide label answers every reuse question.2
6. What decision can it support?
Name the decision and the required level of detail. A national indicator may provide context without resolving a neighbourhood-level expansion question. A historical average may not resolve a near-term operational risk.
Then state the boundary: what can reasonably be concluded, what cannot, and what additional evidence would materially change the answer. This prevents the analysis from promising more than its inputs can support.
A useful evidence record says not only where a number came from, but what someone may responsibly do with it.
Make the answers travel with the data
Grid’s evidence-card builder turns these questions into a small documentation record. It does not verify the source, certify a licence or award a quality score. Its purpose is simpler: help ensure that essential context is not lost when information moves between people, files and decisions.
Sources for the examples
- Electricity Regulatory Authority. Tariff schedules. Catalogue reviewed 21 October 2026; individual tariff values not reproduced or audited.
- ENERGYDATA.INFO. About the platform and terms-of-use summary. Reviewed 21 October 2026.
The six-question checklist is a proposed Grid working method. The examples illustrate specific documentation issues; they do not constitute an audit of the publishers’ datasets.