Working through the formal apparatus, the institutional critique, and the politics of formalism, it becomes apparent that the concept of "state" in "state space" has been operating across the series at several different levels of abstraction that deserve explicit distinction. At the level of computation, a state is a formally specified configuration of variables. At the level of service design, it is a configuration of commitments between agents - what has been promised, what has been fulfilled, what remains outstanding - in the sense that Burgess (2020) and Iqbal (2018) articulate through promise theory. At the level of institutional planning, it is the framework of understanding within which purposeful action becomes possible. And at the level of governance, "the state" is the political apparatus that constructs the categories - citizen, claimant, eligible, ineligible - within which all the other levels operate.
This post is an attempt to clarify those levels. Not because the ambiguity has been careless - the slippage between senses has sometimes been productive - but because the later posts in this series - and my wider research - will need to be precise about which level of abstraction they are working at, and that precision requires making the distinctions explicit.
Sense 1: The formal/computational sense
This is the sense introduced most precisely in "What is a State Space?" and grounded in the AI planning literature. A state space, in this sense, is a formally defined set of possible states a system can be in, together with specified transitions between them - each transition having a source state, a trigger, a target state, preconditions, and effects. Russell and Norvig (2020) and Ghallab, Nau, and Traverso (2004) provide the canonical treatments. The domain model that a classical planner operates on is a state space in this sense; so is the formal structure underlying the Pathway Generator's requirements, though as "The Politics of Formalism" discussed, the genetic algorithm distributes these formal commitments differently from a PDDL-style planner.
The defining feature of the formal sense is precision: states are exhaustively specifiable, transitions have defined preconditions and effects, and the state space as a whole is computationally tractable - meaning an algorithm can operate on it. This precision makes certain things possible: planning (finding paths from initial to goal states), verification (checking whether undesirable states are reachable), and simulation (running the model forward under different scenarios). It also makes certain demands: someone has to have specified the variables, their possible values, the transitions, and the goals. The specification cannot be vague; the formalism does not tolerate ambiguity.
Sense 2: The representational/design sense
This is the sense that operates in the statechart work, in the analysis of service states, and in the critique of journey maps and blueprints for leaving states implicit. A state space in this sense is a shared model of a service system's possible configurations - not necessarily specified with the full rigour of sense 1, but characterised by the discipline of making states explicit as first-class concepts rather than leaving them buried in narrative flows or action sequences.
What distinguishes a service state from a device or interface state is the commitment structure that gives it meaning. This commitment-based conception completes an ontological move that service design theory initiated but left structurally underspecified. The "design for service" literature - Kimbell (2017), Wetter-Edman et al. (2014) - established that services are not products to be designed but relational value-creating processes whose preconditions can be shaped; Kimbell (2017) describes service as "a fluid arrangement of human and nonhuman artefacts, rather than a fixed intangible entity". Vink, Hay and Duan (2025) push this further, arguing for service systems as "the entanglement of manifold relations" in a "constant state of becoming". What this literature does not provide is a vocabulary for describing the specific configuration of those relations at any given moment - a structural unit for the relational ontology it proposes.
The promise vocabulary supplies that unit. Burgess (2020) defines the state of an agent through the promises it makes: agents are distinguished by their commitments, and the states of a system form a configuration of those commitments. Iqbal (2018) builds on this to argue that services are "a set of promises" - demand and supply for affordance, demand and supply for performance - and that the state of a service is the current pattern of which promises have been made, which are being kept, and which remain outstanding. When a statechart of a data access service identifies states such as pending, approved, active, suspended, and terminated, those labels derive their meaning from the promise structure: "pending" means a commitment to provide access has been made but not yet fulfilled; "active" means the core service promises are being kept; "suspended" means promise-keeping has been paused while some condition is assessed. Without the promise vocabulary, these read as database status fields; with it, they describe configurations of commitment between autonomous agents. If services are relational and processual, as the "design for service" trajectory argues, then the question is what those relations consist of at any given moment; the answer - configurations of voluntary commitment between autonomous agents - gives the relational ontology analytical purchase.
The representational sense is looser than the formal sense. A statechart does not enumerate every possible variable combination; it identifies the meaningful configurations a service can be in, the transitions between them, and the conditions under which transitions occur. This is rigorous enough to reveal dead-end states, unassigned transitions, and parallel processes that journey maps conceal, but it does not require the complete formal specification that an AI planner would demand. The discipline is conceptual - thinking in states rather than in actions - and representational - expressing that thinking in a notation that makes states explicit and inspectable.
The representational sense is where most of the practical value for service design sits. It is the sense that allows a designer to ask: what states can this service be in? What triggers transitions between them? Are there states from which there is no exit? Are there transitions with no assigned trigger? These questions can be asked and answered without building a full computational domain model; they require only the discipline of treating states as the primary unit of analysis.
Conventional service design does attempt to represent the experiential dimension of services, typically through the emotional curve that runs along the top of a journey map - a line rising and falling to indicate moments of satisfaction, frustration, anxiety, or delight. Mages and Neely (2023) argue that journey maps offer an "impoverished perspective" on the temporal dimension of experience; the critique extends to their experiential dimension as well. An emotional curve is a single scalar variable: it collapses the multi-dimensional reality of a service encounter - what promises are being kept, what commitments remain outstanding, what expectations have been set and whether they are being met - into a univariate affective signal. A person waiting for a disability assessment is not simply "anxious"; they are in a state where a promise of assessment has been made but not fulfilled, where eligibility remains unresolved, where multiple institutional commitments are outstanding, where the temporal horizon of resolution is uncertain. The emotional label captures one dimension of this; the promise vocabulary captures its structure.
Hay and Vink (2023) identify a related pattern: service design's "systemic turn" toward ecosystem transformation has resulted in what they call a "neglect of actors' emotional experiences" - but the emotional representations the field previously relied on were themselves thin, reducing the complexity of service encounters to a single affective vector. The issue is not that emotion is irrelevant to service experience but that a scalar emotional curve is an inadequate representation of what is actually a multi-dimensional configuration of commitment, expectation, and fulfilment. State-oriented representation does not discard the experiential dimension; it provides the structural vocabulary within which emotional responses become explicable rather than merely plotted.
Sense 3: The institutional/organisational sense
This is the sense that operates in "Beyond Technomagic" and in the broader argument about the relationship between design and planning. The post identifies a specific failure mode underlying technomagic: attempting to plan without a reflexive account of the problem domain - without having enumerated the possible states a system can be in, specified the transitions between them, defined what counts as success, or mapped what actions are available. At this level of abstraction, "state space" is not a term borrowed from computation for rhetorical force; it names a structural requirement for purposeful action that the computational sense (sense 1) happens to formalise with particular precision but does not own. Any form of planning - computational, institutional, governmental - presupposes some account of what the domain looks like, what configurations are possible, and what would constitute progress. The institutional sense identifies the absence of that account as a specific organisational pathology, not merely an incomplete specification.
The institutional sense is distinct from the formal sense not in the structure of its claim but in the degree of specification it demands. A formal domain model must be computationally complete; an institutional understanding of the problem domain need not be, but it must be sufficient for the organisation to distinguish between actions that address the problem as understood and actions that merely perform the appearance of progress. Technomagic fills the gap that absent domain understanding creates: where the problem domain has not been worked out, technology procurement provides concrete, demonstrable action - specifiable and procurable in a way that domain understanding is not - and thereby substitutes the appearance of progress for the conditions that genuine progress would require. Organisations manage to plan, after a fashion, without this understanding; the plans may be poor, the outcomes unintended, the activity procedural rather than purposeful. The institutional claim is not that planning without domain understanding is logically impossible, but that it is epistemically impoverished.
Sense 4: The governmental/political sense
There is a fourth level of abstraction at which "state" operates in this series, one I have been circling without fully engaging: the political state. When policy designers and public administrators refer to "the state", they mean the governing apparatus - the configuration of institutions, laws, entitlements, and obligations that constitutes public authority. But "state" in this political sense is etymologically and conceptually a snapshot: a description of how things stand, a configuration of actors and relations held in a particular arrangement. The "state of affairs" is the current condition; the "state" as polity is the institutional configuration that produces and maintains that condition.
Scott (2020) provides the sharpest account of what this means for the concept of state-space construction. His central argument is that governing requires legibility - the state must simplify complex social reality into standardised categories it can process. Permanent surnames, cadastral surveys, population registers, standardised weights and measures: all are projects of legibility through which the state constructs an informational infrastructure sufficient for governance. "Legibility is a condition of manipulation", Scott writes (2020, p. 11); any substantial state intervention - vaccinating a population, taxing property, conscripting soldiers, distributing welfare - "requires the invention of units that are visible" (2020, p. 11). This is, structurally, state-space construction at the level of governance. The state defines the variables (who lives where, who owns what, who qualifies for what), the values they can take (citizen or non-citizen, employed or unemployed, eligible or ineligible), and the transitions that matter (application to assessment, assessment to entitlement, entitlement to review).
The parallel to sense 1 is not metaphorical - it describes the same structural requirement operating at a different scale. The Pathway Generator's domain model needed formally specified rehabilitation states, transitions, and goal conditions. But the categories within which those states were defined - what counts as "unemployed", what constitutes a "disability", what a "successful rehabilitation outcome" looks like - were not invented by the algorithm's designers. They were inherited from Swedish welfare policy: decades of legislative and administrative work that constructed the categories of eligibility, entitlement, and classification through which the welfare state makes its citizens legible to itself. The Pathway Generator did not just need a state space; it needed one built within a prior governmental state space that defined the domain of vocational rehabilitation.
This is where the governmental sense becomes analytically distinct from the institutional sense (sense 3). Sense 3 concerns what an organisation needs to understand in order to plan meaningfully - the framework of knowledge and shared vocabulary that makes purposeful action possible. Sense 4 concerns the prior political act of constructing the categories within which organisations then operate. When SCÖ's programme governance lacked a coherent account of rehabilitation states, that was an institutional failure (sense 3). But the categories within which vocational rehabilitation is defined as a domain - the administrative classifications, eligibility criteria, and outcome measures - those were constructed at the governmental level, through policy design and legislation. The institutional state space is nested within a governmental one.
Promise theory, which the series engaged with through Burgess (2020), provides a vocabulary that bridges the computational and governmental senses: a state, in Burgess's terms, is a configuration of promises between autonomous agents - the current pattern of who has committed to what, under what conditions, toward what ends. The political state is precisely such a configuration: a set of institutional promises (entitlements, obligations, service commitments) held between citizens and governing bodies. When those promises change - through policy reform, legislative action, or institutional reorganisation - the state transitions. The "state" in "state space" and the "state" in "welfare state" share more structure than their homophony might suggest.
How the four senses relate
The concept's analytical power comes from the connections between these senses rather than from any single sense in isolation. Formal requirements (sense 1) make legible what is absent in institutional planning (sense 3): the Pathway Generator's need for a formally defined domain exposed the fact that SCÖ's programme governance had no coherent account of what rehabilitation states existed, what transitions between them were possible, or what goal states were being pursued. Representational practice (sense 2) provides the bridge: statecharts and state-oriented modelling are rigorous enough to surface gaps and inconsistencies without requiring full computational specification. And the governmental sense (sense 4) reveals that the categories all the other senses work within - the variables, the eligibility criteria, the outcome measures - are themselves products of political construction.
The gradient runs from precision to generality, and from the technical to the political. Sense 1 is the most precise and the most demanding; it enables computation but requires complete specification. Sense 2 occupies a productive middle ground - formal enough to discipline inquiry, loose enough to accommodate the incompleteness and ambiguity of real service domains. Sense 3 captures the institutional prerequisites of purposeful action without requiring formal rigour. Sense 4 is the broadest - it concerns the political processes through which the categories of governance are constructed in the first place. But the gradient also runs in the other direction: sense 4 is logically prior to the others, because the governmental categories constrain what can appear as a variable, a state, or a goal at every other level.
There is a second gradient that cuts across the first: from system configuration to commitment configuration. In sense 1, a state is a configuration of variables - a snapshot of how a system stands, specifiable without reference to any agent's intentions. In sense 2, a state is a configuration of commitments - who has promised what to whom, what has been kept, what remains outstanding. Krippendorff's (2011) trajectory maps onto this gradient: at the utility level, states describe functional performance (system configuration); at the semantic and discursive levels, states describe the negotiated pattern of obligations and fulfilments between autonomous agents (commitment configuration). This is the shift that promise theory articulates: Burgess (2020) and Iqbal (2018) provide the vocabulary for understanding service states not as conditions of a system but as configurations of agency and commitment. The move from sense 1 to sense 2 is not merely a move from precise to loose; it is a move from a mechanistic to a relational conception of what "state" means.
Anchoring in design's own vocabulary
This four-level structure is not an import from AI into design; it maps onto distinctions design theory has been making for decades, though in different vocabularies. Simon (1996) defined design as "devising courses of action aimed at changing existing situations into preferred ones". This is, implicitly, a state-space description: an "existing situation" is an initial state; a "preferred one" is a goal state; "courses of action" are transition sequences. Simon's canonical definition does not use the term "state space", but the formal structure is there - and the series' contribution has been to make it explicit, showing that what design theorists describe philosophically (problem setting, frame construction) has a formal analogue in computational planning, and that this formal analogue makes specific, testable demands.
Boland and Collopy (2004) provide another useful anchor. Their distinction between the "decision attitude" and the "design attitude" maps directly onto the planning/design distinction: the decision attitude assumes the alternatives are given and the task is to choose among them - it assumes the state space. The design attitude recognises that the alternatives must be invented - it recognises that the state space must be constructed. The series' argument is a formalisation of their intuition: what Boland and Collopy describe as an attitude difference is, computationally, a difference in where the domain model comes from.
Krippendorff's (2011) trajectory of artificiality - from utility through semantics to discourse - describes movement through increasingly complex state spaces, and in doing so clarifies why "condition a system can be in" is insufficient as a definition of service-level state. At the utility level, the state space concerns functional performance: the states of a device or interface, where "condition a system can be in" is an adequate description. At the semantic level, it concerns meaning and interpretability: what a state signifies to its users. At the discourse level, it concerns the social negotiation of what the artefact is and does - and here, Burgess's and Iqbal's promise vocabulary becomes essential, because discourse-level states are not configurations of a system but configurations of commitments between agents, contested and negotiated rather than specified and implemented. The trajectory is one of increasing state-space complexity: more dimensions, more stakeholders, more contested definitions of what counts as a state or a goal. Krippendorff does not use state-space language, but his trajectory maps directly onto the distinction between senses 1 and 2 above - and explains why service design, operating at the semantic and discursive levels, requires a richer conception of state than computational formalism alone provides.
What holds across all four senses, and what does not
Having distinguished four senses, I want to test which of the claims I have been making actually travel across all of them.
The foundational claim - state spaces are constructed, not given - holds across all four. Whether we are talking about a formal domain model, a representational statechart, an institutional planning framework, or a governmental apparatus of legibility, someone has to construct it; it does not arrive ready-made from the domain itself. Scott's entire argument is that the categories the state uses to govern are invented, not discovered - that permanent surnames, cadastral surveys, and standardised measures are political projects, not natural descriptions.
The claim that design is logically prior to planning likewise holds at each level, though it means something different in practice. In the formal sense, the domain model must exist before the planner can operate. In the representational sense, the state-oriented model must be constructed before gaps and inconsistencies become visible. In the institutional sense, the problem must be understood before plans can be meaningful rather than merely procedural. In the governmental sense, the categories of eligibility and entitlement must be legislated before service planning can begin - you cannot design a vocational rehabilitation programme without a prior political determination of what "rehabilitation" means, who qualifies, and what counts as success.
The critique of service design's representational toolkit - that journey maps and blueprints leave states implicit - is primarily a claim about sense 2. It is about representational practice, not about what algorithms require, what institutions lack, or how governments categorise.
The institutional critique - that technomagic attempts to plan without a reflexive account of the problem domain - operates primarily in sense 3. It diagnoses an organisational pathology, and while the formal sense gives the diagnosis its force (the algorithm's requirements make the absence vivid), the claim itself is about institutional behaviour.
The governmental sense adds something the other three do not: the recognition that the categories themselves are political constructions. The first three senses all take the domain vocabulary as given - states, transitions, goals - and ask how well it is specified, represented, or understood. Sense 4 asks where that vocabulary came from and whose interests its construction served. This is the level at which the politics of formalism argument begins to bite hardest: not just that different formalisms encode different assumptions (which is a sense-1 observation), but that the variables available to any formalism are already shaped by governmental projects of legibility.
Where this becomes most productive is in tracing the connections across all four levels: governmental categories constraining what can appear as a variable in institutional planning, institutional planning shaping what representational models capture, representational models disciplining what formal specifications demand. The concept of "state space" runs through all of these - and the series' contribution, I think, is in making visible how construction at each level shapes and constrains the others.
References
Boland, R. and Collopy, F. (2004). Managing as Designing. Stanford University Press.
Burgess, M. (2020). A Treatise on Systems Volume 2. ChiTek-i.
Ghallab, M., Nau, D. and Traverso, P. (2004). Automated Planning: Theory and Practice. Morgan Kaufmann.
Hay, A.F. and Vink, J. (2023). The Emotional Neglect in Recent Service Design Developments. Design Studies, 89.
Iqbal, M. (2018). Thinking in Services: Encoding and Expressing Strategy Through Design. BIS Publishers.
Kimbell, L. (2017). The Turn to Service Design. In Julier, G. and Moor, A. (eds.) Design and Creativity: Policy, Management and Practice. Bloomsbury Academic.
Krippendorff, K. (2011). Principles of Design and a Trajectory of Artificiality. Journal of Product Innovation Management, 28(3), 411–418.
Mages, M.A. and Neely, S. (2023). Mapping Temporal Experience: Accounting for Felt Time in Service Design Representations. The Design Journal, 26(5), 750–770.
Russell, S. and Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson.
Scott, J.C. (2020). Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed. Yale University Press.
Simon, H. (1996). The Sciences of the Artificial (3rd ed.). MIT Press.
Vink, J., Hay, A.F. and Duan, Z. (2025). Reimagining Service Design through Relational Perspectives. Journal of Service Research.
Wetter-Edman, K., Sangiorgi, D., Edvardsson, B., Holmlid, S., Grönroos, C. and Mattelmäki, T. (2014). Design for Value Co-Creation: Exploring Synergies Between Design for Service and Service Logic. Service Science, 6(2), 106–121.