Non-Stationary Objective

Prime #

1025

Origin domain

Control Theory Cybernetics

Subdomain

tracking problems → Control Theory Cybernetics

Core Idea

A non-stationary objective is the condition in which the target a system is converging on fails to hold still on the system's convergence time-scale. The load-bearing variable is the ratio of drift rate to adaptation rate: when it crosses one the system becomes tracking-bound rather than converging-bound, and the figure of merit shifts from steady-state error to tracking error under drift.

How would you explain it like I'm…

Chasing the Ice-Cream Truck

Imagine trying to catch a friend who keeps running away while you chase them. If they run as fast as you, you can never quite reach them — you just keep chasing. It's not about how fast you are alone; it's about whether you're faster than they're running. When the thing you're trying to reach keeps moving, catching up turns into keeping up.

The Goal That Keeps Moving

Usually when you're trying to get something right, you assume the goal stays still while you work toward it — like aiming at a target on a wall. A non-stationary objective is when the target itself moves while you're aiming, and it moves about as fast as you can adjust. When that happens, everything changes: your mistakes never fully go away, and a plan you made is already out of date by the time you finish it. The thing that decides whether you can cope isn't your speed by itself or the target's speed by itself — it's the *ratio*: are you adjusting faster than the target drifts? Once the target moves faster than you can keep up, the question stops being 'did I reach it?' and becomes 'how close can I stay?'

Tracking, Not Converging

A non-stationary objective is the condition where the target a system is converging on — its set-point, goal, reward function, or specification — does not hold still on the system's own convergence time-scale. Control, learning, and planning normally assume the objective stays fixed while the system homes in; a non-stationary objective breaks that assumption by moving as fast as, or faster than, the system can converge. The behavior then changes qualitatively: errors stop decaying, learned policies are stale the moment they're ready, and plans expire before they execute. The right figure of merit shifts from *steady-state error* (asymptotic distance to a fixed target) to *tracking error under drift* (running distance to a moving target). Crucially, the load-bearing quantity is not the drift rate or the adaptation rate alone but the *ratio* of the two: a slow tracker is fine against a static goal, and a fast-moving goal is fine if the tracker is faster still.

 

A non-stationary objective is the structural condition in which the objective a system is converging on — its set-point, target, fitness landscape, reward function, or specification — fails to hold still on the system's convergence time-scale. Control, learning, planning, and adaptation systems normally presuppose a stable objective while they home in on it; a non-stationary objective violates that presupposition, because the target moves at a rate comparable to or faster than the system can converge. When this happens the behavior changes qualitatively: errors no longer decay, learned policies become stale on arrival, plans expire before execution, and the relevant figure of merit shifts from *steady-state error* — asymptotic distance to a fixed target — to *tracking error under drift* — running distance to a moving target. The essential commitment is that the load-bearing variable is neither the drift rate nor the adaptation rate in isolation, but the *ratio* between them: a slow tracker is fine under a static objective, and a fast objective is fine if the tracker is faster still. The arrangement has a small recurring cast — an adapter (controller, learner, planner, agent) trying to reach an objective; an objective set by a reference, target, distribution, or landscape; the adapter's natural convergence rate; the objective's drift rate; and the ratio of the two, which fixes the regime. When the ratio crosses one, a qualitative shift occurs: the system becomes tracking-bound rather than converging-bound, and the whole vocabulary of 'did it converge?' gives way to 'how well does it track?'

Broad Use

• Control theory: the distinction between regulator problems (constant set-point) and tracking problems (time-varying set-point).
• Machine learning: concept drift, covariate shift, and reward-function non-stationarity in reinforcement learning.
• Evolutionary biology: Red Queen dynamics, where coevolving species each form a moving fitness target for the other.
• Software engineering: requirements volatility and the agile reframing of delivery as a tracking-rate increase.
• Military strategy: Boyd's OODA loop — victory accrues to the side whose decision loop cycles faster than the situation changes.
• Policy: rule-making chasing innovation, monetary policy under shifting natural rates, the Lucas critique.

Clarity

It relocates the diagnosis from "the system is bad" to "the objective moved," and makes the load-bearing quantity a ratio rather than either rate alone.

Manages Complexity

It collapses model staleness, policy obsolescence, plan expiration, and regulatory chase into a single comparator — drift rate versus adaptation rate — organizing the entire intervention space into three families.

Abstract Reasoning

A qualitative regime shift occurs as the ratio crosses one, and reporting steady-state error for a tracking-bound system is a category error — the system was never in steady state.

Knowledge Transfer

• ML ↔ biology: concept drift and Red Queen coevolution are the same drift/adaptation ratio, with retraining cadence and generation time as convergence rates.
• Control → software: a control engineer recognizes agile delivery as tracking-rate increase.
• Control → strategy: a strategist recognizes OODA-loop superiority as a pure ratio claim.

Example

A deployed model's accuracy degrades in production not because it is bad but because the data distribution drifts faster than the retraining cadence can track — an unfavourable ratio — answered by speeding the tracker (online learning), slowing the target (stabilising the input), or accepting tracking error (continual learning that reports tracking accuracy).

Relationships to Other Primes

Foundational — no parent edges in the catalog.

Children (1) — more specific cases that build on this

• Concept Drift is a kind of Non-Stationary Objective — The file: concept_drift is 'one substrate-specific case' — the ML instance (a data distribution shifting under a deployed model) of the substrate-neutral non-stationary-objective structure. concept_drift is a CANDIDATE (CAND-R2-022-08), not yet canonical, so this is a candidate-link parent rather than a corpus reparent.

Not to Be Confused With

• Non-stationary objective is not concept drift because concept drift is the machine-learning instance, whereas this is the substrate-neutral structure of which drift, Red Queen coevolution, and OODA dynamics are all cases.
• Non-stationary objective is not regime change because regime change is a discrete shift to a new regime, whereas this is continuous drift governed by a ratio.
• Non-stationary objective is not feedback because feedback is a mechanism that drives error toward a reference, whereas this is a condition on the reference itself moving faster than the loop can converge.