Step-Size Throttle¶
Adaptive control — instantiates Bounded Random-Walk Navigation
Governs how large each increment may be — widening steps to explore, shrinking them to settle — on a schedule tied to progress.
A Step-Size Throttle is a governor on the magnitude of each random step. It sets and adjusts how big an increment the walk may take at each point — typically starting large for coverage and shrinking as the process nears a target or as evidence firms up. Its defining move is that it scales steps in place: it neither moves the boundary nor relaunches the walk; it turns the single knob of step size, trading exploration breadth against settling precision. Every other parameter of the walk can be held fixed while this one dial answers the recurring question "how big a step should we take right now?"
Example¶
A clinician is titrating a patient's basal insulin toward a target fasting glucose. Each dose change is a step, and the glucose response is genuinely noisy — diet, stress, and measurement all jostle the reading. Early on, with the patient far above target, the protocol allows large dose steps (say ±4 units) to close the gap quickly. As the fasting glucose approaches target, the step-size schedule shrinks the increments (down to ±1 unit) so the walk settles onto target instead of overshooting into dangerous hypoglycemia. The schedule is tied to distance-from-target, not to the calendar: big steps while far, small steps while close. The result is fast convergence without the overshoot that a fixed large step, or the crawl that a fixed small step, would each produce.
How it works¶
- Set the increment scale. Fix the parameter that controls how large a random step may be.
- Schedule its decay. Shrink step size as a function of progress, time, or firming evidence — large for early exploration, small for late precision.
- Adapt to behavior. Widen again if the walk stalls short of target; shrink faster if it oscillates or overshoots.
It touches only the step-magnitude dial; direction, boundaries, and position are somebody else's job.
Tuning parameters¶
- Initial step size — how bold the early exploration is; too large overshoots from the start, too small crawls.
- Decay schedule — linear, geometric, or performance-adaptive; aggressive decay settles fast but risks freezing early.
- Floor — the minimum step the walk is ever allowed, so it never fully stops adjusting.
- Adapt-up trigger — the condition (a detected stall) under which step size is widened again.
- Oscillation damping — how sharply to shrink when the walk overshoots and reverses.
When it helps, and when it misleads¶
Its strength is that it resolves the coverage-versus-precision tension along a single clean axis — step size — so a process can explore boldly when far from where it needs to be and settle finely when close. This is the same logic as a learning-rate schedule, and its convergence conditions are the classic ones from stochastic approximation[1].
Its failure mode is a mistuned schedule. Shrink too fast and the walk freezes before it reaches the target — premature convergence onto wherever it happened to be. Shrink too slow and it oscillates forever around the target, never settling. The classic misuse is a fixed schedule blind to actual progress: shrinking step size on a timer while the walk is still far from where it should be, so it loses its reach exactly when it still needs it. The guarding discipline is to tie the schedule to observed progress rather than to the clock — decay only as the walk actually closes on its target.
How it implements the components¶
step_size_schedule— the schedule by which step magnitude changes over the course of the walk.random_increment_rule— it sets the scale of the random step the increment rule produces at each point.
It resizes steps while the walk keeps its position, so it does not discard position and relaunch — the random_restart_policy is Random-Restart Schedule, which teleports where this one merely scales — and it drives a real walk rather than a synthetic one, so expected_spread_horizon forecasting belongs to Random-Walk Simulation.
Related¶
- Instantiates: Bounded Random-Walk Navigation — it is the one dial that governs how far each increment reaches.
- Sibling mechanisms: Random-Walk Simulation · Cumulative Displacement Dashboard · Drift vs. Noise Test · Exploration Capture Protocol · Path-Trace Audit · Random-Restart Schedule · Reflecting-Boundary Rule · Walk-Budget Review · Absorbing-State Trigger
Editorial Notes¶
Form Classification¶
Form family: Control, Automation & Runtime
Rationale: Step-Size Throttle operates as a live operational control that automatically routes, enforces, adapts, or responds during execution because it governs how large each increment may be — widening steps to explore, shrinking them to settle — on a schedule tied to progress.
Independent corroboration: The frozen evidence defines Step-Size Throttle as 'Governs how large each increment may be — widening steps to explore, shrinking them to settle — on a schedule tied to progress', so its operative form is Control, Automation & Runtime.
Nearest alternative: Analysis, Modeling & Optimization — Step-Size Throttle includes features of an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution, but its defining operation is a live operational control that automatically routes, enforces, adapts, or responds during execution.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Mathematics
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Adaptive step size balances exploration and convergence in numerical optimization.
Related originating lineages:
- Computer Science & Software Engineering — Algorithms implement schedules.
- Systems Thinking & Cybernetics — Gain shrinks near target.
Review resolution: The blind reviewers agree that mathematics is the primary origin and differ only on origin mode disagreement, domain reach disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain convergent because the combined evidence shows independent disciplinary development. The broader reach of universal records portability separately from historical provenance; encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] Robbins, H., & Monro, S. "A Stochastic Approximation Method". The Annals of Mathematical Statistics 22(3), 400–407 (1951). Establishes convergence conditions for a stochastic-approximation update governed by a prescribed sequence of positive step sizes. registry ↩