Theory — Modularity and Complexity

Housing has to keep changing after it is built. Occupants age, accessibility needs shift, regulations tighten, and the objectives that justify a renovation are themselves contested between the parties who must live with the result. Change of this kind is rarely a single decision; it arrives in feedback-rich episodes spread across years, where each adjustment alters the conditions under which the next one is judged. The governing question therefore concerns keeping adaptation tractable when the criteria are diachronic and disputed across repeated change. We treat that question as the chapter’s organising concern. The theoretical lens we build from modularity and complexity aims at the governability of ordinary change, leaving optimality of any single change aside.

The central claim is deliberately constrained. If representational boundaries, invariants, and transformations are made explicit and transportable, then adaptation can be governed by checking only the boundaries a change actually touches and escalating along declared dependencies, leaving the rest of the system untouched.¹² The claim promises less than it might seem to. It does not dissolve value conflict, eliminate coordination work, or optimise housing systems; those remain open after any representational improvement we could offer. Progress, on our reading, means something narrower and more durable — making ordinary change more intelligible, more local, and more auditable as it recurs. The wager is that intelligibility of this kind, accrued across many small adaptations, outweighs the optimality of any one.

Chapter 2 closed by deriving five meta-requirements, and each carries a precise governance function we inherit and do not re-argue here. Semantic continuity (MR1) requires meaning to survive handover; bounded composability (MR2) requires combination without unbounded side effects; formal expressibility (MR3) requires claims in checkable form; procedural traceability (MR4) requires operations with auditable records; and evaluation-readiness (MR5) requires propositions that declare their own falsifiers, as set out in Section 2.11. These five are settled. The work here is to operationalise them into mechanism commitments precise enough to constrain what follows.

Three outputs convert that standard into theory. The first is a constraint model: housing adaptation treated as operating under coupled wicked-problem and complex-adaptive-system conditions, so that any admissible intervention must stay governable amid contested objectives, delayed effects, and distributed agency.³⁴ The second is a mechanism model, and its operative move is Baldwin-Clark’s publish/hide partition, developed in Section 3.3.⁵ Simon’s near-decomposability supplies the systems intuition and Parnas’s information-hiding criterion supplies the decomposition discipline that make the partition workable.⁶⁷ Platform architecture then extends the same logic to system scale, holding a stable governance core fixed while permitting bounded complement variation — the arrangement that, we argue, lets one set of design rules govern many independent adaptations at once. The third output is a design-theory contract — constructs, principles of form and function, justificatory knowledge, and testable propositions, fixed early so that later chapters are judged against declared standards.⁸⁹

Ten terms form the continuity grammar that binds Chapter 4, the artefact chapters, and the evaluation chapter to one argument line. Six govern boundary and governance operations. An interface is a governed boundary condition for exchange. An invariant is an obligation that must remain true across a permitted change. A transformation is a named operation with declared preconditions and postconditions. A trigger is the condition that activates checking once a change has occurred. A design rule is a stable convention that coordinates otherwise independent evolution. An evidence object is the traceable empirical artefact by which a proposition is later assessed. Four further terms describe representational structure. A semantic primitive, or trigram, is the minimal three-element unit of meaning — source node, edge label, target node — at which individual identification, transport, and checking become possible.¹⁰ A representational plane is a concern layer that governs a distinct coupling profile and tends to evolve at its own rate, so that cross-plane effects may require explicit escalation in place of silent inheritance. A relational graph is the directed labelled graph whose community structure realises representational modularity. Governance-preserving linearisation serialises such a graph into text without loss of governance-relevant structure, a notational operation in its own right.¹¹

The chapter then proceeds through five modules, each contributing one structural layer. Section 3.2 fixes the constraint environment and its governance commitments; Section 3.3 builds modularity theory and its representational substrate — trigram, relational graph, governance-preserving serialisation, and four rival traditions absorbed as design constraints; Section 3.4 scales the logic into a stable-core platform arrangement with three-plane stratification; and Section 3.5 converts the mechanism model into five propositions with baselines, indicators, and falsifiers, before the method obligations Chapter 4 must operationalise are stated. Figure 3.1 maps the arc.

The anti-overlap rule stays strict throughout. We do not re-prove that housing adaptation is difficult, that accessibility needs change over time, or that institutional fragmentation matters; Chapter 2 settles those. We must show that the selected mechanisms jointly explain how explicit representation could restore bounded governability — precisely enough to survive method design and empirical testing.

Three conditions constrain how housing adaptation can be represented at all: the irreducibility of its objectives, the multi-stability of its outcomes, and its distributed agency. Each must be drawn out of the wicked and complex-adaptive analysis before any theoretical mechanism can hope to accommodate it. We develop the extraction in two stages and then join them. Wickedness yields a legitimacy constraint — adaptation proceeds under objectives that cannot be settled once and for all.¹² Complex-adaptive dynamics yield a controllability constraint — local action propagates in ways the actor cannot fully foresee.¹³ Where the two meet, at representation itself, they generate verification debt: the operative cost that links this constraint model to the modular boundary architecture developed in Section 3.3.

Wickedness as a legitimacy constraint

Wickedness matters here because housing adaptation admits no single stable objective function against which a design might be optimised and then declared finished. Rittel and Webber identify wicked problems by three marks: the absence of a definitive problem formulation, the lack of a clear stopping rule, and the way attempted solutions reshape the very conditions of the problem.¹⁴ Churchman’s earlier formulation sharpens the same intuition, treating wickedness as a planning condition in which narrow technical optimisation falls short of the social consequences that action sets in motion.¹⁵ Housing adaptation fits the profile closely. Adequacy depends at once on accessibility, affordability, household change, tenancy or ownership conditions, regulation, and service provision — and no one actor commands all of those criteria, nor does any final design state permanently dissolve their tension.

The legitimacy problem reaches representation directly. When objectives are plural and shifting, a representational system cannot lean on one decisive optimisation target; it must instead preserve enough semantic clarity that stakeholders can examine what is being changed, which obligations remain binding, and where the consequences of a local decision are likely to fall. Wickedness, on our reading, is therefore what obliges the thesis to seek a governance-oriented representational substrate rather than one more single-use design method. Head and Alford carry the argument further: wicked public-policy problems call for arrangements that sustain continuing judgement and adaptation rather than linear execution against fixed goals.¹⁶ A workable representational regime must accordingly support visible justification under disagreement, and not merely efficient information exchange under consensus.

Complex-adaptive dynamics as a controllability constraint

Complex-adaptive-system framing matters because even well-intended local action can trigger delayed and distributed effects beyond the actor’s field of view. Holland characterises complex adaptive systems as populations of interacting agents whose local rules generate emergent behaviour through adaptation, feedback, and recombination.¹⁷ Meadows complements that account by showing why interventions miscarry when feedback structure, information flow, and delay are poorly understood.¹⁸ Housing adaptation displays these dynamics in recognisable form. A change that appears local may reach into circulation, services, compliance pathways, documentation responsibilities, and later alteration rights — and the dependencies are unevenly visible, often surfacing only after a change has been proposed or partly executed.

Complexity becomes a controllability problem because actors work through partial models, role-specific documents, and time-bounded decisions. If the representation they inherit does not show where dependencies sit and which obligations survive a modification, a local actor cannot tell whether a change remains local. The rational response is defensive: actors widen the review scope, or refuse the change outright, because the residual uncertainty about cross-boundary consequence stays too high to absorb. Wickedness and complexity interact in a way that compounds both pressures, since there is no uncontested final state to optimise against once and no reliable local shortcut that may safely ignore feedback (Figure 3.2). Governance quality, then, tends to depend on how faithfully the system preserves and exposes the conditions under which a change can be justified.

Where wickedness and complexity intersect: representation

The point at which legitimacy and controllability meet is representation, for decisions can only ever be checked against what remains semantically and structurally visible. Gallaher and colleagues supply the empirical grounding: their analysis of interoperability failure in capital-facilities work shows that fragmented representations generate avoidable cost through re-entry, translation loss, and repeated checking.¹⁹ The mechanism transfers to housing with little loss: when meaning and obligation do not travel well across handovers, coordination cost rises and assurance work multiplies. This narrows the thesis to a single design question — whether the information regime may be reorganised so that ordinary change queries become answerable at declared boundaries, rather than only after a costly reconstruction of what an earlier representation failed to carry.

Verification debt as the mechanism bridge

Verification debt denotes the deferred, duplicated, or repeated assurance work that arises when fragmented representations force actors to reconstruct obligations, dependencies, and residual validity instead of reusing checks that have already been governed.²⁰ The term names a pattern Chapter 2 repeatedly exposed without formalising: once identity, the meaning of an obligation, and the conditions of a transformation drift into tacit knowledge, each new intervention must rediscover what an earlier one should have preserved. The debt is paid through manual reconstruction, expanded review, or conservative refusal to act.

One distinction does the load-bearing work, and it separates verification debt from Cunningham’s well-known analogy. Technical debt is characteristically elective — a shortcut chosen knowingly under time pressure, taken on with the expectation of later repayment.²¹ Verification debt in housing representation is, by contrast, primarily structural: it issues from the information architecture itself rather than from any actor’s deliberate choice. When a representation does not carry identity, obligation, and transformation logic in explicit form, no actor can escape the burden whatever their diligence — and that single fact locates the remediation target in representational design, not in professional practice. The shift matters because it moves the problem from the conduct of individuals, where it cannot be solved, to the structure of the representation, where it may be.

How verification debt compounds under feedback

Verification debt does not accumulate linearly, and two literatures explain why. From information-quality research, Redman shows that uncorrected quality failures propagate forward through each subsequent decision, compounding cost across the enterprise,²² while Wang and Strong extend the account into a multi-dimensional framework whose representational and accessibility dimensions map onto the governance requirements developed here.²³ From coordination-cost economics, Arrow establishes that every information exchange carries irreducible cost, which makes trust and verification functional substitutes in any governance system,²⁴ and Milgrom and Roberts formalise the escalation: when system elements are complementary, coordination cost rises superadditively with the depth of interdependency.²⁵ Set these results under CAS feedback and the compounding mechanism becomes legible: a missing identity reference in one handover forces reconstruction in the next; that reconstruction introduces interpretive divergence; and every later handover must then accommodate the divergence in turn — the same propagation Holland’s account identifies for system behaviour, now turned on assurance work itself.²⁶

Four governance commitments grounded in the constraint profile

The constraint profile yields four governance commitments, and these are the modularity-theoretic operationalisation of the four critical properties Chapter 2 established as the design targets a modular boundary architecture must make governable — interoperability, transportability, manipulability, and transformability.²⁷ Those four properties correspond to the meta-requirements derived in Chapter 2: interoperability to MR1, transportability to MR2, manipulability to MR3, and transformability to MR4, respectively. The wicked and complex-adaptive analysis does not generate a different set of requirements; what it establishes is why those requirements are non-negotiable rather than merely desirable. Under contested objectives, delayed effects, and distributed agency, a representation lacking any one of the four cannot support bounded re-checking and must fall back on whole-system reconstruction.

Stable referential identity. A revision-governed system must preserve what its objects are called and how they are traced across versions, because comparison becomes impossible when referents drift silently between representations. This commitment operationalises interoperability and its semantic-continuity requirement, MR1, in its most basic form.

Explicit invariants and degrees of freedom. A system cannot govern change if it does not distinguish what may vary from what must remain true. This commitment operationalises manipulability and its formal-expressibility requirement, MR3, by making the space of permissible variation an object the representation declares rather than leaves implicit.

Triggered interface checking. Meadows’ emphasis on feedback visibility implies that checking should follow declared changes along declared dependency paths, rather than defaulting either to blind trust or to whole-system revalidation. This commitment operationalises procedural traceability, MR4, so that every permitted change produces a bounded, re-triggerable assurance obligation rather than an open-ended one.

Serialised transformation logic. If adaptation is to stay auditable over time, permitted operations must be named and represented in a form that preserves preconditions, postconditions, and responsibility for execution. This commitment operationalises transportability and its requirement, MR2: by encoding operations in a medium that survives handover without bespoke reconstruction, it keeps verification chains intact across actor rotations and tool changes. The same constraint profile, applied not to one module’s boundary but to a population of complements co-varying under a shared core, recurs at platform scale in Section 3.4; its evaluation-readiness counterpart, MR5, is taken up through the design-theory framework in Section 3.5.

Boundary: the framing justifies requirements, not full-system modelling

We use wickedness and complexity to justify representational requirements, not to claim that the thesis models the housing system in its entirety. A positive result in the later chapters would not show that housing wickedness has been solved or that complexity has been neutralised.²⁸ It would support the narrower claim that explicit representational governance can make adaptation more tractable — reducing avoidable verification debt and rendering local change more auditable than it is under fragmented, tacit, or tool-bound representations.²⁹ That narrower claim is the mechanism horizon Section 3.3 takes up, translating this constraint analysis into modular boundary logic.

Modularity is the mechanism that brings interdependence under governance, making the structure of dependence operable so a change can be checked at the boundaries it touches. Section 2.8 introduced it as a general principle of bounded interdependence; here we narrow that principle to the one form the representational-governance problem actually demands, which is interface-governed coupling control. A representation is modular, on our reading, only when it permits bounded change by exposing three things at once: the conditions under which change may occur, the obligations that must remain stable while it occurs, and the checks that must follow once a boundary has been touched.³⁰³¹³² The narrowing matters because many housing representations look decomposed while behaving monolithically. Drawings may be layered, models segmented, processes distributed across roles, and yet a modest change still triggers broad reinterpretation, because the interface obligations remain tacit, tool-bound, or inconsistently named. A modularity claim is earned only when post-change verification can be organised through declared boundaries a later actor can read off the representation itself.

That standard is not asserted; it is the invariant that survives systematic comparison across five bodies of evidence. The adopted definition rests on a structured cross-domain corpus of 905 screened records, of which 129 were retained for extraction and 810 evidence rows captured, spanning biology, computer science, engineering, management, and interdisciplinary synthesis over the period 1981 to 2025.³³ Three findings emerge. Definitional variance follows domain logic, each domain performing a different analytical operation on one phenomenon: engineering treats modularity as deliberate partition, biology as discovered covariation, management as governance architecture, network science as detectable community structure.³⁴³⁵ Despite the vocabulary divergence the extracted definitions converge on a single invariant: a candidate module is a cluster whose internal relation pattern is stronger or denser than its external pattern under a declared frame. And definitions and operationalisations, though connected, remain distinct, since a clustering index may detect grouping while leaving interface semantics ungoverned, so definition-measurement alignment is non-negotiable for a thesis whose problem is verification burden under repeated change.³⁶

The convergence yields five non-negotiable properties that any thesis-grade definition must satisfy to do governance work rather than merely descriptive work.

1. Explicit frame-of-analysis dependence. A boundary valid for procurement may not be valid for accessibility verification, so the frame within which a boundary is meaningful must always be declared.
2. Interaction differential as the central organising concept. Internal interactions must be stronger, more frequent, or more governing than external interactions within the declared frame; without that asymmetry there is no principled reason to treat a cluster as a module rather than an arbitrary partition.³⁷
3. Functional-boundary status rather than visual segmentation. A wall line is not a module boundary unless the representation declares what crosses it, what is preserved behind it, and what must be checked once it is altered.³⁸
4. Finite and expressed interface conditions. Exchange across the boundary must be governed by an enumerable, declared condition set whose every condition is stated explicitly.
5. Demonstrably reduced coordination requirements under system-role preservation. Because modularisation can shift rather than eliminate complexity, reduced coordination is interpreted strictly as bounded verification scope after local change.³⁹

The five properties exclude three transfer failure modes that recur across the corpus. Definitional inflation names parts and declares them modular without finite interface conditions. Method absolutism transports a metric across units of analysis where it no longer holds, so a strong graph-partition score guarantees nothing about whether housing obligations are explicit and re-checkable.⁴⁰ Scale confusion leaves unit, interaction type, and time horizon unspecified, which is most consequential under diachronic adaptation, where a boundary adequate at initial design may fail as the conditions of exchange shift across a dwelling’s lifespan.

From that convergence one canonical definition follows, and it governs all subsequent usage: a module is a cluster of system elements whose internal interactions are, within a stated frame of analysis, substantially stronger or more frequent than their external interactions; this interaction differential facilitates a functional boundary governing every intended exchange via an interface of finite and expressed conditions, thereby enabling the cluster’s interaction with the external environment under reduced coordination requirements while sustaining its role within the larger whole. Two commitments follow and bind the remainder of the thesis. Modularity is defined by interaction structure plus boundary governance, never by prefabrication, segmentation, or naming; a representation that labels parts as modules but declares no finite interface conditions is pseudo-modular under thesis rules. And the reduced-coordination claim is testable: it stands or falls on post-change verification behaviour, which architecture diagrams alone cannot carry.

The operative move. What converts this definition from a structural intuition into procedure-bearing control logic is a single distinction we transpose from Baldwin and Clark: between visible design rules, which coordinate independent work, and hidden module parameters, which can vary locally without destabilising the system.⁴¹ Housing adaptation cannot rely on tacit modularity, because the actors who design, approve, procure, verify, and inhabit change share no stable interpretive setting; what is hidden for one is load-bearing for another. The published surface must therefore take an explicit form, which we call an interface contract, with three components: declared degrees of freedom within which a unit may vary, invariants that must remain true when variation occurs, and verification triggers specifying when a change reactivates checking. Two received traditions supply the mechanics this move needs. Parnas’s information-hiding tells us which decisions to hide — the volatile ones, placed behind stable interfaces — and warns that where the conditions on which checking depends are themselves hidden, hidden coupling simply reappears as downstream uncertainty.⁴² Simon’s near-decomposability tells us how checking should propagate: local change checked locally first, escalated through declared dependencies only where the interface conditions require it.⁴³ Both are subordinate; the publish/hide partition is the move.

A modularity claim is accepted, then, only when boundary quality can be stated as an auditable rule set. Ten axioms synthesise Simon, Parnas, and Baldwin and Clark into such a framework.⁴⁴⁴⁵⁴⁶

1. Frame-boundedness — every module claim declares the frame within which its boundary is meaningful.
2. Interaction differential — internal interactions exceed external ones within that frame.
3. Boundary observability — the system can locate the boundary and identify what crosses it.
4. Finite interface conditions — exchange is governed by an expressible, finite condition set.
5. Protected internal variation — internal change is admissible where it does not violate the published interface.
6. Declared invariants — the boundary states what remains true across permitted change.
7. Declared degrees of freedom — the boundary states what may vary without escalating beyond the module.
8. Triggered verification — boundary conditions include the events or states that reactivate checking.
9. Revision-stable role identity — the module’s role stays comparable across versions unless redesign declares otherwise.
10. System-integrability — the module remains assessable within the larger whole through explicit external dependencies.

These axioms reject the weak claim that a module is anything an analyst can draw a line around; a module is something that can carry explicit responsibility for what is hidden, preserved, varied, and checked. From them follow six derived property groups carried into proposition testing — containment, identity and cohesion, boundary governance, scaling, adaptation capacity, and analysability. The groups map directly onto the manageable-change conditions derived in Chapter 2 — containment, manageable interface load, limited coordination span, incrementality, reversibility, and absorptive capacity — read now as boundary-quality properties a representational regime must realise, the same conditions reframed as obligations on the substrate rather than diagnostic criteria for adaptable housing.

Boundary quality is then assessed through three linked interface types. The semantic interface governs meaning: the terms, object identities, and requirement references that must stay interpretable across the boundary. The transformation interface governs admissible change: operation types, allowable parameter movement, and the invariant-preservation conditions under which local variation stays legitimate. The verification interface governs re-checking: it links a change to the evidence, tests, or review procedures that must be reactivated once a boundary condition is touched. The three are analytically distinct but operationally inseparable — a transformation cannot be assessed if the affected object’s semantic identity is unstable, and a verification trigger is indiscriminate if the transformation conditions are unspecified — so we treat them as one contract surface distributed across three governance functions. Three failure modes accordingly make a modularity claim rejectable: interface-as-label, where boundaries are named but degrees of freedom, invariants, or triggers are missing; hidden-coupling persistence, where obligations stay embedded in tacit expertise or tool defaults; and boundary instability, where the interface’s meaning drifts across revisions even as its vocabulary holds.

Rival traditions as design constraints

Four alternative theoretical traditions are declined as the primary framework, yet each raises a genuine challenge and each yields a positive design constraint the artefact chapters must satisfy. The epistemological warrant for engaging them selectively is pragmatist: design research needs a framework that generates design decisions evaluable against their governance consequences.⁴⁷

Actor-Network Theory holds that boundaries are performed through ongoing network enrolment, sustained as achievements of that process.⁴⁸ Reading specification as itself an enrolment of future actors, it yields DC-1 Negotiability — the artefact must expose its degrees of freedom explicitly, alongside its invariants, so actors can find where negotiation is possible without destabilising the governed core. Institutional Theory argues that adoption turns on institutional fit — regulatory, normative, mimetic — as much as technical merit.⁴⁹ It yields DC-2 Institutional compatibility — the artefact must operate within existing arrangements, documenting any departure explicitly. Henderson and Clark show that modular systems remain vulnerable to architectural innovation, reconfiguration that leaves components intact while rendering interface knowledge obsolete and freezing the system in configurations optimised for a prior arrangement.⁵⁰⁵¹ That risk yields DC-3 Revisability — a minimal governed kernel, with interface contracts carrying explicit version semantics so reconfiguration need not mean wholesale replacement. Ontology engineering offers richer formalisms whose expressive power trades against tractability, a tension Levesque and Brachman make precise: each gain in expressiveness shrinks the class of tractably inferable problems.⁵²⁵³ Deliberately simple primitives can carry governance functions that richer formalisms make intractable in practice.⁵⁴ The tradition therefore yields DC-4 Practitioner tractability — compositional complexity built from simple primitives, computable by housing practitioners without specialised inference engines. The four traditions constrain how the modularity framework must be built, and DC-1 through DC-4 are carried forward as evaluation criteria in Chapter 10.

Axiom 4 is the hinge into the substrate. If exchange across a boundary must be governed by a finite, expressed condition set, then the unit of meaning cannot be a full discourse, a drawing, or tacit professional knowledge, for in each of those the condition set is unbounded. Finitude implies atomicity. The minimal sufficient unit is a semantic primitive: the smallest structure that carries relational intent explicitly enough for verification without tacit supplement, stably enough for cross-version comparison, and independently enough for tool-to-tool transport — three conditions following directly from Axiom 4.⁵⁵

What structural form satisfies all three at once? Five independent traditions converge on three elements as both necessary and sufficient for an atomic unit of relational governance. The Peircean Reduction Thesis establishes formal necessity within its domain: genuine triadic relations are irreducible, and Burch’s algebraic proof supplies that necessity direction.⁵⁶⁵⁷ A database-theoretic complement then splits the remaining warrant cleanly: Dechter’s decomposition of any n-ary relation into a tree of binary relations establishes that trigram-level decomposition is sufficient, since higher-arity relations need not be treated as atomic, while Jones and Song’s ternary-irreducibility result establishes that binary reduction loses governance-relevant context — the necessity direction.⁵⁸⁵⁹ The graph-theoretic argument is direct: a directed labelled edge is the minimum encoding a named source, a typed relational claim, and a named target. Relational cognition supplies tractability evidence, ternary processing being the cognitive minimum for systematic inference, with expert schema-formation departing from the unpractised-adult baseline rather than contradicting it.⁶⁰ Simmel’s structural-sociological analogy notes that a third element creates a supra-personal structure whose persistence is independent of either party’s endorsement.⁶¹ And the cross-linguistic subject-verb-object structure encodes the same tripartite organisation across unrelated languages.⁶² The five traditions carry different logical force, so the convergence is a convergent-validity argument: agreement across independent warrants, leaving exhaustive enumeration aside.

This three-element directed structure is the trigram. The source node carries referential identity, naming the governed object in a form that persists across representation changes; the edge label carries relational commitment, stating the relational direction explicitly so it need not be inferred from spatial adjacency; the target node carries constraint scope, the value, bound, or condition that checking can target. The trigram satisfies Axiom 4 structurally: its source-label pair gives it finite referential scope, its explicit direction allowing comparison without reconstructing authorial intent.⁶³ It also binds the three interface types into one operable unit — what flows through semantic interfaces, what identity persistence preserves, what transformation logic operates on, and what verification triggers check.⁶⁴ Semantic primitives thus constitute the substance of the semantic interface itself. The collection of all trigrams over a governed domain forms a relational graph whose topology inherits the modularity properties: named nodes preserve referential identity regardless of tool, typed edges declare the relational claim, and directionality preserves asymmetry. Simon’s near-decomposability has a direct graph-topological realisation here — modules are subgraphs whose internal edge density substantially exceeds cross-boundary density, the community structure whose finite crossing-edge set is precisely the Axiom-4 interface condition set.⁶⁵⁶⁶ Modularity becomes observable at graph level: a drawing must be interpreted before it can be tested, while a trigram graph can be assessed by edge-density asymmetry directly. Graph operations correspond to housing adaptation operations, so changes respecting community boundaries trigger bounded checking, changes crossing them trigger declared escalation, and violations are detectable because the violated conditions are explicit edges.

Text as governance-preserving notation

A graph without a transport medium cannot travel across tools, actors, or time, and so the substrate question closes on serialisation. Text is a notational system whose structural properties constitute the graph as a governable object; it does more than record one governed elsewhere. Haugeland establishes the enabling fact: a digital device identifies tokens positively and reliably, succeeding absolutely and consistently.⁶⁷⁶⁸ Goodman grounds it structurally: a notation requires syntactic disjointness and finite differentiation, both of which text satisfies, whereas drawings are syntactically and semantically dense — “the very opposite of a notation.”⁶⁹ Lagueux shows the autographic/allographic difficulty to be intrinsic to geometric representation, and Morris and Spanoudakis confirm that Goodman’s criteria carry over to formal technical notation, validating the move from art theory to engineering.⁷⁰⁷¹ How the grammar layer secures that disjointness in practice is addressed in Section 7.3.

The governance properties text provides fall into three distinct layers, and collapsing them into a single “text is good” claim would obscure which property the medium supplies and which the grammar must. Text-native properties — discreteness, composability, stability and copyability, human readability — follow from the medium’s constitution independent of any grammar. Tooling-enabled properties — diff-ability, versionability via Myers’s O(ND) algorithm and version-control infrastructure, searchability — exploit native discreteness through added infrastructure.⁷²⁷³ Grammar-imposed properties — parsability, semantic determinacy, round-trip fidelity — require a formal grammar layered over the medium. Text’s native discreteness and composability are what allow it to host a governance-capable serialisation grammar at all; the grammar then completes what those native properties begin. Following Floridi, the layers form a disjoint gradient of abstraction, each making different observables available under one overarching constraint.⁷⁴

This serialisation differs from machine-learning dimensionality reduction, which is lossy by design: it is structurally lossless at the governance level — every node identity, edge label, direction, and community membership is represented — discarding only the spatial layout, never governance-relevant. Shannon’s source coding theorem frames the point exactly: a source can be re-encoded in a different alphabet at any rate approaching its entropy without information loss, and Simon’s description-compression argument adds that nearly decomposable systems admit shorter descriptions by exploiting modular redundancy.⁷⁵ The Larkin-Simon spatial-reasoning trade-off is accepted with open eyes: diagrams can be dramatically more efficient for perception, while governance requires deterministic verification, and “Room A is adjacent to Room B” is deterministically checkable from text in a way a drawing’s scale-dependent adjacency is not.⁷⁶

The substrate is now settled: the trigram as atom, the relational graph as topology, text as transport. Near-decomposability recurs at each — as the interaction differential at the governance scale, as community structure at the graph scale, as hierarchical compositionality at the text scale — one organising principle observed under three abstractions. With the substance now fixed, what remains is a question of scale. A single well-governed boundary does not yet tell us how governance holds across a system of many modules, nested and recombined, where one unit’s interface is another’s dependency. That is the problem the platform architecture in Section 3.4 must address, scaling interface-governed control beyond the single module boundary.

Repeated adaptation cannot be governed through isolated case logic. If each adaptation event is handled as bespoke reconstruction, every new requirement rebuilds the same interpretive and verificatory foundation from the ground up, and the cost recurs in full at each turn — verification debt, now incurred at system scale rather than within a single case. A platform arrangement resolves this by identifying the persistent terms of exchange on which later variation depends. Three lines of work converge on the same lesson. Baldwin and Clark treat visible design rules as the stable surface that makes distributed experimentation possible without systemic collapse.⁷⁷ Gawer and Cusumano locate the same arrangement at industry level: platforms organise innovation by stabilising a common base while permitting complementary variation around it.⁷⁸ Tiwana, Konsynski, and Bush sharpen the point by insisting that platform evolution is governed by the coevolution of architecture, governance, and environmental dynamics rather than by architecture alone.⁷⁹ Taken together, these establish that a stable core is necessary but never sufficient; it earns its keep only when paired with rule structures that hold both variation and checking within bounds.

The strongest housing precedent for this logic is the support-infill split developed in Open Building. Habraken’s separation of a long-life shared structure from short-life occupant-specific variation lets later change occur within a prepared field, and so reduces the scope that must be renegotiated whenever needs shift.⁸⁰ The realised record qualifies the promise. Brand, surveying how building layers move at different rates, finds that embedding faster-changing services into slower-changing structure “may look efficient at first, but over time it is the opposite, and destructive as well.”⁸¹ Kendall and Teicher, documenting built Dutch supports, record that several purpose-built structures “found no economically viable fit-out systems to install within them” — physical layer separation, on this evidence, is necessary but not sufficient, since a working infill market with viable interface specifications must accompany the structural provision.⁸² We adopt the precedent while shifting its plane of action. Open Building separates physical layers and there encounters an unresolved problem of interface specification between them; the present argument operates instead at the governance and information layer, treating the modularity of the representation system — the means through which adaptation is described, verified, and governed — rather than of the building fabric. On that footing the interface-specification problem can be addressed directly, by proposing explicit, governed interface contracts between representational layers as the mechanism through which layer separation acquires operational force.

The governed kernel and its instances. The kernel developed here is representational rather than geometric, and the contrast with Open Building’s physical-fabric core is deliberate. It comprises the persistent terms required for cross-time comparability and bounded verification: stable referential identity, shared semantic anchors for accessibility requirements, a controlled interface vocabulary expressed as design-by-contract obligations, named transformation types, and provenance rules for recording evidence and responsibility.⁸³⁸⁴ These terms are few, and they are meant to stay few. The governed instances, by contrast, are the context-specific module arrangements, case instantiations, transformation sequences, notation expressions, and procedural outputs that vary from one adaptation episode to the next; a governed instance is a local elaboration whose admissibility depends on the kernel rules remaining intact. The arrangement reproduces, at representational scale, the publish/hide partition that organises modular design: the design rules are published and held fixed for any given configuration, while module-internal parameters stay hidden and free to vary. Crucially, the kernel is governed, not frozen: its stability is disciplined revisability — rule-bound, evidence-based, versioned, and auditable — not immobility, and it is precisely this that distinguishes it from the two opposite failures the arrangement must guard against. A frozen core in the pejorative sense expands until all variation is forced back onto one rigid control surface, trading evolvability for uniformity; this is the failure of immobility, and the governed kernel avoids it by carrying an explicit, version-aware revision path rather than forbidding revision outright. ungoverned complements proceed without rule-bounded compatibility, so that hidden coupling accumulates and verification re-globalises across the whole. A viable platform steers between the two — the kernel kept minimal, authoritative, and governed rather than merely frozen, the instances kept variable but rule-bound.

A housing representation qualifies as a platform only when five requirements are met. They follow directly from the four governance commitments of the constraint model in Section 3.2, rather than being added on top of them.

1. Stable identity. Objects, requirements, and versioned states remain referentially traceable across handovers — the platform-scale form of the constraint model’s stable referential identity.
2. Explicit interface publication. The system states what passes between modules, what may change locally, and what stays invariant; this is the constraint model’s explicit invariants and degrees of freedom, lifted to system scale.
3. Stratified concern management. Physical, semantic, procedural, and evidentiary relations remain distinguishable without being disconnected, so that changes escalate by rule rather than by guess — the scaled form of triggered interface checking.
4. Transportable semantics. Meaning survives movement across tools and actors rather than staying embedded in one software environment or one specialist’s tacit knowledge, which is serialised transformation logic carried to platform scale.
5. Governed provenance. The system preserves who changed what, under which rule, and with what verification consequences — the platform-scale form of the change-record obligation that the four commitments jointly imply.⁸⁵⁸⁶

Of the five, stratified concern management names a requirement without yet supplying a mechanism, and the mechanism turns out to be Simon’s near-decomposability applied reflexively to the representation’s own architecture. The principle, established for component boundaries in Section 3.3, is here turned back on the descriptive scheme itself: if concern layers can be partitioned so that within-layer coupling exceeds cross-layer coupling, each layer may evolve semi-independently, changes propagating quickly among elements that share a coupling profile and slowly across the declared transitions between them. Applying a descriptive regularity as a normative design rule asks for justification, and design-science epistemology supplies it. Simon observes that in the artificial sciences “the descriptive and the normative are never far apart,” so a structural regularity may legitimately become a design rule where it fixes the conditions under which an artefact achieves its purpose.⁸⁷⁸⁸ Hatchuel and Weil’s C-K theory gives the move its formal shape: near-decomposability sits in the knowledge space as a descriptive proposition, the K-to-C operator transforms it into a concept-space commitment, and that commitment is grounded in knowledge yet evaluated by its design consequences rather than logically entailed.⁸⁹ The three-plane architecture set out below is precisely such a commitment — stipulated as a design decision, and evaluated by whether the resulting system satisfies the four domain-derived meta-requirements (Meta-Requirement 1 through Meta-Requirement 4), not by deduction from a complexity-science law.⁹⁰

The construal gradient and its cognitive grounding. The three planes are not only a near-decomposability commitment; they instantiate a gradient that cognitive linguistics establishes independently, and the grounding matters because the vocabulary populating the planes in Chapter 6 inherits its structure from it. Cognitive grammar treats meaning as conceptualisation, organised by a baseline-to-elaboration asymmetry: highly schematic categories — space, boundary, objecthood, quality, event — act as cognitive baselines from which more specific concepts are construed by profiling, the foregrounding of a particular facet against a schematic base.⁹¹⁹² A built-environment concept is on this account a construal rather than a bare object: a room is not a box but bounded space profiled as enclosed, functionally identified, and habitable, and a barrier is not a thing but a boundary profiled as obstruction. Developmental work on image schemas supplies the same layering from the bottom up, distinguishing spatial primitives such as path, contact, and boundary from the image schemas built on them — CONTAINER, SUPPORT, PART-WHOLE, SOURCE-PATH-GOAL — and from the schematic integrations that bind spatial form to force and function.⁹³⁹⁴ The CONTAINER schema elaborates into room, SUPPORT and PART-WHOLE into fixture, and SOURCE-PATH-GOAL into path — the same derivations the artefact vocabulary records. Talmy’s force dynamics explains why a passive boundary becomes an active barrier once a blocking force is profiled, and prototype and radial-category theory explains why such elaborated concepts admit graded membership and systematic polysemy rather than definition by necessary and sufficient conditions.⁹⁵⁹⁶⁹⁷

Frame semantics completes the picture and licenses the upper planes directly. Fillmore’s insight that a word’s meaning is fixed by the structured background it evokes — so that context is not metadata applied after interpretation but a meaning-generating frame, and a role exists only as a participant’s place within such a frame — is what warrants treating applicability scope as a plane in its own right rather than as annotation on the others.⁹⁸ Read through this lens the three planes are a construal gradient: the primitive plane carries the schematic baselines and their most direct spatial elaborations, the configurative plane carries the identity and intent that profiling assigns, and the interactive plane carries the relational and framing operations — the construal work itself. Modules then fall out as the gradient’s top layer in the form frame semantics and the interaction differential jointly predict: an interactional gestalt is a cluster of elaborated concepts whose internal interactions are denser than their external ones, bounded by finite interface conditions — Axiom 4’s condition set restated in cognitive terms. This grounding fixes how the vocabulary populating the planes is structured (schematic primitive versus elaborated composite) without displacing the near-decomposability derivation of the planes themselves; the vocabulary, composition, and interface contracts that operationalise it are built in Chapter 6, Section 6.1, and the supporting literature is reviewed in Chapter 2.

The primitive plane. This is the stable metric substrate: discrete geometric primitives, dimensional commitments, coordinate frames, and measurement reference points. Within-plane coupling is strong, since all metric operations share one dimensional ground; cross-plane coupling is thin by design, the plane exposing only what others need for metric grounding while withholding its interior logic. Change here is slow by governance discipline, because each change re-establishes the metric ground for every layer that references it.

The configurative plane. This plane binds identity and intent to assemblies of primitives — assigning functional roles, recording accessibility requirements against named objects, and stating what an assembly is for rather than merely where it is and how large. The semantic primitives developed in Section 3.3 — trigrams carrying referential identity, relational commitment, and constraint scope — are the atomic units through which this plane expresses its semantic content at interfaces. It evolves at a medium rate, more readily than the metric substrate but with more governance overhead than routine operational events.

The interactive plane. This plane carries context and state: as-built deviations, approval records, permitted-change history, and scheduled adaptation events. Each event updates the state record, so change here is frequent — yet bounded rather than free, since every change activates a defined re-checking procedure confirming that primitive- and configurative-plane obligations remain intact, and may not alter lower-plane obligations except through a cross-plane escalation path that meets the higher overhead those planes demand.

So partitioned, the three planes convert geometry-semantics conflation from a probable consequence of monolithic representation into a detectable governance violation. A change to a dimension is checked against primitive-plane invariants and does not, of itself, trigger role re-assignment; a change to a functional role is checked against configurative obligations and does not, of itself, require re-measuring the metric substrate. Cross-plane propagation requires activating an explicit escalation interface, which turns hidden coupling from a default condition into an exceptional one that must be justified before it is admitted.⁹⁹

The plane structure finds a suggestive functional parallel in spatial cognition, which we note as supplementary justificatory material rather than an independent derivation path. Research reviewed in Chapter 2, Section 2.7 distinguishes grid cells, which supply a stable metric substrate firing independently of semantic content, from place cells, which bind functional meaning to spatial position — a separation that mirrors the primitive-configurative distinction. Under the Gregor-Jones framework the convergence offers corroborating knowledge only; it does not alter the model’s theoretical derivation, scope, or evaluation criteria, which rest on the near-decomposability analysis and the governance requirements developed across Sections 3.2-3.4.

This much grounds platform-governed complement evolution as Proposition 4 — a stable shared core paired with rule-governed complement extension. Chapter 3 states Proposition 4 as a proposition and stops there: how the stable core and the governed instance library are actually built and governed is an operational question this chapter deliberately does not answer, deferring it to Chapter 6, Section 6.4, with the variant evidence borne out in Chapter 8 and the compatibility evidence in Chapter 9. Three failure modes mark how the platform claim can fail under evaluation. A frozen core makes variation impossible, every case-specific issue being absorbed into the supposedly stable base. interface drift leaves the nominal rule surface in place while its meaning shifts across actors, revisions, or tools, so that comparability decays beneath apparent consistency. ungoverned complements bypass the shared rules and force later actors back into full-scope reinterpretation. A positive result, conversely, is evidence that meaning travels more reliably across handovers, that local variation stays compatible with shared rules, and that post-change verification can be bounded without denying legitimate adaptation.¹⁰⁰¹⁰¹ The artefact suite maps onto this argument and so makes it testable: Chapter 5 serialises regulatory meaning into stable semantic anchors, Chapter 6 publishes module boundaries as explicit interaction contracts, Chapter 7 encodes transformation and invariant logic, Chapter 9 produces governed local variation, and Chapter 10 tests whether the arrangement reduces reconstruction burden while preserving verification reliability.

Design Science Research accepts a contribution only when its artefact class is specified so that it can be criticised, operationalised, and tested; after-the-fact narration fails this requirement. The design-theory anatomy of Gregor and Jones supplies that form, because it obliges the theory to name its constructs, principles of form and function, justificatory knowledge, boundary conditions, and testable propositions within one coherent structure.¹⁰²¹⁰³ We keep two positioning claims on separate axes, because they answer different questions and might mislead if conflated. On Gregor and Hevner’s contribution-type taxonomy the work is primarily exaptive: modularity theory, relational graph topology, and text as a notational medium are each established in their home domains, and the integrated design theory for housing-governance representation is the contribution we develop.¹⁰⁴ At the theory-type level the work advances a Type V (Design and Action) theory in Gregor’s taxonomy — the one type whose primary obligation is prescriptive rather than descriptive.¹⁰⁵ The exaptive label names the kind of knowledge produced; the Type V designation names the genre in which it is cast. Both axes are necessary, and conflating them would obscure what the two designations jointly establish. Completeness of the Type V claim is itself falsifiable, and it appears to depend on the Chapter 5 to Chapter 9 artefact and evaluation programme rather than on anything settled here.

We use the anatomy as an executable contract, and we name the design-theoretic position it specifies Stratified Functional Structuralism (SFS). The name records three commitments the chapter has already made rather than announcing a new one: a governed representation is treated as a structure — the modular, relationally typed graph of mechanism objects developed in Section 3.3; that structure is stratified into the governed planes of the platform architecture of Section 3.4; and each structural commitment is admitted only for the governance function it discharges, through the principles of form and function set out below. We use “structuralism” in its architectural sense, naming the primacy of relational structure over surface form, with no allegiance to the philosophical movement of the same name. The contract’s constructs and propositions are fixed in advance and bind what follows. Purpose and scope are restricted to repeated housing adaptation under multi-actor, multi-representation conditions, where verification burden materially affects what changes are attempted or abandoned. Seven mechanism objects, fixed earlier in this chapter, serve as the constructs — object identity, interface, invariant, transformation, trigger, design rule, and evidence object. Justificatory knowledge is drawn from wicked problem theory, complex adaptive system theory, modularity theory, platform architecture, and the interoperability and governance evidence base.^{106107108109110} Figure 3.6 maps these components onto the Gregor and Jones anatomy, showing how constructs, justificatory knowledge, and principles of form and function compose into a single scaffold.

The principles of form and function are the architectural commitments developed across Section 3.2 to Section 3.4, each naming both a representational form and the governance function that form enables.

The five Chapter 2 meta-requirements and the five propositions developed here are not in one-to-one correspondence. Each proposition addresses one meta-requirement primarily but also supports adjacent ones, and that adjacency is part of the theory’s integrity rather than incidental overlap. The number of propositions requires its own justification. Proposition 1 through Proposition 4 address governance functions the mechanism analysis identifies as irreducible: semantic identity persistence, boundary-bounded verification, executable transformation, and governed complement evolution. These four cannot be merged without losing evaluative precision, since a representation may achieve identity persistence without bounded verification, or formal grammar without governed variation. Proposition 5 exists for a different reason: the platform argument of Section 3.4 establishes that the integrated artefact suite must be evaluated as a system — an obligation no single meta-requirement generates alone. The matrix below uses P for the primary meta-requirement and s for those a proposition also supports.

Each proposition is then defined by a mechanism, an indicator, a comparator baseline, a body of primary evidence, and a practical falsifier whose triggering carries a diagnostic implication — when a falsifier fires it names which mechanism failed and what re-design the failure implies.

Proposition 1 — semantic continuity as a measurable governance effect. Proposition 1 treats continuity of meaning as a property of the representation, claiming a material reduction in interpretive divergence relative to case-by-case reassembly rather than universal agreement or zero ambiguity. It operates at the trigram granularity fixed in Section 3.3. Identity persistence at system scale is carried by the relational graph’s named nodes, whose referential stability, on our reading, survives linearisation into text because the node identifiers are preserved in the serialised form.¹¹¹

Proposition 2 — modularity as bounded verification scope. Proposition 2 claims that declared interfaces may convert broad reconstruction into local checking plus rule-governed escalation. The deciding evidence is what a change trace supplies: that a local edit triggers a finite, explainable set of checks confined to the affected plane and its declared escalation paths. The mechanism that prevents cross-concern hidden coupling is the three-plane stratification of Section 3.4, under which a bounded edit at one plane triggers checking within that plane and along declared cross-plane paths alone.

Proposition 3 — formal expressibility made executable. A formal language earns its place only when it expresses named transformations, preserved invariants, and admissibility conditions in a form that may be inspected and replayed rather than read as static notation. Replay consistency is assessed at the trigram level (Section 3.3) and against the plane-specific invariants of the stratified architecture (Section 3.4): a replay is consistent when it reproduces the same trigram-level transitions and respects the same cross-plane escalation obligations as the original governed change.

Proposition 4 — variety useful only when governed. Platform logic is justified only where the system generates legitimate local variation without losing comparability, compatibility, or traceability. Proposition 4 rewards only the variation that shared rules continue to govern. The proposition is rejected if additional variety is purchased by bypassing the core interface logic, if exceptions proliferate without rule explanation, or if complements cease to remain compatible with the persistent governance core.

Proposition 5 — integrated utility under diachronic burden. Proposition 5 is deliberately integrative and is therefore the hardest proposition in the set. Practical utility must improve under a composite criterion: burden reduction is insufficient if reliability falls, and reliability gain is insufficient if the burden of use remains prohibitive.¹¹² It carries a non-negotiable falsifiable floor — the suite must not increase verification burden for equivalent assurance quality — and it places a burden on Chapter 4 to design against a learning-effect confound, since burden reduction observed after the suite arrives may partly reflect practitioner familiarity with any new workflow rather than the governance mechanism specifically.

The propositions do not specify artefacts directly; they initiate a derivation chain — propositions to governance functions to representational requirements to a constrained design space to engineering judgements — terminating in the architectural decisions of Chapter 5 through Chapter 8. The chain reaches necessity conditions only; several designs could satisfy the same proposition, so the artefact chapters must show both that a design satisfies the constrained space and why it is preferable among admissible alternatives. That preference rests on three criteria: parsimony, the simplest form satisfying the governance function without sacrificing an atomicity condition; practitioner tractability, governance operations performable by housing practitioners under bounded rationality, which is the direct consequence of DC-4; and institutional compatibility, the requirement to operate within existing Australian regulatory vocabulary and workflows while documenting any departure, which is the direct consequence of DC-2. We disclaim Proposition 4’s operational mechanism here; it is developed in Chapter 6, Section 6.4 as Rule-4 variant inheritance.

The propositions are bounded by context, and those boundaries are part of the contribution rather than a late caveat. The theory applies where adaptation recurs over time, where representational handover materially affects decisions, and where verification burden is a genuine constraint. It does not purport to explain all housing outcomes, policy disputes, or design variation, nor does it replace institutional judgement, occupant negotiation, or political contestation.¹¹³ Positive evidence could justify a mid-range prescriptive claim for SFS, bounded to housing adaptation and related change-intensive domains; SFS makes no universal claim across every built-environment setting. Gregor and Jones, we note, treat scope and boundary conditions as integral parts of a design theory.¹¹⁴ Chapter 4 inherits the obligation to operationalise each indicator, baseline, and falsifier before any later evidence can count.

A mechanism contract is now fixed, and what remains is to record what the chapter has settled and what it has not. The argument moved through four stages. It began with the constraint environment: housing adaptation operates under wicked-problem and complex-adaptive-system conditions, where the governing challenge is revision under contestation and feedback rather than terminal optimisation.¹¹⁵ It then specified the local mechanism — modularity as interface-governed coupling control, grounded in a trigram-graph-text representational substrate, with four rival traditions absorbed as design constraints DC-1 through DC-4.¹¹⁶ From there it scaled the logic into a platform architecture arrangement, a stable representational core plus a governed instance library organised through three-plane stratification,¹¹⁷ and finally converted those mechanisms into a design theory contract of five propositions, Proposition 1 through Proposition 5, each carrying indicators, baselines, and falsifiers.¹¹⁸¹¹⁹

Five commitments are now fixed for the remainder of the thesis. Adaptation as revision-governed. Housing adaptation is repeated change under contested conditions; the encoding of one final layout is not the object. Modularity as coupling control. Modularity means interface-governed coupling control: it is a property of how boundaries govern exchange, independent of physical style or prefabrication.¹²⁰ Platform as governed variation. Platform architecture is the system-scale governance arrangement that holds a stable core while permitting rule-bound complement variation.¹²¹ Design theory as contract. Design theory is a formal contract binding mechanisms to evidence-bearing propositions, after the Gregor-Jones anatomy.¹²² Substrate as enabler. The trigram-relational-graph-text architecture makes the prior four operational: the trigram supplies the atomic governance unit, the relational graph the assessable topology, and text serialisation the governance-preserving medium.

These commitments also define what counts as progress in the chapters that follow. Progress means materially advancing one or more of the five meta-requirements, Meta-Requirement 1 through Meta-Requirement 5, as measured through the corresponding proposition.¹²³¹²⁴ A longer artefact description, a more elaborate demonstration, or a larger volume of generated outputs qualifies only when it advances at least one such requirement. A later chapter that cannot show how its outputs move one or more of those commitments leaves the contribution unstrengthened, however interesting it may be on technical grounds.

Chapter 4 therefore inherits a precise burden: it must operationalise each proposition as a procedure with declared rejection logic, defining at minimum six elements — the unit of change, the invariant format, the mapping from triggers to checks, the baseline against which each proposition is judged, the evidence objects by which each judgement is made, and the adjudication rules that distinguish pass, fail, indeterminate, and boundary-limited acceptance.¹²⁵ The burden is dual. Production rules alone fall short — a transformation log cannot tell the reader what counts as enough improvement, or how conflicting indicators are weighed — so adjudication rules must be specified alongside them.¹²⁶ Temporal comparability adds a further requirement: a comparable starting state, intervention scope, and verification burden across change episodes, which bears centrally on Proposition 2, Proposition 4, and Proposition 5. Proposition 5 carries one more obligation, namely to separate mechanism-caused burden reduction from the learning-effect reduction any practised workflow would show.¹²⁷

Three clarifications bound what is claimed. Representational design does not resolve plural value conflict; wickedness remains.¹²⁸ Modular and platform arrangements deliver benefit only under adequate governance: if interfaces drift, complements escape governance, or the core hardens into rigidity, the mechanism claim fails. Transfer beyond the declared scope is not claimed. If the downstream methodology and evaluation succeed, we will not claim to have solved housing adaptation in general; we will have shown that a representational regime built around stable identity, explicit interfaces, executable transformations, and governed evidence makes adaptation more governable by confining each change’s verification to the boundaries it touches.¹²⁹¹³⁰ On our reading, that is one architectural claim discharged through five evidence-bearing instantiations, an interpretation we revisit in Chapter 11, Section 11.2. Chapter 4 must convert it into explicit procedures, measures, and rejection tests so that Chapters 5 through 9 function as one coherent design-science programme.¹³¹ The first artefact chapter, Chapter 5, addresses the stable-anchor requirement directly.

Next chapter: Chapter 4: Methodology →