Prepare before it publishes
Our method, labelled as ours, for the work the standard will not do for you.
4 pages in this group. Every page states its citation, the evidence a reviewer asks for, and what it does not cover. Last updated: 2026-10-21.
Write the sampling and perturbation plan
Before any run, write down which inputs are sampled, which perturbations are applied, how the sample is split into groups, how many draws are taken, and what result would count as a failure.
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 yo
Assemble the evidence pack a reviewer will test
Bring the system description, the method statement, the plan, the run logs, the results, the uncertainty and the limits into one pack, indexed so a reviewer can follow a claim to a record. L
Map the work to the AI management system and the risk process
Place the statistical robustness work inside the management system: the impact assessment at A.5.2, verification and validation at A.6.2.4, data quality at A.7.4, the documented information,