[deferred to copy pass]
A system can be locally correct in every output it produces and globally wrong in what those outputs add up to. This is not a contradiction. It is the predictable consequence of optimizing many decisions for local defensibility without an interpretive frame for what the decisions, taken together, are supposed to mean. The drift is structural. It is also, from inside the system, almost invisible.
The phenomenon is straightforward in description and disorienting in experience. Each output passes its review. Each metric improves or holds. Each function reports the numbers it was built to report, and the numbers are real. And the organization, taken as a whole, ends up somewhere no one would have chosen if anyone had been asked where the organization was going.
This is the shape of drift under local correctness. Nothing went wrong. Everything was working. The trajectory diverged anyway.
The metrics that confirm local correctness are precisely the metrics each function was built to produce. This is not a flaw in measurement. It is what measurement is. A function measures what it can measure, and what it can measure is its own local output. The supply chain reports cost reductions. The service function reports resolution rates. The marketing function reports engagement. Each report is accurate. Each is also bounded by the function that produced it.
What the reports cannot capture is whether the functions, taken together, are still doing what the organization needed them to do. That question is not local. It cannot be answered by aggregating local metrics, because aggregation is arithmetic, and the question is interpretive. It asks what the activity means in context: whether the cost reductions are reducing costs that mattered, whether the resolutions are resolving the right things, whether the engagement is engaging the people the organization is for.
These are not bad-faith questions. They are the questions an organization was answering, implicitly, before the systems began producing outputs at a scale that exceeded any individual's capacity to interpret. The interpretation used to happen in conversation, in judgment, in the small frictions of human review. When the volume increased and the reviews thinned and the friction came to be seen as a cost to be reduced, the interpretive layer thinned with it. The metrics remained. The interpretation that the metrics used to be checked against did not.
People inside the system are not stupid. They notice that something is off. The drift produces a low-grade dissonance: outputs that pass review but don't quite cohere, decisions that defended themselves at every step and still seem wrong in retrospect, the feeling that the work is fine and also not. The dissonance is real. It has no obvious cause. So it gets attributed.
The attributions are predictable and almost always plausible. It is a change management problem: people have not adapted to the new tools. It is a data quality problem: the inputs are noisier than they should be. It is a training problem: the systems need more examples. It is an alignment problem: the functions are pulling in different directions. It is a maturity problem: the deployment is early; the wrinkles will smooth out.
Each attribution is sensible. Each one is also a displacement. The cause is located somewhere that admits a remedy, because a cause without a remedy is intolerable, and a remedy without a cause is incoherent. The attribution closes the loop and lets the work continue. What it cannot do is name the structural condition the attribution is displacing. The condition is that the system is producing locally correct outputs in the absence of any layer that determines what local correctness, in this situation, is supposed to mean.
The displacement is what allows the drift to continue. The dissonance is acknowledged and reassigned. The work proceeds. The drift compounds.
The drift is hard to see from inside because the system was never structured to produce a view of it. There is no dashboard for global coherence. There cannot be, in the sense that coherence is not a quantity. It is a relationship, between what the system produces and what the situation requires, and the relationship is only legible from a vantage that no role inside the system occupies.
The function lead sees the function's metrics. They are good. The executive sees the executive's metrics. They are also good, in their own register. The board sees what the executive reports. The customer sees what the customer encounters. Each vantage is partial. Each is locally correct. None of them sees what the vantages, taken together, are producing, because no one is positioned to occupy a vantage that aggregates them in any way other than additively. And the aggregation that would matter is not additive. It is interpretive.
This is why the drift does not produce a moment. There is no point at which someone says: here is the failure. There is no failure, in the conventional sense. There is a slow divergence between what the system produces and what the organization is for, and the divergence does not announce itself, because it is not visible from any of the seats from which announcements are made. By the time the divergence becomes large enough to be visible at the executive level, it has compounded across so many decisions that no specific decision can be held responsible. The drift has no author. That is part of its durability.
The felt experience has a specific shape, and people who have been inside it tend to recognize it when it is named. The shape is this: things are working and also not. The metrics are green and the situation isn't right. The work defended itself at every step and arrived somewhere no one would have chosen. The conversations about what to do next are productive and also strangely unmoored. The participants agree on the next move and feel, privately, that the next move is not going to be the move that resolves what they cannot quite say is wrong.
There is no fault line. There is nothing to point at. The dissonance cannot be filed against anything. It cannot be escalated, because escalation requires an incident, and there has been no incident. It cannot be addressed in a retrospective, because retrospectives presume a failure to look back at, and the system did not fail. It produced. The outputs arrived. The metrics held.
What is being experienced, in that condition, is what it is like to be inside a system that is locally correct and globally wrong. The wrongness is not in any of the parts. It is in the relationship between the parts and the situation that no part is structured to interpret. The system is working exactly as it was built. The drift is what working looks like when no one was watching what working was for.
That is the shape of the feeling. It is worth being able to name it, because what cannot be named tends to be reassigned, and what is reassigned tends to compound.