Keep Answers Relevant

For content to be useful, both the question asked and the answer provided need to be relevant.

Sometimes, a question may be clearly relevant to your dataset, but the fidelity and framing of the answer can easily make it irrelevant to the reader. For a question to be relevant to a dataset, it must be applicable across the data lifecycle, to different parts of the dataset (data points, subsets, or derivatives), and important for the readers you have in mind. The same criteria apply to answers. In addition, the level of detail provided in an answer affects its relevance and importance.

Ask, "How do readers experience dataset complexity in content?"

While every reader benefits from clarity and simplicity, too much simplicity can introduce vagueness. Inversely, too much information can become overbearing. Consider your readers’ expertise, proficiency, and tasks to determine when and how to add complexity when answering questions.

The Overview section of the More Inclusive Annotated Images Data Card (top) sets the tone for what readers can expect. It recognizes researchers and practitioners as their primary audience, which reflects the level of complexity present throughout their Data Card. In contrast, some readers might find the Open Images Extended Crowdsource Data Card (bottom) inefficient for their needs because reader expectation was not defined upfront.
In the WikiDialog-OQ Data Card, complexity at different abstractions is layered across the content. The "Dataset Subject" provides an overview that sets the tone what what readers can expect. The "Example: Data Point" directly speaks to the more technically inclined reader who may, whereas the "Data Fields" addresses the needs of readers who want to understand what is meant by each feature.
In addition to a code-based example and a description of each field in the datapoint, the WikiDialog-OQ Data Card also presents a graphic for non-technical reader groups.

Don't disregard a question in the Data Card that appears irrelevant too quickly.

Sometimes questions that are hard to answer can appear irrelevant. These present magic moments to proactively anticipate new ways in which your dataset might be used, second- and third-order outcomes of dataset usage, and institute mitigative measures where necessary.

Pay special attention to near-relevance, and consider what changes to context can make these relevant. For example, could a new use case or the addition of new features to the dataset change the relevance of certain Data Card questions?

The More Inclusive Annotated Images Data Card was released with the first version of the dataset, so dataset owners are able to anticipate that there may be other annotations used in conjunction with their dataset. Their Data Card helps readers understand future contexts in which more uses might become available.