Why it matters
Providers of high-risk systems must draw up technical documentation before the system is placed on the market or put into service, keep it up to date, and include at least the elements in Annex IV (Article 11). Small and medium-sized companies, including start-ups, and now also small mid-caps, may provide them in a simplified form using a form the Commission is to provide (not yet confirmed as published).
The data point, word for word
Annex IV point 2(d) asks for, where relevant: “the data requirements in terms of datasheets describing the training methodologies and techniques and the training data sets used, including a general description of these data sets, information about their provenance, scope and main characteristics; how the data was obtained and selected; labelling procedures (e.g. for supervised learning), data cleaning methodologies (e.g. outliers detection)”. See Annex IV.
What to write
- General description of each data set and its main characteristics.
- Provenance and scope: where it came from and what it covers.
- How it was obtained and selected.
- Labelling procedures and guidelines.
- Cleaning methods: which checks ran, what they flagged and what you decided.
Related points
Annex IV also asks for development methods including pre-trained systems or third-party tools (point 2(a)), design choices and assumptions (2(b)), validation and test procedures with information about the test data, and dated and signed test logs (2(g)), and specifications on input data (point 3).
A datasheet makes this easier
If you keep a datasheet per data set while you build it, this section is mostly copy and update. See how to document a dataset. This guide is general information, not legal advice.