What does the Test domain ask of a person?
The Test domain of AIRS-J/1 looks at how a person checks evidence, including evidence an AI system produced, before acting on it. This section sets out its three dimensions, the threshold for each, and what the domain does not assess.
Judgment is how well a person decides once AI has given the answer. AIRS-J/1, the Standard for judgment in AI-enabled work published by AIR APAC, describes it in nine dimensions grouped in four domains.
Where Test sits among the four domains
- Structure asks the person to name the real decision and to set what is fixed, open and unknown.
- Test asks the person to check the evidence for an answer before acting on it.
- Engage asks who must agree, who must know, and what the person may decide alone.
- Perform asks the person to commit in proportion to the evidence and to take responsibility for what follows.
The Standard gives the Test domain one line: "Interrogating evidence, including machine-generated evidence." (AIRS-J/1 Section 5). In plain words, evidence is what shows that an answer is supported, or wrong, and to check is to look for what would show it is wrong.
The sentences come from Section 5 of AIRS-J/1 and the anchors from Section 7.5. Each sentence describes the whole dimension. The Standard scores some parts only at higher levels, so each Level 2 anchor asks for less than its sentence. The scale runs from Level 0 to Level 4, and Level 2 is the "Threshold level" (Section 7). All three Level 2 anchors below use the word "unprompted": the person does it without the exercise pointing towards it. A response that appears only after prompting can score no higher than Level 1 for that observation (Section 7.1).
T1. Evidence Verification
The Standard's sentence. "Traces claims to source. Distinguishes generated, retrieved, asserted and verified content. Applies verification effort proportionate to consequence."
Plain reading. The person traces a claim back to where it came from. The person can tell whether a statement was generated by a system such as an AI model, found in a source, simply stated by someone, or checked. The person checks more when more depends on the claim.
The threshold (Level 2). "Checks the load-bearing claim against an available source before acting, unprompted. Distinguishes generated content from retrieved content." A load-bearing claim is one the decision rests on: if it is wrong, the decision changes.
T2. Assumption and Alternative Testing
The Standard's sentence. "Surfaces the assumptions a decision rests on, names competing explanations, and takes action capable of discriminating between them, including against their own preferred position."
Plain reading. The person names what the decision takes to be true. The person names at least one other explanation. The person then takes a step that can show which explanation holds, even when that step could show the person's own view to be wrong.
The threshold (Level 2). "Names, unprompted, at least one assumption the decision depends on and at least one competing explanation, and takes an action capable of discriminating between them."
T3. Risk and Failure Anticipation
The Standard's sentence. "Identifies plausible failure modes, the pathway to each, who bears the cost, and which harms are being accepted rather than mitigated."
Plain reading. The person names how the decision could fail, how each failure would come about, and who would pay. The person also says which harms are being accepted rather than reduced.
The threshold (Level 2). "Names, unprompted, at least one plausible failure mode, the pathway by which it would occur, and who carries the cost."
Why setting an AI answer aside unread is not checking it
A note under the T1 anchors reads: "rejecting AI output wholesale without examining it is a failure to verify, not caution" (Section 7.5). A person who sets an AI answer aside without looking at it has not checked it. That sits at Level 0, beside acting on an unverified claim that the decision rests on when checking was possible.
Under Section 7.3, an assessor may not score on inferred motivation, attitude, confidence or personality. So the anchors describe what a person does, as it appears in the recorded session. The domain scores whether the check appears there, not whether a person seems careful.
How Test is decided for certification
Certification is non-compensatory: strength in one domain does not offset weakness in another (AIRS-J/1 Section 12.1). The Test domain score, the average of T1, T2 and T3 (Section 12.2.1), must be at or above 2.0, and no single dimension may be below 2.0. These are two of six rules in Section 12.2. All six must be met, including an overall average of 2.3 or above.
What this does not claim
- The Test domain does not assess what a person knows, skill with any AI tool, knowledge of a professional field, intelligence or personality (AIRS-J/1 Sections 2 and 2.1).
- This section gives no person a level. Levels come only from an Assessment, the certifying session. Every reading is scored by published rules and traces to what the person did.
- It shows the Level 2 anchors only. It is not preparation material for an Assessment.
AIR APAC publishes the standard for judgment in AI-enabled work, certifies people against it through its official instrument, and stands behind every result.