How does a person name the ways a decision could fail, and who pays? (T3 Risk and Failure Anticipation)
What T3 asks a person to do before committing, in the Standard's words, and one fictional decision that shows it.
Judgment is how well a person decides once AI has given the answer. T3 is one of the three dimensions of the Test domain in AIRS-J/1.
What the Standard says
T3. Risk and Failure Anticipation. "Identifies plausible failure modes, the pathway to each, who bears the cost, and which harms are being accepted rather than mitigated." (Section 5)
In plain words: before committing, the person names how it could fail, who pays and which harms are being accepted.
The published anchors (Section 7.5) describe what appears in the transcript (Section 7.3):
- Level 2, the threshold: "Names, unprompted, at least one plausible failure mode, the pathway by which it would occur, and who carries the cost."
- Level 3: "As Level 2, and distinguishes reversible from irreversible failure, and states which harm is being accepted rather than mitigated."
- Level 4: "As Level 3, and names a leading indicator (an observable that would appear before the failure) and who is responsible for watching it."
Its Level 3 note: "naming an accepted exposure is not a weakness in the answer. Failing to name it is."
One fictional decision, worked through
A fictional example. Illustrative figures.
The situation. Arvind Raj leads customer service at a broadband provider in Kuala Lumpur. He has one week to approve a plan to move billing questions from phone to chat.
What the AI produced. BateleurAI drafted the plan: "Moving billing questions to chat from 1 March saves MYR 1.2 million a year. Based on this year's call volumes and the cost of each contact by phone and by chat. I am 89 percent sure." It names one risk: the saving is lost for each month of waiting.
The check. Before signing, Arvind writes down how the change could fail. A customer who cannot use chat, such as older customers or some with a disability, cannot reach a person about a wrong bill before the payment date. The bill goes unpaid and the service is cut off. The customer pays first, then the company, in complaints and refunds. The days without service cannot be given back; a longer chat wait can be reduced.
The decision, and who was told. Chat starts on 1 March, with a phone option for any customer holding a cut-off notice. The plan states that longer chat waits in the first month are accepted, not mitigated. Its leading indicator is customers who contact chat twice about one bill within 7 days; the billing team lead watches it weekly. Arvind tells the finance director that the saving is smaller, then tells the team leads.
What happened after. Three months later: repeat contacts rose in week 2; the team lead found a confusing chat menu and fixed it within a week. About 300 customers a month used the phone option.
Writing down the accepted wait is the act the Level 3 note names.
What this does not claim
- This section explains one dimension. Reading it confers nothing toward certification.
- The example is made up. Arvind holds no level, band or result; those come only from an Assessment.
- Certification rests on all nine dimensions and the decision rules, not on T3 alone.
AIR APAC publishes the standard for judgment in AI-enabled work, certifies people against it through its official instrument, and stands behind every result.