John Holland

Holland defined a CAS as a population of *adaptive agents* whose interactions, governed by conditional IF–THEN rules, produce system-level behavior that emerges from the bottom up rather than from central control. He reduced every CAS to seven elements: four *properties* - aggregation, nonlinearity, flows, and diversity - and three *mechanisms* - tagging, internal models, and building blocks. Source: Holland, *Hidden Order: How Adaptation Builds Complexity* (1995; reviewed at DOI 10.2307/20047667). The reviewer's opening move on any space proposal is to ask: *who are the adaptive agents, what conditional rules do they actually follow, and what behavior emerges that no agent intends?*

Holland's foundational invention: a search procedure that maintains a population of candidate solutions encoded as strings, and improves them through selection, crossover (recombination), and mutation. His *schema theorem* proved that short, low-order, above-average "schemata" - partial solution patterns - receive exponentially increasing trials over generations. Source: Holland, *Adaptation in Natural and Artificial Systems* (1975; MIT Press edition DOI 10.7551/mitpress/1090.001.0001; reviewed in SIAM Review DOI 10.1137/1018105). The reviewer uses this to judge any space-systems optimization that invokes "evolutionary" or "genetic" methods: *is the encoding meaningful, is recombination actually exploiting building blocks, or is it a glorified random restart?*

Adaptation works by discovering useful low-order partial solutions ("building blocks") and recombining them into higher-order structures; complexity is *built*, not designed whole. Holland insisted that crossover, not mutation, is the engine, because it assembles independently discovered blocks. Source: Holland, *Adaptation in Natural and Artificial Systems* (1975/1992); facetwise-model treatment in Goldberg/Sastry, "John H. Holland, Facetwise Models, and Economy of Thought" (DOI 10.1093/oso/9780195162929.003.0008). For the reviewer, this is the test of whether an architecture is *modular and recombinable* or monolithic and brittle.

The defining phenomenon Holland chased across his career: lawful, persistent macro-level patterns (markets clearing, ant colonies foraging, immune systems learning) that arise from simple local rules and cannot be read off the rules of any single agent. Emergence is "much coming from little," and is the reason reductive analysis of agents fails to predict system behavior. Source: Holland, *Emergence: From Chaos to Order* (1998); CAS framing in *Hidden Order* (DOI 10.2307/20047667). The reviewer demands that any system-level claim ("the constellation will self-stabilize," "operators will internalize risk") be derived from the agent rules, not asserted.

Holland's agents carry *tags* (markers that enable selective interaction and aggregation), and *internal models* that let them anticipate - to act on predicted, not just current, state. Adaptation is the continual revision of these rules under credit assignment (his "bucket-brigade" algorithm in classifier systems). Source: Holland, *Hidden Order* (1995) and "Holland and the Evolution of Economics" (DOI 10.1093/oso/9780195162929.003.0020), which documents how Holland's agent model reshaped complexity economics at Santa Fe. The reviewer asks whether a space-governance scheme models agents as *anticipating, learning, tag-sorting* actors, or as static rule-followers.

CAS rarely settle to a fixed equilibrium; they exhibit perpetual novelty because each adaptation changes the fitness landscape for every other agent (co-evolution). Holland warned against equilibrium-based analysis of adaptive systems. Source: Holland, *Hidden Order* (1995); economics linkage in DOI 10.1093/oso/9780195162929.003.0020. The reviewer flags any space model that assumes a stable end-state equilibrium where agents are in fact co-evolving (operators, regulators, debris).