Turning a source like a WPA slave chronicle into a row of data feels, at first, like a loss; Charlie Aarons’s “I don’t know, but I sure been ’round here long time” becomes, in a spreadsheet, just “Not stated.” But that loss is also what makes certain questions askable in the first place. Closely reading one interview alerts you about one life. Arranging five interviews into equivalent fields, as we did in this module’s activity, lets you ask questions no single tale could answer: how many of these five people were enslaved in more than one state? How did stated ages at liberation contrast across interviews? Paul Escott’s much more enormous version of this same move, coding eighty-one variables across 2,358 narratives, is what let him make assertions about patterns across a population rather than the texture of one life (Escott, Slavery Remembered, Appendix A, 1979).

Johanna Drucker makes the fundamental point unswervingly: data are not “given”; they’re “captured” – capta – through acts of collection and organizing, and that structuring is what makes enumeration, cleaning, and comparison possible at all (Drucker, “Data Modeling and Use,” The Digital Humanities Coursebook, 2021). Miriam Posner’s edging of “humanities data” as an essential paradox captures the tension well: humanists need the term to get their evidence taken seriously as infrastructure, even though the signal itself counterattacks the orderliness the word implies (Posner, “Humanities Data: A Necessary Contradiction,” 2015).

What has to stay visible, though, is the process. Ahnert and Ahnert’s Tudor Networks of Power is a good model for this: turning 130,000+ early modern letters into network data took nine months of named, documented labor, a custom “Disambiguation Engine,” and an explicit guiding principle, “it is better not to collapse John Smith and J. Smith unless we could find evidence they were the same person,” because an over-connected network can fabricate historical significance that was never there (Ahnert & Ahnert, Tudor Networks of Power, Oxford University Press, 2023, Introduction).

Hoekstra and Koolen argue this kind of disclosure should be routine rather than exceptional: every choice to select, enrich, connect, and analyze historical data changes it, and those choices are “rarely reported and discussed” even though they shape the findings (Hoekstra & Koolen, “Data Scopes for Digital History Research,” Historical Methods 52, no. 2, 2018). Our own activity bore this out directly: comparing our Aarons data model against the instructor’s example and against Escott’s categories showed how differently two careful readers can structure the same interview, and our cleaning pass showed how much interpretive judgment hides inside something as small as whether a blank cell means “not asked” or “not applicable.” Making that visible- the categories chosen, the judgment calls, what’s missing and why, and the source’s own prior history of handling- is what keeps data-driven humanities arguments honest.