What are the six decision rules for certification?
AIR APAC certifies a person against AIRS-J/1 only when all six rules are met. This page sets out each rule in the Standard's words, what each rule does, and why every rule must be met on its own.
Judgment is how well a person decides once AI has given the answer. Under AIRS-J/1, the Standard for judgment in AI-enabled work, a certification result is decided by six rules. Section 12.2 states the condition in two sentences: "All six must be met. Failure of any one results in Not Certified."
Each observation of what a person did is scored from 0 to 4 (Section 7). Level 2 is the threshold level. At Level 2, the behaviour "appeared unprompted and was adequate to the situation". The six rules read these scores and the evidence behind them. The Standard scores nine dimensions, grouped into four domains: Structure, Test, Engage and Perform. See https://airapac.org/standard/dimensions/.
The six rules
| # | Requirement | The rule in the Standard's words (Section 12.2) | What the rule does, in plain words |
|---|---|---|---|
| 1 | Dimension floor | "No dimension below 2.0" | Each of the nine dimensions, taken alone, reaches the threshold. |
| 2 | Domain floor | "Each of S, T, E, P at or above 2.0 (mean of its dimensions)" | Each of the four domains reaches the threshold on average. |
| 3 | Composite | "Overall mean at or above 2.3 (Section 11.1)" | The mean of all nine dimension scores is at least 2.3, so the overall level sits above the threshold, not only at it. |
| 4 | Evidence | "Section 8.3 satisfied for all nine dimensions" | Each dimension has at least three independent observations from distinct decision points. Two observations from the same decision point count as one. |
| 5 | Stakes coverage | "Section 6.4 satisfied" | The Assessment spans at least three distinct primary consequence families. At least one of them involves irreversible or non-financial harm of the kind Section 6.4 lists. Those are people's safety or welfare, information or privacy, institutional or public trust, and the environment or community. |
| 6 | Confidence | "Moderate or High" | Certification is issued only at Moderate or High confidence (Section 10). Provisional and Insufficient results do not certify. |
What each rule does
Section 12.3 explains the rules "so the rules are not mistaken for redundancy". Rules 1 to 3 govern the level. Rules 4 to 6 govern whether the level is knowable.
- The composite decides a flat profile. A person at exactly 2.0 on all nine dimensions has a composite of 2.0. That profile meets rules 1 and 2 and fails rule 3.
- The floors stop an uneven profile. Three high dimensions could lift the composite above 2.3 while one dimension sits at 1.0. Rules 1 and 2 stop that result.
- Stakes coverage stops a narrow set of scenarios. Rule 5 stops a result earned on financially framed scenarios (Section 12.3). Under Section 6.4, a candidate assessed only on financial and operational scenarios is scored Insufficient Evidence.
- Evidence and confidence decide whether a level can be read at all. Rule 4 sets a minimum for each dimension; rule 6 reads the evidence as a whole. Confidence describes how much evidence there is, not the candidate's ability (Section 10).
How the scores are built
Section 12.2.1 sets the arithmetic. A dimension score is the mean of its valid observations. A domain score is the mean of its dimension scores. The composite is the mean of all nine dimension scores.
Why all six must be met
The rules are non-compensatory. Section 12.1 states: "Strength in one domain does not offset weakness in another." A result that meets five rules and misses one is Not Certified. Section 12.2 gives the rules no order of importance.
After the six rules
- The band. A result that meets all six rules is reported in a band (Section 12.4). The three bands are Certified, Certified with Merit and Certified with Distinction.
- A critical-review pattern. Some Level 0 observations match a critical-review pattern in Annex D. Such an observation makes the affected dimension Insufficient Evidence, pending a targeted reassessment (Section 12.2.2). A single trigger "is not an automatic Not Certified decision".
- A result close to the cut score. Before release for a hiring or promotion decision, AIR APAC checks whether a one-level change in one observation would change the outcome. Where it would, the candidate is offered "one further held-back mission on an unseen form at no charge before release" (Section 12.8.1).
Where the thresholds came from
AIR APAC set the thresholds on the review board's advice as "initial cuts" (Section 11.1). They come from the anchor meanings, with Level 2 as the threshold behaviour, and from synthetic panels (checks V1 to V3). They are reviewed each year against the evidence programme. A change to a threshold is a major version of the Standard.
What this does not claim
- Meeting the six rules does not say that a person is right for a particular job. A result describes the person's judgment in AI-enabled work at the level certified (Sections 2.2 and 12.6).
- The thresholds are initial cuts, reviewed each year. This page does not present them as final (Section 11.1).
- A certified result is reported as a band with a confidence level, not as a raw composite (Section 12.4). It is never a percentage or a rank of people (Sections 12.4 and 12.7).
- This page explains the rules. It is not preparation advice 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.