The dashboard says zero. Nobody measured it.
A missing observation can become a confident answer as it passes through software. The person reading the final screen may never see what was lost.
An absent measurement deserves a visible place in the model.AI-generated conceptual setting.
Consider two fictional sensor records.
In the first, a measurement was taken and the result was zero. In the second, no measurement arrived. The dashboard shows zero for both.
A person looking at that screen could reasonably believe both checks happened. The uncertainty has vanished, even though nobody resolved it.
This is one of the questions behind the scientific-software assurance work in my research material. Much of it concerns detailed rules and reference cases. The human issue is easier to recognise: a clean answer can make a person stop looking for information that is still missing.
Follow the idea
Which record actually says zero?
A fictional sensor example. It contains no patient, medicine, dose or clinical recommendation. The comparison illustrates missing information, not a validated product.
Unknown and measured zero are different states.
No measurement arrived
The record is blank. It does not establish that the measured quantity was zero.
Matching the default in a test will not recover the source meaning.
The dashboard supplies a number
A default turns the blank into zero. Calculations and tests may accept it, but the screen now implies a measurement happened.
Unknown and measured zero are different states.
Keep the question visible
Show that the value is unavailable and why, when known. A real measured zero still remains usable.
The answer gets more certain on its way to the screen
The first record says something about the world. The second says something about the limits of the record. They deserve different treatment.
Yet each part of a software pipeline can make a locally convenient choice. An empty field becomes a default value. A calculation accepts it. A chart plots it. A summary describes the chart.
No one step has to look dramatic. By the time the result reaches a reader, it can be difficult to recover the original gap.
That is particularly relevant when AI is helping produce explanations as well as code. If it receives only the finished number, an otherwise fluent account can explain a certainty the source never contained. Better phrasing cannot recover information that was discarded earlier.
The FHIR vocabulary for absent data makes distinctions such as unknown, not asked and temporarily unknown. That standard is specific to healthcare, but the distinction is understandable without knowing the standard: a failed measurement, an unanswered question and a result of zero are different events.
The test can agree with the mistake
Suppose a developer writes a test expecting a blank field to produce zero. The program passes.
The test has established consistency with the chosen rule. It has not established that the rule is a faithful interpretation of the source.
The work therefore needs a place where someone can challenge the expected answer before it becomes the thing every later check protects. A reference case can preserve the original record, the accepted interpretation and the reason for it. If the interpretation changes, the code and its tests need to follow.
Google’s code-review guidance asks whether tests are valid. Here, that means being willing to question the answer in the test, even when changing it breaks a passing build.
The opposite mistake is possible too. A genuine zero should remain a usable value. Replacing every difficult case with “unknown” would lose information just as surely as inventing a number.
An unanswered question still needs a useful interface
A blank box alone may leave the reader wondering whether the page is broken. A useful display can say that a measurement is unavailable, show why when the reason is known, and identify what would resolve the gap.
That lets the person decide whether to obtain the missing information, proceed with an explicit limitation or stop the dependent calculation. The interface has preserved a choice that a false zero would remove.
The source material includes a planned assurance approach and a separate research harness using fictional records. Its reported checks are evidence about that method. They are not clinical validation or proof that a released product follows it. This article uses no patient, medicine or dose example.
The point of the blank is not to look cautious. It is to keep an unanswered question available to the person who still has to make a decision.