Using official sources without sounding like an authority

A product can rely entirely on data published by institutions and still remain independent. Citing an authority does not transfer its mandate, certainty, or right to speak on its behalf.

The distinction sounds obvious. In practice it is easily lost, especially when several sources are combined on one screen and the product calculates new indicators or summarises the situation in a sentence.

I encountered this problem while working on Dunărea, an independent monitor of what official institutions publish about the river.

A source is official for a particular purpose

An institution may be the right source for a measurement, an act, or an administrative area and have no authority over a broader conclusion. For each source, I record:

• who publishes it and in what capacity;
• what it measures or describes;
• the area and period it covers;
• how often it is updated;
• the units and method it uses;
• reuse and attribution requirements.

Without this information, two similar-looking series can be compared incorrectly. A clean interface cannot repair a meaningless comparison.

Keep the label beside the value

In a data product, a value may be measured, estimated by a model, or calculated by the application. All three may be useful, but they are not interchangeable.

The distinction should survive from import to the final screen. For each value, I want the source, publication time, transformations, and quality state. If the label is added only in the interface, different kinds of data may already have been mixed upstream.

For example, today's output from a hydrological model can be compared with the history of the same model. It should not be presented as a physical gauge measurement. Likewise, a percentile calculated by the product should expose its reference period and minimum sample count.

Missing data is information too

Many products hide gaps. A field disappears, an old value looks current, or an estimate silently fills the place of a measurement.

I prefer the product to say, “we did not find this value in the sources we checked.” It can also retain the last check, the institution that may publish the information, and how much the gap affects the conclusion.

In Dunărea, the missing-data register is not an apology page. It is part of the method. It helps separate a missing detail from a gap that could change the result.

Calculations should be reproducible

When a conclusion can be expressed as a clear rule, I prefer a deterministic one. Conversions, time windows, thresholds, and refusal conditions can be documented and tested.

AI can help find documents, extract data, or write an explanation. It should not quietly replace a calculation that can be stated and verified directly.

For a public indicator, the path should be reproducible:

1. retain the document or response received from the source;
2. normalise the data without losing provenance;
3. check dates, units, and required fields;
4. apply a documented, versioned rule;
5. generate the explanation from the calculated result;
6. retain enough information to reproduce the outcome.

Observation, effect, and decision are different claims

A hydrological condition, a confirmed effect on infrastructure, and a risk to the energy system are three different statements. They use different sources and may be confirmed by different institutions.

A useful product can place them on the same page, but should not collapse them into one alarm. The user needs to see what was observed, which effect was confirmed, and which decision still belongs to an operator or authority.

State independence plainly

An independent product should explain which sources it uses, how it transforms the data, and where its conclusions stop. That boundary belongs in the introduction, method, and data labels, not only in a footer note.

Credibility does not come from looking “official.” It comes from allowing someone else to verify the source, calculation, and limits of the claim.