The Data Is the Org Chart

The first time I traced a broken forecast to its source, I expected to find a bug. What I found was a project manager who had learned, correctly, that the cost-code field did not matter for his bonus and did matter for his Friday afternoon. He filled it with whatever cleared the validation rule. So did forty of his colleagues. Our model was faithfully learning the shape of their indifference.

You cannot fix that in the pipeline.

---

Data is exhaust from work, and work follows incentives. Wherever a field is entered by someone who never sees it used, that field rots. Wherever a number determines someone’s standing, that number is groomed. Both failure modes are invisible in the warehouse: the rotted field and the groomed field arrive equally well-typed, equally non-null, equally green on the quality dashboard.

This is why “we have a data quality problem” is almost always a category error. The data is fine — it is a precise record of what the organization actually rewards. Read this way, the warehouse is the most honest org chart the company has. It shows you who is measured on what, which handoffs are real and which are theater, and where the official process diverges from the practiced one.

---

The practical consequence: before I model anything now, I ask who enters this field, and what happens to them when they enter it carelessly. If the answer is “nothing,” the remediation is not a cleaning script. It is closing the loop — making the person who enters the data a consumer of the thing built from it. Farmers eat what they grow. Data entered by people who taste the result is the only kind that stays fresh.

Doctrine: garbage in is not a data problem. It is a report on your incentives, delivered nightly.