19.18 Decision Control Error

Decision control error is a failure in the regulatory process through which a decision-making system monitors outcomes, compares them to targets, and adjusts its decisions to close the gap between actual and desired states. It is a systemic error rather than an error in any single decision: it manifests as a persistent inability to achieve or maintain target outcomes despite ongoing decision activity, because the control logic — the rules, procedures, and feedback mechanisms that govern how decisions respond to observed deviations — is defective in some way. Decision control errors represent breakdowns in the cybernetic core of decision systems, where the feedback-driven self-correction that should keep the system on track fails to function as intended.

The Nature of Control Error

In cybernetic terms, a controlled decision system works by comparing observed outcomes to target states, detecting deviations, and generating corrective decisions that drive the system back toward the target. Decision control error occurs when any component of this process malfunctions in a way that prevents effective correction.

Unlike a single poor decision — which may produce a bad outcome that a subsequent decision corrects — decision control error produces a pattern of persistent failure: the system repeatedly makes decisions that miss the target, or makes corrections that overshoot, or fails to detect deviations, or detects them but generates corrections in the wrong direction. The error is in the control logic itself, not in any particular decision that logic produces.

Types of Decision Control Error

Decision control errors can be classified according to which component of the control cycle is defective:

Sensor error occurs when the monitoring component of the decision system fails to detect actual deviations from target. The deviation exists in reality, but the feedback mechanism does not register it — either because the relevant variables are not measured, because the measurement system is insensitive, because data collection is delayed, or because intermediaries suppress or distort the deviation signal before it reaches the decision maker. A system that cannot detect its own errors cannot correct them.

Reference error occurs when the target against which deviations are measured is itself wrong. If the decision system is calibrated to drive outcomes toward a reference state that does not correspond to genuine objectives — because goals are poorly specified, because proxy measures diverge from actual objectives, or because the reference state was appropriate under old conditions but not current ones — it will produce decisions that optimize for the wrong target. Reference error is particularly insidious because it can cause a system to perform well on its measured criteria while performing poorly on the criteria that actually matter.

Gain error refers to miscalibration in the magnitude of the correction applied in response to detected deviations. A system with insufficient gain makes corrections that are too small: it detects deviations but applies corrections insufficient to eliminate them, causing the system to drift persistently away from the target. A system with excessive gain makes corrections that are too large: small deviations trigger large corrections that overshoot the target in the opposite direction, generating oscillation around the target rather than convergence to it. Both forms of gain error produce control failure, one through undercorrection and the other through overcorrection.

Phase error occurs when the timing of corrections is wrong relative to when they are needed. A correction applied with a delay that exceeds the system's response time may be appropriate when issued but counterproductive when it takes effect — because conditions have changed between decision and effect, and the correction is driving toward an outdated target. Phase error is especially common in systems with long decision-to-effect lags, where feedback and correction cycles cannot keep pace with system dynamics.

Sign error is a catastrophic form of control error in which corrections are applied in the wrong direction — positive feedback is generated where negative feedback was needed, or vice versa. Instead of reducing deviations from target, corrections amplify them, producing runaway dynamics that drive the system progressively further from the desired state. Sign errors often result from fundamental misunderstandings of the causal structure of the system being controlled.

Communication Failures as Sources of Control Error

From a communication perspective, decision control errors are often traceable to communication failures in the feedback pathway:

Transmission delay in the feedback channel introduces phase error by ensuring that the correction signal reaches the decision maker later than the deviation occurs. The larger the transmission delay relative to the speed of system dynamics, the more severe the resulting phase error.

Distortion in reporting introduces sensor error by modifying the deviation signal as it passes through intermediaries. Intermediaries may downplay the severity of deviations, filter out information that reflects poorly on their own performance, or summarize in ways that lose the diagnostic detail needed for accurate correction calibration.

Goal displacement in metrics introduces reference error by substituting measurable proxies for the genuine objectives of the decision system. When decision makers are evaluated on metrics that imperfectly represent actual goals, they calibrate corrections to the metric rather than to the goal, producing behavior that optimizes the metric at the expense of genuine objective achievement.

Diagnosing Decision Control Errors

Diagnosing decision control errors requires distinguishing persistent systematic failure from random variation. A system that occasionally misses its target but centers on it over time is performing within normal bounds; a system that consistently misses in the same direction, or that oscillates with increasing amplitude, is exhibiting control error. Statistical analysis of outcome time series — looking for systematic bias, sustained oscillation, or progressive drift — can identify control errors that would be invisible in any single decision review.

Effective diagnosis also requires distinguishing between the different error types, since each has a different remedy. Sensor errors require monitoring system improvement; reference errors require goal clarification; gain errors require correction calibration; phase errors require faster feedback channels or reduced decision-to-effect latency; sign errors require fundamental revision of the causal model underlying the control logic. Applying the wrong remedy can leave the underlying error intact while generating additional confusion about why performance does not improve despite remediation effort.