EVIDENCE & LIMITS · DESIGN NOTE
Adaptive Questions Are Not Inferred Answers
Context can help choose the next question. It must not be used to write an answer on a person’s behalf.
Adapt the path, not the response
A reflective conversation can use a person’s stated goal and visible gaps to decide whether to pause, explain a construct, or offer a smaller question bundle. That is adaptation of pacing and sequence. It is not evidence that the assistant knows how the person would answer an item.
Noesis proposes a strict separation: context can inform which question is offered next; an assessment response exists only when the person explicitly provides it. A model summary, a prior conversation, or a plausible inference is not a substitute response.
Keep score provenance visible
When a score is shown, a person should be able to distinguish explicit answers from explanatory text and from any optional confidence note. If an answer is missing, the honest state is missing—not a guessed value that quietly completes a scale.
This distinction supports meaningful review, correction, and withdrawal. It also prevents adaptive language from being mistaken for psychometric evidence.
A cited source is not product validation
Testing standards and adaptation guidance help frame what should be documented when a measure changes language, administration, scoring, or interpretation. They do not independently validate a particular product, translation, norm, or individual conclusion.
For that reason, a public evidence record should name the source, version, scope, and limitation together. It should not turn a citation into a promise about accuracy or a life decision.