jev-why bundles a sample of a third-party dataset for its runnable example.

deepset/prompt-injections
  Source:  https://huggingface.co/datasets/deepset/prompt-injections
  Licence: the dataset card carries two markings -- "apache-2.0" at the card
           level and "cc-by-4.0" in dataset_info. Both are permissive and both
           require attribution, so this notice cites both rather than choosing.
  Used in: examples/data/prompt_injections_test.jsonl (the 116-row test split,
           unmodified).

The dataset card does not document label semantics, collection method, or
language. examples/injection_screening.py verifies the label convention
empirically before computing any calibration figure, and the README states what
the dataset does and does not document.

Nothing in examples/data/composed_page.md comes from the dataset except one
sentence, quoted verbatim from a label-1 row and identified in the comment at
the top of that file. The comment is stripped before the document is sent to
Jev, so the example never leaks the answer into the input it is asking about.
That document is a constructed illustration, not a dataset sample.
