Simulation Rescaling Sweep¶
Simulation method — instantiates Scale-Invariance Testing
Runs a model across a planned range of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full scale.
Simulation Rescaling Sweep runs a model — not the real system — across a planned series of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full size. Its defining idea is in-silico search: because testing at true scale is too expensive or too dangerous, you encode the mechanism in a model and sweep a scale parameter across a designed range, reading off the emergent scaling relation and the points where it bends. Everything it produces is a hypothesis about the real world, gated on the model being right; it stops at diagnosis and hands the field the transformation and the decision.
Example¶
A fulfillment center is planning how many autonomous floor robots to buy. Buying 400 to see what happens is the expensive way to learn, so the team instead builds an agent-based simulation of the warehouse floor and sweeps fleet size from 50 to 600 robots in defined steps. Throughput per robot rises with fleet size, then flattens near 320, then falls: congestion at aisle intersections and contention for charging stations create a deadlock threshold around 380, where robots begin to gridlock. The sweep's curve shows the superlinear cost of congestion that no single-fleet-size test would reveal. It cannot hand them the true operating number — the model omits the variance of human pickers sharing the floor — but it hands them the shape: buy in stages, instrument congestion at intersections, and do not exceed roughly 320 without redesigning the aisle layout.
How it works¶
- Design the sweep plan. Set the range of scales and the step density — where to sample densely (near suspected thresholds) and where coarse steps suffice.
- Run the model at each scale. Hold the mechanism fixed and vary only the scale parameter, with enough stochastic replicates to separate signal from Monte-Carlo noise.
- Read off the scaling relation and the breaks. Fit how the emergent quantity moves with scale and mark the thresholds where behavior changes regime — the curvature, the saturation, the tipping point.
Tuning parameters¶
- Sweep range and step density — a wide range with dense steps near thresholds finds regime changes but multiplies runtime; sparse sampling risks stepping over the tipping point.
- Replicates per scale — more stochastic runs at each point tighten the estimate of the scaling curve at the cost of compute.
- Model fidelity — how much of the real mechanism is encoded. Higher fidelity narrows the sim-to-real gap but is slower to build and run, and can overfit to assumptions.
When it helps, and when it misleads¶
Its strength is cheap, safe exploration of regimes that would be ruinous to probe in reality: it finds thresholds and saturation before deployment, when a plan can still change. Its failure mode is model validity — a sweep of a wrong model yields confident, precisely-wrong thresholds, since "all models are wrong; the question is whether they are useful."[n1] The classic misuse is treating a simulation's output as a real transfer decision rather than as a hypothesis to be checked against the world. The guarding discipline is to validate the model against at least one known real scale before trusting its extrapolations, and to carry the sim-to-real caveat forward rather than laundering it away in a tidy curve.
How it implements the components¶
Simulation Rescaling Sweep realizes the pre-deployment, in-silico slice of the archetype — the exploratory search that runs before real evidence exists:
scale_sampling_plan— the designed range and step density of the sweep is its plan; deciding where to sample is the core method.scaling_ratio— it reads the emergent scaling relation off the swept model rather than assuming one.breakpoint_detection— it locates the modeled thresholds, saturation points, and tipping points across the sweep.
Everything here happens in a model, not the field, so it stops at diagnosis: the real-rollout scale_transformation, the bounded transfer_limit, and the go / stage / stop action_update_rule are Pilot-to-Scale Validation's to own.
Related¶
- Instantiates: Scale-Invariance Testing — it is the archetype's what-if engine, mapping scale behavior before commitment.
- Sibling mechanisms: Pilot-to-Scale Validation · Normalized Metric Check · Dimensional Scaling Test · Per-Unit Invariance Check · Stratified Scale Sampling · Breakpoint Review Table · Log-Log Scaling Check
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Simulation Rescaling Sweep operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it runs a model across a planned range of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full scale.
Independent corroboration: The frozen evidence defines Simulation Rescaling Sweep as 'Runs a model across a planned range of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full scale', so its operative form is Analysis, Modeling & Optimization.
Nearest alternative: Experiment, Test & Rehearsal — Simulation Rescaling Sweep includes features of an active test, trial, simulation, drill, or rehearsal that generates evidence through a deliberate attempt or perturbation, but its defining operation is an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Sweeping a planned scale factor and detecting curvature, threshold, or saturation is experimental sensitivity analysis. NASA's parameter-sweep work formalizes systematic variation and output response; engineering supplies scale-up consequences.
Related originating lineages:
- Data Science & Analytics — Data science, analytics, and operational monitoring supplies a parallel or contributing lineage for the mechanism's defining operation: runs a model across a planned range of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full scale.
- Engineering & Design — Engineering design, reliability, and systems-safety practice supplies a parallel or contributing lineage for the mechanism's defining operation: runs a model across a planned range of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full scale.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: runs a model across a planned range of scales to hunt for curvature, thresholds, and saturation before anything is built or deployed at full scale.
- Operations Research — Capacity and congestion models identify scale-dependent operating regimes.
- Physics — Scaling laws and dimensional analysis predict where behavior changes with size.
- Systems Thinking & Cybernetics — Feedback and saturation can create qualitative regime shifts during scaling.
Review resolution: The blind reviewers disagree on primary lineage (engineering_design versus statistics_experimental_design). Authoritative or primary research supports statistics_experimental_design as the best historical origin: Sweeping a planned scale factor and detecting curvature, threshold, or saturation is experimental sensitivity analysis. NASA's parameter-sweep work formalizes systematic variation and output response; engineering supplies scale-up consequences. The cited NASA, Single-Parameter Sweep and Scale Analysis directly supports the mechanism's defining operation. All independently supported contributing domains are retained without an arbitrary cap. origin_mode=convergent records lineage, while domain_reach=multi_domain records later applicability separately from provenance.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
Notes¶
[n1] "All models are wrong, but some are useful" — the statistician George Box's dictum. A rescaling sweep inherits it in full: its thresholds are only as trustworthy as the model that produced them, which is why validation at a known scale precedes any extrapolation. ↩