Clause 6, Robustness properties measurable using statistical methods
The public titles name four topics: general, in-domain robustness assessment, specification of expected behaviour, and other considerations that contribute to robustness assessment. The body text is withheld. The titles decide two things a buyer must write down: the domain the claim covers, and the expected behaviour that turns a change into a pass or a fail.
The requirement in plain English
The public titles name four topics: general, in-domain robustness assessment, specification of expected behaviour, and other considerations that contribute to robustness assessment. The body text is withheld. The titles decide two things a buyer must write down: the domain the claim covers, and the expected behaviour that turns a change into a pass or a fail.
What the public record says
The public table of contents lists clause 6 and subclauses 6.1 General, 6.2 In-domain robustness assessment, 6.3 Specification of expected behaviour and 6.4 Other considerations that contribute to robustness assessment. The body text is not public.
| Subclause | Public title | Body |
|---|---|---|
| 6 | Robustness properties measurable using statistical methods | Withheld |
| 6.1 | General | Withheld |
| 6.2 | In-domain robustness assessment | Withheld |
| 6.3 | Specification of expected behaviour | Withheld |
| 6.4 | Other considerations that contribute to robustness assessment | Withheld |
Why it matters to you
In-domain is the word that keeps a robustness claim honest. A result measured on the deployment domain says something different from a result measured on a shifted or adversarial domain, and the two cannot be swapped. The expected-behaviour subclause is the other half: without a written expectation, a change in output is just a change, not a failure. The fourth subclause is a reminder that other considerations contribute, so a statistical result is one input and not the whole assessment.
Evidence a reviewer asks for
- The property register: the property, the domain and the expected behaviour, one row each.
- The definition of the in-domain data and how it differs from the test sample.
- The expected behaviour specification, signed before the run.
- The log of other considerations, with the reason each was included or excluded.
Checks to run this month
- Is the domain of the claim stated in the same sentence as the result?
- Was expected behaviour written before the run?
- Which other consideration could change the conclusion?
Dates
FDIS registered 2026-08-21. The clause body is not public.
Cross-references
- Clause 3.1, Stability
- High-dimensional input and why exhaustive testing fails
- A statistical approach to robustness
What this page does not state
The body text of clause 6 and its four subclauses is not public.