The Same Number, Twice - The Structural Read
The Structural Read
SubscribeSign in
The Same Number, Twice<br>When a measure stops helping people see and starts teaching them to hide<br>Jun 23, 2026
Share
Organizations measure engineering to get more good engineering. But the most valuable part of engineering is exactly what measurement cannot create, and often what it quietly destroys: judgment, care, the choice to do the harder thing because it is correct.<br>This is not a complaint about bad metrics. It is about what measurement does when it is used to judge.<br>A number can reach a person in two ways. As information, it describes how a system behaves. It tells you something about your own work, and most people want that. As a target you are judged against, it does something else: the work slowly stops being its own reason and becomes work done for the number.<br>Psychologists named this long ago . The overjustification effect: attach a reward or a judgment to something a person already did for its own sake, and the outside reason can replace the inside one. The real danger is not that motivation drops. It is that the behavior now exists only where the number is watching. Away from the measure, it stops.<br>AI Engineers know the same failure from machine learning. Give a model a score to maximize, and it finds the cheapest path to the score, not to the thing the score was meant to represent. Goodhart wrote the general law: when a measure becomes a target, it stops being a good measure. Campbell wrote the harder version for human systems: the more a number decides what happens to people, the more it corrupts the behavior it was meant to track.<br>An organization runs the same failure, only slower and harder to see. Estimates inflate. Activity counts reward motion instead of progress. Incidents get reclassified until they no longer count. None of this needs anyone acting in bad faith. It is what reasonable people do once a number starts to carry weight it was never built to carry.<br>Here is the part the usual framing leaves out. Whether a number informs or corrodes is not a property of the number. It is a property of the structure around it : who owns it, whether the people measured had any say in the target, what happens to them when it reads low, how much trust was there before the number arrived. The same deployment graph is a useful signal on one team and a quiet threat on the next. Nothing in the graph separates the two. The meaning is set by the control structure, not by the metric.<br>This is why you cannot fix a measurement problem by fixing the measurement . The pathology does not live in the chart. Change the metric, and a structure that judges people will load the same weight onto the new one.<br>DORA is a clean test of this . For a decade it sorted teams into four tiers: Elite, High, Medium, and Low, the industry’s standard ladder. In 2025 it removed them and replaced them with profiles that include burnout and friction. The stated reason was AI, not leaderboards, so this is a hint and not a proof. But the change lands exactly where the theory predicts. A four-step ladder is the easiest thing in the world for a team to read as a verdict on itself rather than a description of its work. Built to reveal patterns, it was widely used as a scorecard.<br>Does removing it help? Only partly, and that is worth being honest about. A culture that wants to rank will rebuild a ladder from whatever you hand it: archetypes, burnout scores, anything. What changes is the default. A ready-made ranking with years of authority behind it is the path of least resistance, and taking it away raises the cost of ranking without removing the option. That is a real effect, and a small one. It is also the most any change at the metric layer can do, because the metric was never the cause.<br>So if the number cannot produce the behavior, what can? Only the things that were never controlled by a number in the first place. Craft moves between people. You work beside someone whose judgment you trust, and you absorb how they think. This is apprenticeship, and what researchers call communities of practice: knowledge that travels through relationships, not thresholds. It comes with no clean percentage, and anyone who offers you the exact share of a team you must convert is selling a precision that does not exist.<br>This is also why those channels survive being measured: they never depended on a measure to begin with. The same property that makes craft impossible to install with a metric is what makes it durable once it is real. It does not collapse the moment someone turns it into a target, because it never depended on one.<br>Very little of this is about numbers. You can build the best ruler the field has ever seen, and it stays two different instruments depending on whose hand holds it and what happens to that hand when the number is low. The measure is never the lever. The structure around it is.<br>The one thing worth holding loosely is the view from outside. The...