On this page
A chart can make a result understandable. It cannot, by itself, establish that the result has been calculated from the right data, using the right operation, for the intended question.
The distance between a number and a decision
In many analytical applications, the visible result is the end of the journey. A table or chart shows the number, perhaps accompanied by a short explanation. That is usually sufficient for a casual enquiry. It becomes less satisfactory when the result is surprising, contested or going to influence an important decision.
At that point, the reader may need to know what question was actually answered, which fields were used, what calculation took place and whether there are reasons to treat the outcome cautiously. A reassuring sentence does not answer those questions.
The design problem is how to keep the supporting material available without making every user follow a technical audit trail to read an ordinary answer.
Evidence is a responsibility, not a visual ornament
I have approached this as an architectural separation. The analytical calculation is established through the governed execution path. Supporting evidence and result information are then attached to that process. Explanation and presentation may render or summarise what has been established, but they should not create a new factual basis.
That separation allows the same governed result to be approached at different levels. A person may begin with a sentence stating what was found, examine a chart or table, look at the calculation and, where supported, follow the underlying record of the analysis.
The purpose is not to force everyone to become an auditor. It is to make scrutiny possible when it matters, and to avoid presenting confidence as a substitute for evidence.
Correct, important and relevant are different questions
A calculation can be correct but unimportant. It can reveal a change that is too small to warrant attention, or a difference that adds little to what is already known. Conversely, a modest movement may matter greatly in a particular context.
That is why determinIQ does not treat every valid analytical output as a finished insight. The architecture includes qualification and attention processes through which supported results may be considered for greater prominence. These processes do not give the presentation layer authority to rewrite the underlying numbers.
It is reasonable for a governed discovery system to return no notable findings. An interface that must always produce something noteworthy risks confusing a requirement to fill a space with the existence of genuine significance.
Evidence must survive communication
The analytical journey rarely ends with the person who first asked a question. Findings are discussed in meetings, included in reports and shared with people who were not present for the original analysis.
As soon as a result moves into a new document, there is a temptation to keep the conclusion and lose the conditions on which it rests. The result becomes a polished statement whose origin is difficult to recover.
The reporting model in determinIQ has been developed around preserving the connection between governed analytical outputs and their explanatory context. A report is intended to communicate established findings rather than introduce new analytical truth. Different audiences may need different levels of explanation, but they should not receive incompatible calculations dressed in different language.
This is a design and implementation principle, not a claim that every conceivable export format has already achieved identical provenance behaviour. Each publishing route needs its own checks, especially where a long-lived document can outlast the software view from which it originated.
What can be demonstrated
The current application has distinct result, evidence, historical record and reporting components. The October 2026 white paper documented the progressive-disclosure approach, including the conceptual path from an answer to its explanation, evidence, calculation and provenance. It also documented the principle that the final business judgement belongs to a person or organisation, not to the analytical engine.
The public claim should remain proportional to that evidence. It is fair to describe the architecture and supported ways of inspecting calculations. It would not be fair to suggest that no erroneous conclusion is possible: incomplete source data, an incorrectly understood question or misleading business assumptions can still lead people astray.
The lasting value is the ability to distinguish those questions. When an outcome is in doubt, we should be able to ask whether the issue arose in the source data, the interpreted request, the permitted calculation or the explanation. Without that separation, the answer is much harder to challenge usefully.
Beyond the answer
This study connects to two other parts of the determinIQ work. Natural-language interpretation needs a contract before calculation because a precise calculation for the wrong question is still wrong. Privacy needs explicit boundaries because evidence should be inspectable without unnecessarily revealing personal information.
The wider research question is whether these properties can coexist with an accessible interface. The point is not to make every answer look more official. It is to leave people with an understandable result and a practical way to establish what supports it.
Research context
Research programme: Making data analysis understandable, reproducible and accountable
Topic: Evidence and verification
Explore further
Supporting material is listed below.
Further reading is listed below.