Set the acceptance threshold and the decision rule
Fix the metric, the baseline, the threshold and the action that follows a pass or a fail. The public text names robustness properties but states no threshold, so the number must come from your risk appetite, your quality model or a contract.
The requirement in plain English
Fix the metric, the baseline, the threshold and the action that follows a pass or a fail. The public text names robustness properties but states no threshold, so the number must come from your risk appetite, your quality model or a contract.
What the public record says
The public text states no threshold, no metric and no pass mark. It places robustness inside a quality model and describes assessment as one step to build trust.
Why it matters to you
Without a threshold, a result cannot fail, and an assessment that cannot fail is not an assessment. The threshold is a risk decision, not a statistical one, which is why it belongs to the risk owner and needs a date and a signature. The decision rule also protects the team: it says in advance what happens if the result is adverse, so the conversation after the run is about the action, not about the number.
Evidence a reviewer asks for
- The metric definition and the baseline.
- The threshold, the risk owner's signature and the date.
- The decision rule for a pass, a marginal result and a fail.
- The escalation path and the remediation owner.
Checks to run this month
- Is the threshold signed by the risk owner?
- Is the baseline recorded on the same data?
- Is the action after a fail written down?
Dates
Set before the first run. Review at the evidence freeze, 2027-01-17.
Cross-references
- Write the sampling and perturbation plan
- Uncertainty, complexity and model confidence
- Map the work to the AI management system and the risk process
What this page does not state
No threshold, metric or pass mark is stated in the public text.