Clause 8, Selecting datasets representative of a robustness dimension
The public titles cover the general case, the datasets needed for a robustness assessment, the acquisition of data, and the measurement of adversarial robustness. The body text is withheld. The titles show that dataset choice is treated as part of the method, and that adversarial robustness has its own measurement step.
The requirement in plain English
The public titles cover the general case, the datasets needed for a robustness assessment, the acquisition of data, and the measurement of adversarial robustness. The body text is withheld. The titles show that dataset choice is treated as part of the method, and that adversarial robustness has its own measurement step.
What the public record says
The public table of contents lists clause 8 and subclauses 8.1 General, 8.2 Datasets needed for robustness assessment, 8.3 Acquisition of data for robustness assessment and 8.4 Measuring adversarial robustness. The body text is not public.
| Subclause | Public title | Body |
|---|---|---|
| 8 | Selection of datasets representative of a robustness dimension | Withheld |
| 8.1 | General | Withheld |
| 8.2 | Datasets needed for robustness assessment | Withheld |
| 8.3 | Acquisition of data for robustness assessment | Withheld |
| 8.4 | Measuring adversarial robustness | Withheld |
Why it matters to you
Every statistical claim is a claim about a sample, and the sample is a dataset decision. The clause titles make that explicit: the dataset has to represent the dimension being tested, it has to be acquired lawfully, and adversarial robustness has its own measurement. That last point matters because an adversarial set is not a random sample; it is constructed to find failures, so it supports a different claim. If the pack does not say which dataset supports which sentence, a reviewer cannot tell what the result covers.
Evidence a reviewer asks for
- The dataset register: source, licence, size, date and the dimension it represents.
- The representativeness argument, with the known gaps.
- The acquisition record, including personal data and retention.
- The adversarial set description and the method used to generate it.
- A note on which claim each dataset supports.
Checks to run this month
- Does each dataset name the dimension it represents?
- Is the acquisition lawful and recorded?
- Is the adversarial set described as constructed, not as a sample?
Dates
FDIS registered 2026-08-21. The clause body is not public.
Cross-references
- Write the sampling and perturbation plan
- Adversarial examples and data concentration
- Corruption robustness and its efficient assessment
What this page does not state
The body text of clause 8 and its four subclauses is not public.