Skip to content

Discrepancy-Driven Correction

Version
v1 · 2026-08-24 · History
Prime #
801
Origin domain
Systems Thinking & Cybernetics
Subdomain
goal directed control → Systems Thinking & Cybernetics
Aliases
Negative Feedback

Core Idea

A goal-directed loop that holds a target, observes the current state, computes the signed gap, selects a corrective action keyed to that gap, applies it, and re-observes — repeating until the gap falls below threshold. The gap is the load-bearing object: nothing fires without it.

How would you explain it like I'm…

Close the Gap

Imagine pouring water into a cup until it reaches the line you want. You keep looking: too little, pour more; close enough, stop. You watch the gap between where the water is and where you want it, and you keep fixing it until the gap is gone.

Close The Gap Loop

Discrepancy-Driven Correction is a loop that fixes a gap. First you decide on a target, like the temperature you want a room to be. Then you check where things actually are, and find the gap between the goal and now. You take an action to shrink that gap, then check again, and keep repeating until the gap is small enough. The whole loop only runs because there is a gap, and everything it does is to make the gap smaller.

Gap-Driven Correction Loop

Discrepancy-Driven Correction is a goal-directed loop built around one object: the signed gap between a target and the current observed state. The system holds a target, observes its state, computes the gap, chooses an action as a function of that gap, applies it, and re-observes — repeating until the gap is small enough or the target gets reconsidered. It is sharper than feedback in general: a runaway loop that amplifies noise is feedback too, but it has no target, so it isn't this. It is also broader than checking something once, because the commitment here is to the loop — act on the gap, then re-measure. A characteristic way it fails is target drift, where the target quietly slides to match what you observed instead of the other way around.

 

Discrepancy-Driven Correction is the iterated, goal-directed loop in which a system holds a target, observes its current state, computes the signed gap, selects a corrective action as a function of that gap, applies it, and re-observes, repeating until the gap falls below an acceptance threshold or the target is reconsidered. The gap is the central object; the loop exists only to reduce it. The target may be a setpoint, specification, predicted value, plan, or ideal; the observation a sensor reading, test result, or judgment; the action a control signal, code change, or hypothesis revision. What unifies these is shape, not content. It is the precise intersection of three commitments no single neighbor names alone: goal-directed (a target exists), iterated (a loop with re-measurement), and gap-driven (action is a function of the signed discrepancy). Six primitives recognize it anywhere: target, observation, gap function, action selector, action application, termination condition. Its characteristic failure mode is *target drift*: silently moving the target to match the observation instead of the reverse.

Broad Use

  • Control theory: setpoint versus plant output; the error signal drives the controller's actuation.
  • Biology: homeostatic regulation of temperature, glucose, or blood pressure toward a setpoint.
  • Software: the edit–compile–test cycle, where the gap is the bug symptom and the action is a code change.
  • Medicine: differential diagnosis, where the gap is the unexplained finding and the action is hypothesis revision.
  • Science: anomaly investigation, where the gap is the residual between model and data.
  • Project management: variance analysis, where the gap drives scope, resource, or schedule corrections.

Clarity

It separates action without a target from genuine correction, one-shot validation from iterated correction, and exposes target drift — closing the gap by moving the target rather than the observation.

Manages Complexity

It reduces a sprawl of "iterative work" practices to one checklist: target, observation, gap, action, threshold, cycle time — with loop gain and cycle-time matching as portable levers.

Abstract Reasoning

Match action gain to gap magnitude, match cycle time to how fast the gap evolves, audit the comparator, and set the termination criterion deliberately rather than by fatigue.

Knowledge Transfer

  • Engineering ↔ medicine: a debugger and a clinician working a fever of unknown origin run the same loop and fail the same ways — gap miscomputed, action too aggressive, target drifted.
  • Control ↔ coaching: matching action gain to gap magnitude is one insight in PID tuning and in a coach's correction cadence.
  • Control ↔ neuroscience: predictive processing casts perception itself as prediction-error minimisation, the same loop run continuously.

Example

A PI controller stabilising a reactor compares temperature to a 350 K setpoint and opens a coolant valve in proportion to the signed error; too much gain and the loop oscillates, too slow a sample and it cannot converge.

Relationships to Other Abstractions

Local relationship map for Discrepancy-Driven CorrectionParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Discrepancy-DrivenCorrectionPRIMEPrime abstraction: Feedback — is a kind ofFeedbackPRIMEDomain-specific abstraction: Missile Guidance — is part ofMissile GuidanceDOMAINDomain-specific abstraction: Guide Star — is a kind ofGuide StarDOMAINDomain-specific abstraction: Rectification (law) — is a kind ofRectification(law)DOMAINDomain-specific abstraction: Skills Management — is a kind ofSkillsManagementDOMAINPrime abstraction: Calibration — is a kind of, typicalCalibrationPRIMEPrime abstraction: Homeostasis — is a kind of, typicalHomeostasisPRIME

Current abstraction Discrepancy-Driven Correction Prime

Parents (1) — more general patterns this builds on

  • Discrepancy-Driven Correction is a kind of Feedback Prime

    'feedback's goal-directed specialisation — feedback structured around an explicit target and a signed gap that the action is keyed to.' A strict subspecies of feedback (which has no target).

Children (6) — more specific cases that build on this

  • Guide Star Domain-specific is a kind of Discrepancy-Driven Correction

    The minimal proposed parent is prime:discrepancy_driven_correction.

  • Rectification (law) Domain-specific is a kind of Discrepancy-Driven Correction

    The proposed strict upward parent is prime:discrepancy_driven_correction.

  • Skills Management Domain-specific is a kind of Discrepancy-Driven Correction

    Discrepancy-Driven Correction is the strongest live parent.

  • Calibration Prime is a kind of, typical Discrepancy-Driven Correction

    Calibration is typically a specialization of Discrepancy-Driven Correction, retaining the parent's defining structure while adding the child's specific commitments.

  • Homeostasis Prime is a kind of, typical Discrepancy-Driven Correction

    'homeostasis is one biological OUTCOME of a stabilising correction loop...

Hierarchy path (1) — routes to 1 parentless root

  • Discrepancy-Driven Correction → Feedback

Not to Be Confused With

  • Discrepancy-Driven Correction is not Feedback because correction adds a maintained target and a gap-keyed action, whereas feedback names the bare fact that an output is routed back to influence an input.
  • Discrepancy-Driven Correction is not Homeostasis because correction is the general loop mechanism that can chase a moving target or revise a hypothesis, whereas homeostasis is one biological outcome of a stabilising loop.
  • Discrepancy-Driven Correction is not Refinement because correction is gap-keyed and terminates on a threshold, whereas refinement is open-ended improvement toward "better" with no signed gap.