Working papers

Before We Say ‘System’

Two questions the word leaves undeclared, and a third that only the world can answer

Terry Cooke-Davies. Working paper. 21 October 2026. Comments and disagreement welcome.

Why a word matters

The word system is applied, in ordinary and scholarly use alike, to gas in a cylinder, a living cell, a motor car, a railway network, a legal jurisdiction, a project, and the processes by which people and institutions judge and act. The variety is usually treated in one of two ways: as loose usage to be tightened, or as evidence that the word means several different things. This paper takes a third course. It does not claim that the word has one meaning; its uses may well form a family in Wittgenstein’s sense, overlapping and criss-crossing without a single shared feature. The claim is more modest and more useful. One working definition organises a remarkably wide range of these uses, and once it does, it shows the word to be missing not a meaning but two pieces of information which every use requires and none supplies.

This matters beyond terminology, for two reasons. The first is that the long dispute between ‘hard’ and ‘soft’ systems thinking, which has run for half a century, turns on one of those missing pieces, and continues partly because the vocabulary gives the participants no way of noticing it. The second is larger. Much of what modern living depends upon — firms, projects, markets, jurisdictions, supply chains — is spoken of as a system, and the word carries no warning about what most often goes wrong with such arrangements: that the consequences of what they do do not travel where their authority does.

The definition

To call something a system is to draw attention to it in a particular way: to treat it as a demarcated whole whose parts interact to produce behaviour that cannot be worked out from the parts taken one at a time. Attending to something in this way is always possible. Whether it is apt is a separate matter, settled by the purpose in hand and, in the end, by where the consequences of acting upon it travelled.

What varies from one use to the next is what justifies the boundary and what sustains the coherence, and the word carries no indication of either. Every use is therefore incomplete until two things are declared.

The first question: what holds the boundary in force?

There are four answers. A boundary may be self-maintained, held by the thing itself at continuing cost, as a cell membrane is. It may be sanctioned, held in force by an authority whose recognition constitutes it and whose withdrawal would dissolve it, as a licence or a charter is. It may be reciprocal, held by mutual acknowledgement among peers with no authority above them, as the boundaries of nation states are. Or it may be drawn by an inquirer for a purpose, and sustained by nothing but that purpose.

The second question: what holds the whole together?

Again there are four answers. Coherence may be constitutive, requiring no production at all, as gravity holds a solar system. It may be self-produced, by the parts’ own continuing operation, as in a cell or a court. It may be externally supplied, built and maintained from outside, as in a machine. Or, commonly in human affairs, it is mixed.

The two questions are independent of one another. An institution may be demarcated by sanction while its coherence is self-produced; an inquirer’s boundary may be laid over a coherence that a living body genuinely produces for itself. The characteristic error is not using the word in various ways, which is unavoidable. It is importing the answers from one case while speaking of another: treating what an inquirer drew as though the thing maintained it, or treating what sanction holds in force as though it had been found in nature.

Sixteen ordinary uses

The table groups everyday uses of the word by what holds the boundary, so that the independence of the second question can be seen: within each group the answers to it vary. The table is not a scale from hard to soft, or from real to notional. The immune system and the legal system give the same answer to the second question; so do the living cell and the banking system.

Three observations follow. First, inquirers’ boundaries differ greatly in how far the world constrains them. A watershed drains as it drains whether or not anyone is looking, and the terrain leaves almost no discretion about where the basin ends, which is why such a boundary feels found rather than drawn. But the decision that this basin is the whole under attention is still an inquirer’s act. Causal structure in the world and demarcation in the description are different things, however tightly the first constrains the second.

Second, most real cases carry several boundaries and several sources of coherence at once. A motor car is type-approved on one boundary, insured on another and taxed on a third. A project is coordinated by its management and also held together by relationship, routine and professional identity. Where the coherence formally supplied and the coherence people produce for themselves part company is, in social cases, where most of what is worth studying is found.

Third, reciprocal boundaries are recursive. Who counts as a peer is itself a boundary held in force, and the same two questions apply to it. A definition that brings this to the surface is doing its work.

Two familiar uses fall outside the definition altogether. A filing system has no interacting parts; what holds it together is an ordering scheme, not a process. ‘Newton’s system’ or ‘a system of thought’ belongs to the same second sense: an ordered arrangement of formulations rather than parts in interaction. That is where genuine double meaning lies, and it is a different word in all but spelling.

What this shows about the hard–soft dispute

Peter Checkland’s own comparison of the two stances is explicit. The hard stance assumes that the world contains systems which can be engineered, and holds that systemicity lies in the world. The soft stance treats systems models as intellectual constructs to aid debate, and holds that systemicity lies in the process of inquiry into the world.

The asymmetry in that pairing is easily missed. The hard stance asserts a location: systemicity lies in the world. The soft stance says that it lies in the process of inquiry, which does not assert that it fails to lie in the world as well. It is a relocation accompanied by silence, not a denial. Only one of the two positions carries an ontological commitment, while the dispute has been conducted as though both did. The charge of relativism regularly levelled at the soft position presupposes a symmetry that is not there.

That is why the tension does not resolve. The two stances are not rival answers to one question. The hard thinker hears the soft thinker’s silence as a claim that nothing is out there; the soft thinker hears the hard thinker’s confidence as a failure to notice that perception is doing work. Neither has said what the other heard, and because the word carries no marker for which question is being answered, or about what, neither can see the misattribution taking place. The dispute sustains itself not because its participants are careless but because the vocabulary affords no way of noticing.

The more recent form of the argument, between realists and constructivists in systems engineering, has the same shape. Realists assert that a boundary is maintained by the thing itself; constructivists that it is drawn by an inquirer. Each treats the other as the only alternative. But these are two of four answers. Sanction and reciprocity are not positions in that argument at all, and they are precisely the answers that fit the cases the argument is usually fought over: boundaries that are neither maintained at physical cost by the thing itself nor free for an inquirer to redraw. A jurisdiction, a registration, a charter, a licence, a contract can be pointed at, litigated over and enforced; they also dissolve if the recognition that constitutes them is withdrawn. A project belongs here, which is why the question whether a project really is a system has never been settled by either of the two available answers. Its boundary is neither found nor merely notional. It is recognised, and recognition is a third kind of thing.

Each camp will try to absorb this answer rather than concede it. A realist will say that a sanctioned boundary is real, since it holds independently of any particular observer. A constructivist will say that 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 like that. The distinction earns its place by what it lets a reader check.

Where this stands in the tradition

A definition offered to a field with seventy years of definitions owes an account of what it adds and what it borrows. It was set against sixteen others, from Norbert Wiener and Ross Ashby to the international systems-engineering standards. The findings can be stated briefly.

Most definitions are silent on both questions. Bertalanffy, Ackoff, Jackson, Meadows, the systems-engineering standards and the dictionary all say 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 literature about the definitions is not silent, however. A group of Fellows of the International Council on Systems Engineering, surveying more than a hundred definitions, found that the inconsistencies between them arise from implicit and undeclared ontological commitments, which definers rarely spell out. They then devoted most of their effort to arguing the realist case against the constructivist one — the first question, in another vocabulary, and answered once more as though it had only two answers.

The most striking antecedent is Ross Ashby, writing in 1960. A system, he says, is not a thing but a list: whatever set of variables an observer selects from the infinity available on a real machine, and the right to select arbitrarily cannot be waived. That is the inquirer’s answer to the first question, two decades before Checkland. But ten sections later he asks the second question, and answers it quite differently. If an active and relevant variable has been left out, the system’s behaviour becomes capricious, not reproducible at will. 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. One is free to draw the line anywhere, and not free to make the world behave reproducibly across a badly drawn one. That relation between our maps and the world — freedom in the drawing, answerability in the outcome — is the relation on which the rest of this paper depends, and it was already present in the definition of the word in 1960.

Wiener had earlier described the self-maintaining, self-producing case. Living organisms, he wrote in 1950, are local enclaves of organisation within a general increase of entropy, maintained by negative feedback, carrying forward a special environment of their own only until they begin to decay faster than they can reconstitute themselves. Maturana and Varela later made that case the definition of the word itself, as autopoiesis. It is a powerful position, but it purchases its clarity by excluding most of what the word is ordinarily used for, including every arrangement held in force by sanction.

C. West Churchman, in 1970, placed the boundary of a system where the decision maker’s power to change things ends. That ties the boundary to authority rather than to the inquirer’s convenience, and it is the nearest approach to sanction in the literature examined. It is not the same thing: the decision maker’s reach fixes the edge, rather than a third party’s recognition that constitutes the boundary and dissolves it when withdrawn. On the coherence question, David Snowden’s typology of constraints — rigid, governing, enabling, absent — is a better instrument than the four answers given here, because it distinguishes constraints by how tightly they determine what follows rather than by where they come from. The two cut the same question differently and can be laid across each other.

The diagnosis that the word is doing two jobs is not new either, and neither are attempts to cure it. Ashby’s remedy was stipulation: in his book the word would mean the abstraction. Checkland in 1988 proposed surgery on the vocabulary, reserving ‘system’ for the intellectual construct and adopting ‘holon’ for it. The systems-engineering Fellows in 2017 proposed a framework of typologies onto which existing definitions could be mapped. None has settled the matter. This paper proposes a fourth remedy: keep the word, and require that each use declare its two answers. A requirement of disclosure asks less of the field than a new vocabulary, a private convention or a framework that must be adopted whole, and it can be taken up one paper, or one meeting, at a time.

What the definition borrows, then, is the inquirer’s answer (Ashby, Churchman), the self-producing case (Wiener, Maturana and Varela), a test of whether a boundary was well drawn (Ashby) and the finer account of coherence (Snowden). What it adds is the naming of the two questions as a pair; the demonstration that they are independent; sanction and reciprocity as answers to the first, which none of the definitions examined supplies, though institutional and legal theory may hold something close; and the proposal that the answers be declared each time the word is used.

A third question: where do the consequences go?

The two questions above can be answered by inspecting the thing. A third cannot. Where do the consequences of acting within this boundary actually travel, and who bears them? The answer is not derivable from the other two. A living cell and a banking system give the same answer to the second question, yet they lie at opposite ends of this one: the cost of a cell’s failure falls on the cell, while much of the cost of a banking system’s failure falls on people outside it. The answer has to be established by tracing what happened.

The question is deliberately not whether the boundary was wrong. A boundary is rarely simply wrong; it is fit for some purposes and not for others. A project’s sanctioned boundary may be entirely adequate for authorising expenditure and wholly inadequate for registering ecological consequence. Adequacy is judged afterwards, from where the consequences went, rather than decided beforehand from the map.

The question is also kept out of the definition, deliberately. The definition adjudicates nothing about where systemicity resides, and anyone in the systems tradition may accept it whatever their ontology. The third question belongs to a different kind of claim.

Taken together, the three questions form a progression in which more of the answer passes, at each step, from us to the world. The first is ours to declare. The second the world confirms or refuses, as Ashby saw, by behaving reproducibly or capriciously. The third only the world can supply, and it supplies it over time, in bodies — human and other — and in the living systems they draw upon. We are free to draw our maps, but not free to choose what the world does in reply. A map remains accountable only so long as what the world does in reply reaches whoever holds the map.

The tradition has come close to this question several times, and each approach is instructive. Wiener 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 very definition of control. But 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 somebody else’s body lies outside his frame. Churchman reported a seminar in which twenty-one of twenty-two studies concluded that the organisation examined was serving the wrong client — the hospital its doctors rather than its patients, the company its senior managers rather than its customers — and observed that practitioners hope their mistakes will not come back to haunt them, and try to arrange matters so that no error proves fatal. Werner Ulrich, in 1983, asked who should witness for those affected by a design but not involved in it, and traced how planning actually selects whom it serves. But his remedy closes in argument: the affected gain a voice and a witness, not standing to halt, and nature appears only as environment or through a human who chooses to speak for it. His loop closes in discourse; the loop at issue here closes in consequence.

Niklas Luhmann gives the strongest opposing answer, and gives it openly. For him a social system is made of communication connecting to communication, and the consequences of action reach it only once they have become topics of communication. Dead fish are floating in the Rhine: what is sociologically interesting, he said, is which systems communicate about this and with what consequences — not the fact that the fish are dying. He may be right as a description. But the fish go on dying whether or not anyone is talking about them, and until someone does, the cost falls on whoever lives with the river: the fish, the people downstream, the river itself.

Once one asks where consequences travelled, one commonly finds that they did not return to the party whose sanction keeps the boundary in force. Authority is bounded; consequence is not. Consequences propagate through a widening cone whose extent depends on which consequence is traced and over what span of time. Where that happens, correction fails for structural rather than moral reasons: those positioned to change the arrangement are not those who register what it does, and nothing in the arrangement carries the one to the other. This is a claim about arrangements, not about systems, and it is to be tested case by case rather than assumed. But the word system, as ordinarily used, imports no warning of it.

What the three questions are for

The formulation the project field reaches for — the project as a system, the megaproject as a system of systems — is not false in the way it is usually accused of being. A project does have a real boundary, held in force by contract, charter and board approval. The error is not that its boundary is fictitious. It is that consequences do not respect it. That mismatch can be investigated, which is the practical gain. Rather than debating whether a megaproject really is a system, one may ask: what boundary governs the authority to decide; what holds the undertaking together; where do the consequences of decisions taken under that authority actually propagate; and who is positioned to register them?

The same questions apply wherever the word is used to speak of something people have arranged: a public service, a firm, a supply chain, a programme of reform, a technology released into ordinary life. They are not a method, and they are not offered as one. They are a discipline of disclosure. The first two make any use of the word checkable by a reader. The third keeps the arrangement answerable to the world it acts upon.

Read in this light, the half-century dispute between hard and soft systems thinking was never really about whether systems are real. It was about who is entitled to say where a boundary lies. The more pressing question, for the arrangements on which modern living now depends, is whether what happens beyond the boundary ever comes back to those who hold it in place.

Sources

W. R. Ashby, Design for a Brain: The Origin of Adaptive Behaviour, 2nd edn (Chapman & Hall, 1960), §§2/5 and 2/15.

P. B. Checkland, Systems Thinking, Systems Practice (Wiley, 1981; revised edn 1999); ‘The case for holon’, Systems Practice 1(3) (1988); with C. J. Atkinson, ‘Extending the metaphor system’, Human Relations 41(10) (1988); with S. Holwell, ‘“Classic” OR and “soft” OR — an asymmetric complementarity’, in M. Pidd (ed.), Systems Modelling: Theory and Practice (Wiley, 2004).

C. W. Churchman, ‘Operations Research as a Profession’, Management Science 17(2) (1970), B37–B53.

M. C. Jackson, Systems Thinking: Creative Holism for Managers (Wiley, 2003); with P. Keys, ‘Towards a system of systems methodologies’, Journal of the Operational Research Society 35 (1984).

N. Luhmann, ‘System as Difference’, Organization 13(1) (2006), 37–57 (a lecture given at Bielefeld in 1991).

H. Maturana and F. Varela, Autopoiesis and Cognition (Reidel, 1980).

H. Sillitto, D. Dori and colleagues, ‘Defining “System”: a comprehensive approach’, INCOSE International Symposium 27(1) (2017); D. Dori and H. Sillitto, ‘What is a system? An ontological framework’, Systems Engineering 20(3) (2017).

D. Snowden, on constraints, after A. Juarrero, Dynamics in Action (MIT Press, 1999).

W. Ulrich, Critical Heuristics of Social Planning (Haupt, 1983).

G. Vickers, The Art of Judgment (Chapman & Hall, 1965).

N. Wiener, The Human Use of Human Beings: Cybernetics and Society, revised edn (1954).

L. Wittgenstein, Philosophical Investigations (1953), §§65–71.

Also compared: L. von Bertalanffy, General System Theory (1968); R. L. Ackoff, ‘Towards a system of systems concepts’, Management Science 17 (1971); D. Meadows, Thinking in Systems (2008); ISO/IEC/IEEE 15288; A. D. Hall and R. E. Fagen, ‘Definition of system’, General Systems 1 (1956).

This paper combines and supersedes two notes to file: 260808a, What We Mean by ‘System’ (version 5), and 261020a, What Others Have Meant by ‘System’ (version 6). The comparison with earlier writers is summarised here rather than reproduced; the fuller record of readings and corrections remains in the second note. The stance on knowledge behind the third question is developed elsewhere as Answerable Realism, and the account of why consequences fail to return as Recognition Theory; nothing in this paper depends on either name.

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.