Question 1: What Is Actually Being Controlled Here?
The short version
Every purposeful system (a person, a team, an organisation, a government) is constantly acting to keep something stable. That something is called the controlled variable: the thing the system’s behaviour is organised to hold constant, which routinely diverges from what the system says it is trying to achieve.
When you are trying to change a system and failing, the most common reason is that you are targeting the stated goal rather than the operational goal. You are pushing on a variable the system is not actually controlling, so the system routes around your intervention without effort, and nothing changes.
The first diagnostic question is: what is this system’s behaviour actually organised to maintain?
Why this question is non-obvious
We are accustomed to taking stated purposes at face value. An organisation says it exists to serve customers. A policy says it is designed to reduce poverty. A school says its goal is to develop capable, curious people. A person says they want to get healthy.
Perceptual Control Theory (developed by the engineer and physicist William T. Powers) shows that stated purpose and operational purpose systematically diverge, and that behaviour is always organised around the operational purpose, not the stated one.
Powers demonstrated this with what he called the rubber band experiment. Two people hold opposite ends of a rubber band. One person’s task is to keep the knot at a fixed position. The second person tries to predict the first person’s hand movements by watching their own movements.
It is impossible. The first person is controlling a perception (the position of the knot), not producing a particular behaviour. Their hand moves in whatever way is necessary to keep the knot still, which depends entirely on what the second person does. The hand movements are unpredictable. The knot position is stable.
The critical insight: you can only predict a system’s behaviour by identifying what it is controlling. The same stimulus produces different responses. The same response can come from different stimuli. What is consistent is the controlled variable: the thing the system is acting to hold stable.
This inverts the common approach to understanding (and changing) behaviour. Most analyses ask: what is this agent doing? The more useful question is: what perception is this agent’s behaviour organised to maintain?
How systems end up controlling the wrong thing
Operational goals drift from stated goals through a consistent mechanism: feedback loops attach to what is measurable, not what matters.
A system acts on what it can perceive. If the thing that matters (genuine customer satisfaction, actual learning, real poverty reduction) is hard to perceive in real time, the system attaches its feedback loop to something easier to measure. Revenue. Test scores. Benefit claimant numbers. The feedback loop then maintains that proxy variable, not the thing it was meant to represent.
Over time, the proxy variable becomes the operational goal. The system is now genuinely, functionally trying to hold the proxy stable: this is the normal operation of its control architecture, not a deviation from it. The staff and managers are doing exactly what the system rewards. The system is maintaining the variables its feedback loops are attached to.
This is why measuring the wrong thing does not just give you bad data. It changes what the system is trying to do. It changes the operational goal.
Beer put the information version of this observation simply: “Information is what changes us. Up until the point it changes something, it is merely data.” The inverse is equally true: what a system acts on (what actually changes its behaviour) is its real information. Everything else is noise to that system, regardless of how important it seems to an outside observer.
Applying the question
To identify what a system is actually controlling, do not ask what it says it is for. Observe its behaviour under stress.
When a conflict arises between two apparent goals, which one does the system sacrifice?
A school that says it values creativity and curiosity but responds to budget cuts by eliminating arts and reducing unstructured time is telling you something. It is controlling its cost structure and its measurable academic outputs. Creativity and curiosity are values in its stated identity, not variables in its control loop.
A company that says it puts customers first but, when revenue dips, immediately cuts customer service headcount is telling you something. The operational controlled variable is the revenue line. Customer experience is a stated value that floats free of the feedback loop.
A government that says it is reducing poverty but measures success through benefit claimant numbers (then tightens eligibility criteria when the numbers rise) is telling you something. It is controlling the perception of poverty (the count of people officially in the system) rather than the conditions that produce poverty.
In each case, the organisation is not lying when it states its goals. It genuinely holds those goals at some level of its identity. But the operational control loop (the mechanism that actually drives behaviour) is attached to something else.
Three diagnostic moves:
- Find the feedback loop. What gets measured, reported, and acted on in real time? That is what the system is controlling.
- Apply stress. When the system faces a genuine trade-off between two objectives, which one drives behaviour? The winner is the operational controlled variable.
- Observe what triggers action. What has to change before anyone acts? Systems respond quickly to changes in their controlled variables and slowly (or not at all) to changes in everything else.
The hierarchy problem: when the goal of the goal is different from the goal
PCT identifies a further complication. Goals are not flat: they are organised in a hierarchy. Lower-level goals serve higher-level goals, which serve still higher-level goals, up to the level of identity and values.
This matters because an intervention aimed at a lower-level goal will fail if it conflicts with a higher-level goal that the lower-level goal was serving.
A person who says they want to lose weight may also, at a higher level of their control hierarchy, be controlling their sense of social belonging (food as connection, eating as social participation) or their emotional regulation (eating as anxiety management). An intervention that addresses only the lower-level goal (diet plans, calorie counting) will fail or produce rebound, because the behaviours it is trying to change are serving higher-level goals the intervention has not touched.
An organisation that says it wants to innovate may, at a higher level, be controlling its stability and its current power structures. Innovation that threatens those structures will be reliably routed around, regardless of how many innovation programmes are launched, because the operational System 5 (the identity function) is organised around stability, not change.
The diagnostic implication: when an intervention consistently fails despite genuine effort, look one or two levels up in the goal hierarchy. The behaviour you are trying to change is probably serving a higher-level goal that your intervention is threatening. You will not change the behaviour until you either address that higher-level goal directly, or find a way to serve it that does not require the behaviour you are trying to eliminate.
What psychological distress looks like through this lens
PCT offers a specific account of where distress comes from: conflict between control levels.
When a person controls “be honest” at one level and “be kind” at another, and these come into conflict in a specific situation, neither goal can be achieved without violating the other. The conflict produces oscillation at lower levels (the person cannot settle on a behaviour) which manifests as anxiety, rumination, and apparently irrational indecision.
Standard interventions address the symptoms: the anxiety, the indecision, the specific behaviours. They do not address the conflict at the level it actually exists. The Method of Levels therapy, derived directly from PCT, works by helping the person become aware of and resolve the higher-level conflict, not by managing its lower-level manifestations.
This same structure appears in organisations. An organisation that is told to “innovate” while also being held to tight quarterly targets is experiencing a genuine conflict between a higher-level goal (stability, predictability, short-term performance) and a lower-level mandate (novelty, risk, deviation from the current path). The organisation will oscillate. It will launch innovation programmes, then defund them when revenues wobble, then relaunch them, then defund them again. This is not poor execution. It is the mechanical output of a genuine control conflict that has not been resolved at the level where it exists.
The failure mode this question reveals
The most common failure mode this question uncovers is what might be called proxy capture: the feedback loop becomes attached to a proxy variable, the proxy variable becomes the operational goal, and interventions aimed at the underlying reality are reliably neutralised because the system is not, in fact, controlling the underlying reality.
Some domain examples:
In healthcare: A hospital system says it is controlling patient outcomes. Its feedback loops are attached to waiting times, throughput, and budget adherence. When there is pressure, it reduces beds to hit budget targets, which worsens throughput, which distorts waiting time figures. The operational controlled variable is institutional survival and compliance with regulatory metrics, not patient outcomes, which are too slow and diffuse to function as real-time feedback signals.
In education: A school system says it is developing capable, curious people. Its operational feedback loops are attached to standardised test scores, which are visible, comparable, and tied to institutional funding. Teachers teach to the test because the test score is what the system’s feedback loop is attached to. Student curiosity is outside the control loop. It deteriorates unmeasured.
In public policy: A poverty reduction programme says it is controlling poverty rates. Its operational feedback loop is attached to programme participation numbers and budget spend. A programme that reduces poverty while spending less would trigger alarm signals (budget underspend = poor performance). A programme that grows participation while poverty increases would trigger positive signals (meeting targets). The system is controlling its own operational metrics, not poverty.
In organisations: A technology company says it is controlling user wellbeing. Its operational feedback loops are attached to engagement metrics: time on platform, clicks, shares. Engagement and wellbeing are positively correlated in some circumstances and negatively correlated in others. When they diverge, the system maintains engagement. User wellbeing is not in the control loop.
In every case: the intervention you need is not aimed at the stated goal. It is aimed at the feedback loop itself: specifically, at what the feedback loop is attached to.
What the answer tells you
Once you have identified what the system is actually controlling, you have three possible intervention points:
1. Change what the feedback loop is attached to. Replace the proxy with a more direct signal. This is harder than it sounds: good leading indicators of the thing that actually matters are rare and expensive to generate. But it is the only intervention that changes what the system is actually controlling.
Example: A school that replaces end-of-year exam scores with weekly formative assessments and measures of student engagement is not just adding a new metric. It is changing the controlled variable. The system will start to optimise for the new signal.
2. Resolve the conflict in the goal hierarchy. When lower-level behaviour is serving a higher-level goal that your intervention threatens, you must either address the higher-level goal directly or redesign the intervention so it serves rather than threatens it.
Example: An innovation mandate will fail in an organisation where stability is the higher-level operational goal until and unless the leadership explicitly resolves the conflict: by making innovation a component of stability rather than a threat to it, or by changing System 5 (what the organisation believes itself to be) rather than issuing lower-level mandates.
3. Accept the actual controlled variable and work with it. Sometimes the controlled variable is legitimate and the problem is that the stated goal was wrong, not that the system is. A hospital that is operationally controlling institutional survival under severe resource constraint may be doing the only viable thing. The intervention is not to change what it is controlling, but to change the resource constraint that has forced it into that mode.
Before moving to Question 2
Question 1 identifies the gap between what the system says it is doing and what it is actually doing. This is often enough to reframe the intervention entirely.
But it raises a further question: even if you correctly identify what the system should be controlling, can it? Does the system have the capacity to regulate a variable as complex as genuine patient outcomes, or real student development, or actual poverty levels?
That is the question of variety, whether the controller has enough complexity to match the complexity of what it is trying to control. It is the subject of Question 2.
Reference
The theoretical foundation for this question is Perceptual Control Theory, developed by William T. Powers.
Primary source: Behavior: The Control of Perception (1973, second edition 2005).
Key extension to therapy: Method of Levels, developed by Tim Carey, based directly on PCT’s account of control hierarchy conflict.
The PCT framework was developed independently of but is deeply consistent with Stafford Beer’s Viable System Model and Karl Friston’s Free Energy Principle. All three converge on the same core insight: purposive systems act to maintain perceptions, not to produce behaviours, and understanding what perceptions a system is maintaining is the only reliable basis for predicting or changing its behaviour.