Controllability indicates whether an external agent
can guide or drive a system's state to a desired condition using
available inputs or interventions.
How would you explain it like I'm…
Can You Steer It?
Imagine a toy car with a remote. If pushing the buttons can drive it anywhere in the room, the car is steerable. If the buttons only make it spin in a circle, it isn't. Controllability is just asking: can the tools you have actually move the thing where you want it to go?
Whether You Can Steer It
Controllability asks whether the moves you are allowed to make are enough to push a system into any state you want. A car with a working steering wheel and gas pedal is controllable: you can get to any spot on the road. A car with a stuck wheel isn't. The same question shows up in medicine (does the medicine actually move the patient toward health?), software (can we actually deploy a fix?), and government (can a policy lever really change the outcome?). If the answer is no, no clever plan will rescue you — you need a different system.
Controllability
Controllability is the structural property that tells you whether the inputs available to an agent can drive a system from any starting state to any desired target state. If yes, the system is controllable; if no, certain states are simply out of reach no matter how clever the strategy. In control engineering, this is checked formally with the rank of the controllability matrix for linear systems. But the idea extends everywhere: a hospital can only heal what its treatments can affect, a policy can only move variables its tools touch, and software can only be fixed if engineers can actually deploy changes. Recognizing uncontrollability is itself useful, because it redirects effort away from impossible goals and toward redesigning the system to add the missing levers.
Controllability is the structural property that determines whether an agent's available inputs can steer a system's state into any desired region. Formally, a linear time-invariant system dx/dt = Ax + Bu is controllable if and only if the controllability matrix [B, AB, A²B, ..., A^(n-1)B] has full rank (Kalman 1960). Nonlinear analogues use Lie bracket algebra and Chow's theorem. Controllability is the information-theoretic dual of observability: observability asks whether outputs reveal state, controllability asks whether inputs steer state. The concept matters because it is the structural precondition for intervention — without it, desired states are unreachable no matter how clever the strategy. Software without deployable fixes cannot be healed; patients without effective treatments cannot be cured; policies without available levers cannot reshape outcomes. The abstraction generalizes across control engineering, SRE and infrastructure (deployability, rollbackability, feature flags), medicine, governance, systems biology (Barabási-Liu-Slotine 2011), and climate science. In each case it sharpens the same question: can intervention actually move the state to the target, or must we first redesign the system to add the missing levers?
Control Systems: A system is controllable if a suitable
input sequence can move it from any initial state to any target
state.
Software & Infrastructure: Admin privileges or APIs let
operators push updates, reroute traffic, or fix bugs
systematically.
Healthcare Interventions: Physicians must apply the "right
inputs" (medications, therapies) to direct a patient's condition
toward recovery.
Policy & Governance: Laws, taxes, or incentives "control"
social or economic variables within certain bounds, though real
systems may be only partially controllable.
If a system is uncontrollable, no effort or
input can shift certain undesirable states—knowing that spares
wasted resources on impossible transformations.
Reveals the parallel to "observability":
full system mastery often requires both seeing internal states
(observability) and having the levers to shift them
(controllability).
Spacecraft thrusters give mission control full
(though finite) range to maneuver a satellite; if thrusters fail or
lack certain axes, some orientations become uncontrollable.
Controllability is not Observability because Observability is the property that internal states can be inferred from outputs, while Controllability is the property that inputs can drive the system to desired states.
Controllability is not Linearity because Linearity is the property that outputs are proportional to inputs, while Controllability is the property that the system state space can be reached from a given initial state through appropriate inputs.
Controllability is not Traceability because Traceability is the ability to track how outputs depend on inputs, while Controllability is the ability to achieve desired states through input choices.
Controllability is not Nonlinearity because Nonlinearity is disproportional input-output relationships, while Controllability is the structural property of state reachability under input choices.
Controllability is not Reflexivity (Self-Reference) because Reflexivity is a relation with itself, while Controllability is the ability to drive system state through external control inputs.