Working papers

What Others Have Meant by ‘System’

A comparison table, and an assessment of what my working definition of system adds

Terry Cooke-Davies. Note to file, version 6, 20 October 2026. Comments and disagreement welcome.

This note, and the earlier note it refers to as 260808a, have since been combined into a single paper, Before We Say ‘System’, which replaces both. This note is kept here for its comparison table and the record of how the definition was tested.

Purpose

My note 260808a proposes that every use of the word system is incomplete until two things are declared: what holds the demarcation in force, and what sustains the coherence. This note asks whether anyone had already asked those questions, and what, if anything, remains for the new definition to add.

The third and fourth columns of the table below are not the authors’ own categories. They record how each definition stands towards the two questions of 260808a — including, in most cases, that it does not address them at all. That silence is the finding.

The table

What the comparison shows

Most definitions are silent on both questions

Bertalanffy, Ackoff, Jackson, Meadows, the systems-engineering standards and the dictionary all specify what a system is made of — elements, interactions, a whole not reducible to its parts — and say nothing about what warrants the boundary or what holds the thing together. This is not carelessness. A definition of composition is what most purposes require. The silence was already present at the field’s founding: Hall and Fagen’s definition, in the first General Systems yearbook of 1956, specifies objects, relationships and attributes and leaves the system’s relation to anything outside it entirely to one side.

An earlier version of this note took that silence to mean the two questions had gone unasked in the mainstream of the literature. The claim holds for the definitions. It does not hold for the literature about the definitions, and the four systems-engineering documents supplied by Ian Roderick, read on 21 August, are what showed it. Sillitto, Dori and seven other INCOSE Fellows surveyed over a hundred definitions and found that the inconsistencies between them arise from implicit, undeclared ontological commitments — the assumptions that must be accepted for a definition to be true — which definers rarely spell out. They then give most of their length to arguing the realist case against the constructivist one, which is the first question in another vocabulary, and they state explicitly that for a certain class of systems there is an objective identification of the system and its boundary, mounted against the constructivist position of the systems engineering body of knowledge.

The first question has four possible answers, not two. A demarcation may be maintained by the thing itself at continuing cost, as a membrane is; held in force by sanction, where recognition by some authority constitutes it and withdrawal dissolves it; held reciprocally, by mutual acknowledgement among peers with no authority above them; or drawn by an inquirer for a purpose and sustained by nothing else. The realists assert the first and the constructivists the fourth, and each treats the other as the only alternative. Sanction and reciprocity are not positions in that argument; they are the answers neither side has on its board — and they are the answers that fit the cases the argument is usually fought over, where a boundary is neither maintained at physical cost by the thing itself nor free for an inquirer to redraw. A four-way account therefore cuts across the dispute rather than joining it, which is a more useful thing to be able to say than that nobody asked.

Each camp will try to absorb the third answer rather than concede it, and the reply should be ready. A realist will say a sanctioned boundary is real, since it holds independently of any particular observer. A constructivist will say sanction is social construction under another name. Both absorptions are available, and both destroy the information the declaration exists to carry: that this boundary depends on a named party’s continuing recognition, and would dissolve if that recognition were withdrawn. Neither a cell membrane nor a watershed on a map behaves that way. The distinction earns its place by what it lets a reader check.

Each of the two questions can be answered in two different ways, and they should not be confused. One says how you would recognise that the boundary, or the coherence, is really there. The other says what holds it there. Ashby answers the coherence question in the first way: a badly drawn selection announces itself as caprice. Miller, and the INCOSE Fellows after him, answer the demarcation question in the same way: a real boundary announces itself as a discontinuity in organisation. The four answers in each column of the table are of the second kind. On recognition, the field has one answer to each question. On what holds the boundary or the coherence in place, each source gives a single answer and treats it as what a system is. None treats it as a question with more than one possible answer, to be declared each time the word is used.

Wiener occupies the self-produced cell first, and names its mechanism

This section was new to v5, and rests on a direct reading of The Human Use of Human Beings (revised edition, 1954). Wiener offers no definition of system. But in describing the living organism he answers both questions, and answers them in the cell the table otherwise credits to autopoiesis. Organisms, he writes, maintain and may even increase their organisation as local enclaves within a general increase of entropy, and they do so by homeostasis: negative feedback mechanisms of the kind found in mechanical automata. The pattern so maintained is, for him, the touchstone of personal identity. That is coherence self-produced, with its mechanism named. The demarcation is present too, though not as a boundary: we live within a very special environment which we carry forward with us, and only until we begin to decay more quickly than we can reconstitute ourselves. That is a demarcation self-maintained at continuing cost, with its failure condition stated — twenty-two years before Maturana and Varela.

His best-known image needs handling with care. We are, he says, whirlpools in a river, patterns that perpetuate themselves rather than stuff that abides. The image does the work he gives it, which is to locate identity in pattern rather than in material. Taken as an account of provenance, it would place us in the wrong cell. A whirlpool’s pattern is held by the flow passing through it, so its coherence is externally supplied; the organism spends energy manufacturing and maintaining the components that carry its pattern, and the whirlpool does not. Wiener’s own word, reconstitute, marks the difference his image elides.

What Wiener adds against the existing rows is precedence and mechanism: the self-produced cell described in 1950, and its coherence attributed to feedback rather than to the production of components. His larger bearing on this note lies with the third question, and is taken up below.

Ashby asks both questions, in two places, without naming them as two

This is the most significant finding, and it is the reason for v2 and all subsequent versions. In §2/5 Ashby writes that because any real ‘machine’ has an infinity of variables, from which different observers with different aims may reasonably make an infinity of different selections, there must first be given an observer or experimenter; a system is then whatever set of variables he selects from those available on the real ‘machine’. It is a list, nominated by the observer, and quite different in nature from the real ‘machine’. He then stipulates that throughout the book, ‘the system’ will always refer to this abstraction and not to the material thing.

That is the inquirer answer to the first question, in the 1960 revised edition, twenty-one years before Systems Thinking, Systems Practice. The relocation of systemicity to the inquiry cannot therefore be credited to Checkland without qualification. What Checkland did was carry the relocation into methodology and make it consequential for practice; the definitional move was already made. The correction costs 260808a nothing, since it strengthens the claim that the first question is where the dispute turns — it shows the question was live at the foundation of the field and still went undeclared thereafter.

Ten sections later, in §2/15, he asks the second question. He restates that a system is any arbitrarily selected set of variables and that the right to arbitrary selection cannot be waived — and then argues that if a system is to be studied with profit its variables must have some naturalness of association. The test he supplies is behavioural: if an active and relevant variable is left unobserved or uncontrolled, the system’s behaviour becomes capricious and not capable of being reproduced at will. Formalised, a particular input value and a particular state must determine uniquely which transition follows, and this must hold across every pair.

The two sections together are more than the sum of the parts, and this is what the first version of this note missed. Demarcation is the observer’s and freely his. Coherence is not his at all: the world decides whether the selection was well drawn, and delivers its verdict as caprice. The boundary is nominated; the coherence is discovered. That is Answerable Realism inside the definition of the word, in 1960 — one is free to draw the line anywhere, and not free to make the world behave reproducibly across a badly drawn one. Caprice is what a poor demarcation feels like from the inside.

What 260808a still adds against him is now more precisely stated. He has one answer to the first question and treats it as the only possible answer; self-maintenance, sanction and reciprocity are absent, so he cannot say that a cell holds its boundary at metabolic cost whether or not anyone is watching. And his answer to the second question is a test rather than a source: he says how coherence would be detected, not what produces it. The four-way answer concerns provenance, his concerns detection, and the two are compatible. They could usefully be stated together.

Churchman locates the boundary at the edge of someone’s power, and asks where the benefit goes

This section was new to v4, and rests on a direct reading of ‘Operations Research as a Profession’ (1970). Churchman says plainly that the determination of the bounds of the system, and hence also the nature of the decision maker, are strategic problems of the OR inquiring system. That is the inquirer answer, a decade after Ashby and a decade before Checkland. But he adds something Ashby does not: the limits of a system and the decision maker are correlative. The environment is what influences effectiveness but the decision maker cannot change; resources are what he can. The boundary therefore falls where somebody’s power to act ends. That is demarcation by reach — adjacent to sanction, since it ties the boundary to authority rather than to the inquirer’s convenience, but not sanction, since it is the decision maker’s power that fixes the edge and not a third party’s recognition that constitutes it and dissolves it on withdrawal. The table entry is amended accordingly, from ‘nearest approach to a third answer’ to ‘adjacent to sanction’.

Two further things in the paper bear on 260808a. First, Churchman names the characteristic error that 260808a warns against: practitioners assume that the law lies in the environment of the decision maker and translate it into rigid constraints, when laws are ambiguous and a good lawyer knows that treating one as fixed may be a very poor strategy — what sanction holds in force treated as though found in nature. Second, he asks the third question before it had a name. He reports a graduate seminar in which twenty-one of twenty-two papers concluded that the social system studied was serving the wrong client: the hospital the doctors rather than the patients, the company its top managers rather than its customers, the counter-insurgency division the Pentagon rather than the natives. Where the benefit travels is not derivable from the definition; it has to be traced. And he records, as a mystery of the profession rather than a method, that most practitioners pray their mistakes will not come back to haunt them, and that the astute practitioner tries to create an environment in which no error will ever prove fatal. That is the severance of consequence described from inside the profession, by its founder, fifty-six years ago.

Autopoiesis collapses the two questions the other way; Luhmann collapses them and then answers the third

Maturana and Varela define a system as one that produces its own components and thereby its own boundary. Demarcation and coherence are made to coincide by definition. That is one cell of the table — self-maintained, self-produced — elevated into the definition of the word. It is a coherent position and a powerful one, but it purchases its clarity by excluding most of what the word is ordinarily used for, including every arrangement held in force by sanction. The cell itself was not new: Wiener had described it as homeostasis in 1950, without making it the definition of the word.

Luhmann carries the same move to social scale, and a direct reading of his 1991 Bielefeld lecture ‘System as Difference’ (published in English in 2006) makes the entry more exact than v3 had it. A system is the difference between system and environment; it creates itself as a chain of operations, and the difference arises merely because one operation produces a subsequent operation of the same type — for social systems, communication connecting to communication. Demarcation and coherence are one operation. But he then does what neither Maturana nor Varela does: he keeps two concepts of environment apart. There is the environment as the system defines it for itself, through its own external reference, and there is the environment an external observer sees, which includes the system and its environment together; the observer may see many more and quite different things than are accessible to the system. That is the inquirer’s demarcation, acknowledged and held separate from the system’s own — the independence of the two questions, conceded by the one theorist who fused them.

He also answers the third question, and the answer is the reason this note now treats him as more than one silent source among sixteen. The outside world, he says, has only limited importance for the consequences of communication; communication never exits the system, since the next connection is always provided for within it. Dead fish are floating in the Rhine: what is sociologically interesting is which systems communicate about this and with what consequences — not, he says in terms, the fact that the fish are dying. Bodies are not in the social system at all; consciousness is a separate closed system, and he counts it a good thing that no one can inject a perception into another’s. Consequences reach the social system only after they have become topics of communication. This is the operation 260808a and its companion papers describe as the short circuit — representations checked against representations — stated as the constitution of the social rather than as a failure of it. Luhmann is therefore the strongest opponent the companion argument has: he grants the mechanism and denies that it is a fault. What the companion argument adds is not a better description but the observation that the fish die whether or not the communication continues, and that somebody’s body pays for the interval between the dying and the saying.

One further thing from the lecture belongs on the record. Luhmann states that every social theory is part of the social domain and as such part of what it describes, and that sociologists can adopt the attitude of external observers ‘who we can never be in reality’. Whatever else divides him from 260808a, he declares the observer’s position rather than assuming it.

Snowden answers a different question about coherence, and answers it well

Cynefin does not define system; it classifies by constraint, after Juarrero — rigid, governing, enabling, absent. Read against 260808a, that is a four-way answer to the coherence question, and a more discriminating one than constitutive, self-produced, externally supplied and mixed, because it distinguishes constraints by how tightly they determine what follows rather than by where they come from. The two typologies are not rivals; they cut the same column differently and could be laid across each other. Snowden offers nothing at all on the demarcation question, which is worth noting given how often Cynefin is invoked to settle what kind of thing one is dealing with.

Four remedies to one problem, across sixty-six years

This is the finding most worth having, and it is a sequence rather than a contrast of two. Ashby’s remedy in 1960 was stipulation: he announces that in this book the word will refer to the abstraction and not to the real ‘machine’, and proceeds. That binds his own text and nothing else. In 1988 Checkland published ‘The case for holon’, and with Atkinson in the same year argued that the notion of system as an adaptive whole surviving in a changing environment is an unexamined given that ought to be challenged. His diagnosis was this note’s own — the word is doing two jobs and the confusion is systematic — and his remedy was terminological surgery: reserve ‘system’ for the intellectual construct, adopt ‘holon’ for it.

Neither remedy took. Stipulation binds only the stipulator; almost nobody outside the immediate circle says holon; and the word went on doing both jobs throughout. A third was attempted in 2017, and attempted at scale: the INCOSE Fellows proposed a framework of typologies onto which existing definitions could be mapped, together with a family of definitions in neutral language. The dispute it was built to settle is still running. 260808a proposes a fourth: keep the word, and require that each use declare its two answers. A disclosure requirement asks less of the field than a change of vocabulary, less than a private convention, and less than a framework that must be adopted whole, and it can be taken up one paper at a time. Whether that is why it might travel where the others did not is a claim about uptake and cannot be settled from here — but four remedies to one problem across sixty-six years is a better shape for the argument than three, and the shape belongs in the IJPM editorial rather than being spent here.

One qualification belongs with the fourth remedy, and it narrows the claim rather than unsettling it. Hybertson attaches a statement of ontological commitment to his own definition of system, and the INCOSE Fellows name him, together with Sillitto’s 2014 book, among the rare cases where such commitments are spelt out. The primary text was sought on 25 August and was not reasonably obtainable — an expensive edition, poorly reproduced, and a delivery that could not be relied on — so the assessment here rests on the secondary account and on the author’s stated position, where the textual descriptions of systems and the ontological claims made for them do not appear to sit consistently together. On that evidence this is disclosure practised once, in an author’s own book, rather than proposed as a discipline others must follow. The claim made here is for the requirement, not the instance: two named questions, answered in a form any reader can check against the paper’s own argument. Doing it once is not proposing that anyone else must, and an inconsistent instance is further from a requirement rather than nearer to it. The claim would be weakened, though not destroyed, if a direct reading showed Hybertson arguing that all definitions ought to carry such a declaration.

Sanction appears to be genuinely absent

Nothing in the table supplies it. Churchman comes closest, and the direct reading above makes the distance precise: his boundary is fixed by the reach of the decision maker’s power, not by a recognition that constitutes it and whose withdrawal dissolves it. Ulrich, likewise, treats boundary-setting as an act for which the analyst is answerable — the analyst’s act, not a third party’s recognition. On the evidence assembled here, the third answer to the first question is the paper’s own contribution. That finding remains provisional: it rests on a survey of the best-known definitions, not on a systematic search, and institutional and legal theory has not been looked at, where something close may well exist.

One antecedent, tested directly

Werner Ulrich’s Critical Heuristics of Social Planning (1983) sets out twelve boundary questions, among them who ought to serve as witness for those affected by a design but not involved in it. That is close to the third question — where do the consequences travel, and who bears them — and close to Recognition Theory’s bearer-side concern more broadly. Two candidate differences were pencilled in here. The text was read directly on 12 August and one of them failed.

The first held that Ulrich’s questions are asked prospectively, by a designer deciding where to draw a boundary, while the third question is asked retrospectively, by tracing where consequences actually went. That is wrong. His leverage-points figure for tracing institutional selectivity is retrospective and symptom-led: four points of selectivity, each with observable symptoms, reconstructing the actual client map of administrative planning against the claimed one. The negative finding is worth more than the resemblance was, and it is recorded here rather than quietly dropped.

The second survives, and is sharpened by the loss of the first. Ulrich’s tracing terminates in argument. Once the actual client map is exposed, the affected employ boundary judgements polemically, shifting the burden of proof onto the planners. They gain a voice, a witness, argumentative symmetry. They do not gain standing to halt, and it is nowhere asked whether the harm returns to those whose sanction keeps the map in force. His loop closes in discourse; this one closes in consequence. That is the single load-bearing difference, and it is cleaner than either of the two it replaces. What Recognition Theory adds to Ulrich is therefore not the affected — he has them — but the lever and the return.

A further finding bears on the demarcation question rather than this one, and it extends the same difference to the living world. Ulrich’s own architectonic declares its root metaphor as the social group rather than machine or organism, and in his diagram of the two boundary judgements both the involved and the affected are social actors. The natural world is not in the affected circle; it appears as environment, filed under sources of control, or through a human who elects to speak for it. His scheme can represent it in the discourse but cannot let it stop anything. The exclusion is declared by the author, not discovered by a critic, which makes it citable rather than contestable. The steelman is that rejecting the organism was an emancipatory move against organicism in social theory, which by 1983 carried real authoritarian baggage; the cost was that in rejecting the organism as a model of society he also lost it as the environment of society.

Summary of what 260808a adds

First, it puts the two questions together and names them as two. Ashby asks both, ten sections apart, and does not mark them as a pair; Luhmann fuses them and then, in keeping the observer’s environment apart from the system’s, concedes their independence without saying so; nobody else in the table asks them together at all. Second, it shows they are independent, which is what the grouping of the three-column table in 260808a demonstrates and what no source above states. Third, it supplies sanction as a third answer to the first question, which appears to be new: Churchman’s demarcation by reach is the nearest antecedent, and it is not the same thing. Fourth, it answers the second question by provenance where Ashby answers it by detection, and where Snowden answers it by the tightness of constraint. Fifth, it offers a disclosure requirement — two named questions, checkable by any reader — where Ashby offered stipulation, Checkland offered a new word, and the INCOSE Fellows offered a mappable framework.

What it does not add: the inquirer answer, which is Ashby’s, and Churchman’s a decade later; the self-maintained and self-produced cell, which Wiener described as homeostasis in 1950 and Maturana and Varela formalised as autopoiesis; a test for whether a demarcation was well drawn, which is Ashby’s §2/15 and which 260808a lacks; and a superior account of coherence, where Snowden’s constraint typology is the better instrument.

One thing the comparison now shows that it did not before. The third question — where the consequences travel, and who bears them — is not merely absent from the tradition. Wiener in 1950 defined feedback as the control of a machine on the basis of its actual rather than its expected performance: the return of consequence written into the definition of control. His monitors are attached to the machine they monitor, so the coupling is assumed rather than asked about; the case in which the actual performance registers in someone else’s body lies outside his frame. Churchman asked the third question in 1970 and recorded that practitioners arrange to be out of the way of the answer; Luhmann answered it in 1991 by defining the social so that the answer can only ever arrive as a topic. Between the cybernetician who made the return of actual consequence the definition of control, the founder who saw its evasion as a mystery, and the theorist who made the evasion a constitution lies the space the companion argument occupies.

Record of corrections

The correction to v1 is worth stating plainly, since this note is the record. Version one placed ‘not addressed’ in Ashby’s coherence column, on the reasoning that a system which is merely a selection has no coherence to account for. §2/15 refutes that. The error came of reading the definition in §2/5 and stopping there.

Version two rewrote the Ashby row and commentary after reading §§2/5 and 2/15 directly and corrected the dating to the 1960 second edition. Version three replaced the passage on Ulrich after reading Critical Heuristics of Social Planning directly on 12 August, in which one of the two candidate differences failed and is recorded as having failed.

After version three the note developed along two branches, each unaware of the other, and both reached a version five. The first, issued as 260825b, qualified the premise following the four systems-engineering documents supplied by Ian Roderick and read on 21 August, made the sequence of remedies four across sixty-six years rather than three across thirty, and then settled the disclosure claim when the primary text for Hybertson proved not reasonably obtainable. The second, issued as 260918a and 260927a, corrected the provenance note of versions two and three, which said that apart from Ulrich and Ashby the sources had been consulted at second hand — true of the initial trawl, drafted in dialogue with Claude from outside the documents of the project, but never true of the author’s reading; amended the Churchman and Luhmann rows and added commentary on them after reading the 1970 paper and the 1991 lecture directly; corrected the Hall and Fagen citation to what it actually rests on; and added a row and commentary for Wiener after reading The Human Use of Human Beings directly, correcting the attribution of the self-maintained, self-produced cell, which earlier versions credited to autopoiesis alone.

Version six reconciles the two branches and carries both without loss. Where they conflicted, on the premise, the qualified form is kept: the September branch, having diverged before 21 August, restated as surviving inspection the claim the August branch had already withdrawn. The table, the commentary on Wiener, Churchman and Luhmann, and the provenance note follow the September branch; the qualified premise, the four answers set against the realist–constructivist dispute, the symmetry of detection, the four remedies and the Hybertson qualification follow the August branch; the Ulrich section combines both readings. The number of sources compared is sixteen.

Sources consulted for this note, and how

Read directly by the author: N. Wiener, The Human Use of Human Beings: Cybernetics and Society, revised edn (1954), in the Kindle edition with a foreword by Brian Christian, read in September 2026, chiefly the passages on feedback, homeostasis and the whirlpool; W. R. Ashby, Design for a Brain: The Origin of Adaptive Behaviour, 2nd edn (1960), §§2/5, 2/6 and 2/15, in a searchable text, and An Introduction to Cybernetics (1956), to which §2/15 cross-refers; L. von Bertalanffy, General System Theory (1968); C. W. Churchman, ‘Operations Research as a Profession’, Management Science 17(2), 1970, B37–B53 (The Systems Approach (1968) and The Systems Approach and Its Enemies (1979) have not been consulted directly, and the twelve planning categories of the latter are known through Ulrich and Reynolds); R. L. Ackoff, ‘Towards a system of systems concepts’, Management Science 17 (1971), and later work; H. Maturana and F. Varela, Autopoiesis and Cognition (1980); G. Vickers, The Art of Judgment (1965) and papers including ‘Science and the appreciative system’, Human Relations 21 (1968); W. Ulrich, Critical Heuristics of Social Planning (1983), read on 12 August 2026; N. Luhmann, ‘System as Difference’, Organization 13(1), 2006, pp. 37–57, the edited translation of a Bielefeld lecture of December 1991 (Social Systems (1984/1995) has not been consulted directly); P. B. Checkland, Systems Thinking, Systems Practice (1981), ‘The case for holon’, Systems Practice 1(3) (1988), and C. J. Atkinson and P. B. Checkland, ‘Extending the metaphor system’, Human Relations 41(10) (1988); M. C. Jackson, Systems Thinking: Creative Holism for Managers (2003) and Critical Systems Thinking: A Practitioner’s Guide, both read in part; ISO/IEC/IEEE 15288 and the INCOSE definition; D. Meadows, Thinking in Systems (2008); D. Snowden on constraints, after A. Juarrero, Dynamics in Action (1999); the Oxford English Dictionary.

Supplied by Ian Roderick and read on 21 August: Sillitto, Dori and colleagues, two INCOSE papers on defining ‘system’ and on an ontological framework for it (2017) [full citations to be completed]; Ing, on rethinking systems thinking [full citation to be completed]; Walls and Flach, a conversational book asking whether systems exist [full citation to be completed].

Cited at second hand: A. D. Hall and R. E. Fagen, ‘Definition of system’, General Systems 1 (1956), p. 18, as quoted by Bertalanffy (1968); J. G. Miller, Living Systems (1978), and Sillitto (2014), as carried by the INCOSE papers; D. Hybertson, on declared ontological commitment — primary text sought on 25 August and not reasonably obtainable, assessed on secondary evidence only, and the limitation is stated where the assessment is used.

The comparison table was first drafted in dialogue with Claude from outside the project’s documents; every entry except Hall and Fagen has since been checked by the author against the source named.

Version six, superseding 260927a and 260825b. Four further rows recommended by the 21 August reading — Miller; Sillitto and colleagues; Emery and Trist; and Weinberg as carried by Walls and Flach — are pending and are not yet in the table, which therefore under-represents the systems-engineering literature. Entries in the third and fourth columns are this note’s reading of each source against the two questions, not the authors’ own words. Definitions are compressed rather than quoted.

Developed in dialogue with Claude (Anthropic) as part of an ongoing AI-enhanced appreciative practice. The thinking and every word are the author’s own.