Requirement
When a surface presents content that models a real-world artifact — a standard, a form, a policy document, a report — the content must be faithful to that artifact: correct scope, terminology, and structure. Where full fidelity is not yet available, the content must be explicitly labelled illustrative or placeholder, in-product and in the decision record. "Format is right but the specifics are invented" is a fail for any surface used in user testing, because testers judge credibility on the specifics, not the shell.
Rationale
The triggering incident: a mock report graded outcomes for a category that does not apply at that level, and showed an outcome worded in a way no practitioner would recognise — both read as fake to a real practitioner. The first case was caught and fixed by hand; the wording case remained flagged as illustrative pending a domain reviewer's confirmation. Two evaluator passes graded this ad hoc against the sprint contract because no control named the failure mode — a naming/voice/tone-clean surface can still read as fake if the domain details inside it are wrong. CNT-1..3 grade error anatomy, naming, and voice mechanics; none of them checks whether the content is true to the thing it claims to model.
Passes when
- A domain detail matches the real-world artifact it models (correct level, subject, terminology, structure) — verifiable against the source, or by a named domain reviewer.
- Content that cannot yet be verified is explicitly labelled illustrative/placeholder, both in the UI and in the decision record.
- A named domain reviewer (for example, a team lead in that domain) has signed off before the surface is shown to real practitioners for user testing.
Fails when
- A category or level detail a practitioner would recognise as wrong — e.g. a category graded at a level where it does not apply.
- Invented specifics presented as real in a surface used for user testing, with no illustrative label.
- Placeholder content ships with no label at all, so a tester cannot tell it from real data.
Evaluator guidance
Judgment, with a domain-reviewer dependency the evaluator cannot substitute for: read the content against the real-world artifact it claims to model (a standard, a form, a policy). Where the evaluator lacks the domain expertise to judge a detail directly — most domain specifics will be exactly this — say so and flag the item for a named domain reviewer rather than guessing. Confirm either (a) a named domain reviewer's sign-off is recorded before the surface reaches user testing, or (b) the content carries an explicit illustrative/placeholder label, in-product and in the decision record. A surface with neither is a finding regardless of how polished it otherwise reads.
Do not flag
- Content explicitly labelled illustrative/placeholder, in-product and in the decision record — the label is the control working as intended, not a defect to chase away.
- A detail that looks unusual to a non-specialist evaluator but carries a named domain reviewer's recorded sign-off. The reviewer's judgment settles the question; do not override it with a generalist read.
- Settled, verified reference facts restated verbatim from an authoritative source (for example, a quoted regulation line) — the fidelity requirement is about invented specifics, not about the source material itself.