The less familiar the domain, the harder checking is — and that is exactly when people ask the system

What it is

Someone filing taxes for the first time most needs the system to fill deduction lines, and is least able to see that “home-loan interest” was written as another class of deduction. People who can check can often fill the form themselves. The expertise–verification paradox is that checking skill and the motive to ask run in opposite directions: the less domain knowledge, the harder errors are to see — and that is why the task was handed to a generator.

Whether there is enough clock time is a different object. What is missing here is discriminability, not minutes.

Why it happens

Checking depends on a “this does not look real” pattern: legal ranges, usual pairings, impossible combinations. Those patterns come from domain experience. Novices have no such base map and can only check the surface: does the sentence parse, is a field empty. Generators almost always pass on the surface, so the novice toolkit is empty.

The motive to ask comes from the same gap. People rarely hand over control on tasks they can do; on tasks they cannot, the generator becomes the only path that looks complete. A design that defaults to “the user will verify” builds its safety on the group least likely to start it.

Studying it

Split people by a domain-knowledge pretest into novice / expert. Give the same error-bearing artefacts and free checking (search, ask someone, open a source). Dependent variables: detection rate, checking strategy (surface vs. comparison with an authority), rate of adopting the error. Independent variables: whether a novice-usable comparison is provided (official clause text side by side, not “please search”).

Do not let a student sample stand in for novices. Students may lack this course, not the ability to search. A true novice condition is: not knowing which keywords to search.

Where it stops holding

When a strong external authority exists and the interface opens a comparison aligned to the sentence, novices can check “is this word-for-word the same” without domain knowledge — provided the object exists and is aligned. Pure aesthetic or preference tasks have no true/false, so the paradox does not apply. Experts under fatigue or time pressure can behave like novices; that is an attention-budget problem, not a knowledge gap. This entry does not treat whether high-stakes checking should be left with the person on scene.

Applying it

• For tasks novices do often, do not rest correctness on “please verify.” Provide a side-by-side with an authority, or restrict fillable values to a legal set.
• Beside the output, say “you may not be able to judge this yourself” when a pretest or the task type already shows the user is a novice.
• Treat “I came because I cannot do this” as the default portrait, not the expert who can check.
• Check: give real novices for that task an error-bearing output and an open network. If they detect almost nothing yet are willing to submit, the checking mechanism is empty for actual users.

Related

• Same group: L3.03.1 Fluent wording is not the same as being correct · L3.03.2 The cost of checking can exceed doing the work oneself · L3.03.3 High-stakes settings must not leave checking entirely to the user · L3.03.4 When errors sit inside correct content, checking is sentence by sentence and costs about as much as rewriting · L3.03.6 Certainty of wording has no relation to reliability of content · L3.03.7 Errors in numbers, dates, and names are the hardest to notice and the most damaging · L3.03.8 Putting the checking duty in a disclaimer does not reduce how far errors actually spread
• Nearby: L3.02 Source Attribution · L1.02 Expressing Capability Boundaries
• Search terms: expertise–verification paradox · novice fact-checking · knowledge gap