A System Is a Way of Looking

Definition

Weinberg argues that "system" names a set, and a set has no existence independent of the act of selecting its members. What counts as a system's parts, and how finely they are discriminated, is set by an observer's "scope" (what kinds of observations she can make) and "grain" (how finely she can distinguish within each kind) — not read off some pre-existing structure in the world. As the third of his three baseball umpires puts it, describing how he calls balls and strikes: "They ain't nothin' 'til I call 'em."

In the Book

Weinberg opens Chapter 3 by contrasting definitions of "system" that treat it as a self-evident "set of objects" (citing Hall and Fagen) with his own insistence that someone must first choose the objects, and that the choosing is where the real content lives. He builds a working example from Robert Herrick's poem about a mistress's disordered dress: Herrick's "scope" as an observer is the set {Dress, Disorder}, and his "grain" within Dress is {lawn, lace, cuff, ribbands, petticoat, shoestring}. Combining scope and grain via the Cartesian product generates every observation Herrick is capable of making — but the resulting model can be too broad (crediting Herrick with discriminations he cannot actually make, an "error of composition") or too narrow (missing a dimension he is actually using). Weinberg illustrates the latter with a psychology experiment: a pigeon was thought to be responding to a card's {Color, Shape}, but the apparatus made a faint click that differed by card, and the pigeon's real scope was {Color, Shape, Click} — invisible to the experimenter until tested.

Why It Matters

This reframes "what are the parts of this system" from a factual question with a right answer into a design question about what an observer needs to notice for a given purpose. It explains why two competent analysts can draw the "same" system completely differently (an economist's firm and an engineer's firm) without either being wrong, and it puts the burden of justification on the modeler's chosen scope and grain rather than on the "true" structure of the thing being modeled — a discipline useful anywhere models get built, from software architecture to org charts to scientific theory.