Dignity in the Columns: The First Value of the DoGoDa Oath

Rotterdam’s welfare-fraud model never met the woman it scored. It only knew she was a parent, a woman and someone whose Dutch was poor, and each of those raised her fraud-risk score. When journalists from Lighthouse Reports took the city’s model apart in 2023, that variable list was the whole picture: roughly 30,000 welfare recipients in Rotterdam were in scope, and the score influenced which of them got investigated. She was not a fraud risk waiting to be discovered. She was a person who had the misfortune of matching several columns.

The columns decide

The way I see it, the moment a data team ships a customer score, a segment or a queue position, it has taken a decision about human dignity. That decision has a precise address: the columns you accepted as a description of a person. She has no idea your table exists and she never chose those columns, yet they settle how your organisation treats her: which offer she gets, how long she waits, whether someone doubts her.

Being a parent, being a woman, speaking poor Dutch: someone at a desk accepted those as stand-ins for fraud risk. The choice felt defensible at the time, or it was made too quickly to be examined, or it was buried in a join on a lookup table that everyone had forgotten was there. But the person in the row experienced the result.

Same choice, different desk

The Rotterdam case is one data point. In 2019 researchers took apart a care algorithm applied to millions of US patients. The algorithm used money spent on a patient as a stand-in for the care that patient needed, which sounds plausible until you look at the distribution: less money is spent on Black patients with the same health conditions, so they had to be considerably sicker to receive the same risk score. Correct that one design choice and the share of Black patients flagged for extra care rises from 17.7% to 46.5%. One column, accepted without a signed rationale, skewed the care trajectory of millions of people.

In 2020 the Dutch privacy regulator found that the Belastingdienst had let nationality stand in for fraud risk, in breach of the GDPR. Different organisation, different sector, different country, different decade: the same move. A column accepted as a truth about a person, by someone at a desk who did not write down why.

Models need columns, segmentation is ordinary work, and I build scoring models myself. All true. But every stand-in above was typed by a person with access to a keyboard and a dataset, and most organisations can name who approved the dashboard while almost none can name who approved money spent as a stand-in for care needed.

Nobody signed it

That accountability gap is not primarily a legal problem, though the GDPR makes it one when the processing meets the Article 22 threshold for solely automated decisions with significant effects. The gap shows up before the legal tests kick in: at the desk where someone first accepts a column as a description of a human being.

The columns-decide argument is easy to accept as an abstraction; it is harder to sit with when you are the one who chose the column. That discomfort is productive: it is where dignity as a working value starts, not in a policy document but in the moment of deciding which stand-in ships.

The neighbor rule

So this is my rule, and it is where the DoGoDa Oath starts, with its first value, dignity: build every scoring or profile model as if the person in the row is your neighbor. The test is practical. In your next model review: write each stand-in as one sentence about a person (“we treat money spent as a stand-in for care needed”), put your name next to it, and ask whether you would defend that sentence to your neighbor’s face. A stand-in you would not defend does not ship.

The neighbor rule does not require a committee, a new process or a compliance tool. It requires the person at the desk to pause before accepting a column and to write down what that column is actually standing in for. That written sentence is what dignity means in daily data work: not a poster value but a habit, practised at the level of the schema review.

The DoGoDa Oath walks this value first because it is the foundation the others rest on: a data practice that is not dignity-first cannot become transparent about it later. The next Oath value builds from this one.

Happy to think this through with you.