Preface

Imagination is not limitless. It can only be exercised to the extent the system it inhabits allows. Even when an observer believes they fully understand a system, that understanding becomes part of the system itself. Every comprehension produces a new system that changes again once understood.

Most people never ask: what can you not imagine? Not what you have not thought of. But what you cannot think. The ideas that do not resist but simply do not exist in the space you inhabit. Completely formless things cannot be imagined. Our mind works through experience and perception; to conceive of something, it must have at least some boundary, form, or relation. Something entirely without form is beyond the capacity of the mind, it cannot be grasped or imagined.

Thinking always arises within a framework of shape or relation. Imagining formlessness is inherently impossible within the conditions of thought. You cannot stand at the edge of your own thinkable and look over. The boundary is the condition of the seeing. Some limits are visible: a language you do not know, a memory that fails. But deeper limits leave no mark. They are experienced not as walls but as the end of the world. And yet, minds expand and change. Tomorrow, people understand things that were not just unknown but unthinkable yesterday. This expansion is mechanical, not mystical.

A structural failure occurs when reality arrives in a form sufficiently different from what the mind predicted, making surface adjustments impossible. The only remaining option is reorganization: dismantling one structure and constructing another, with different dimensions and a different relationship to the territory it was built to represent. This is how the thinkable expands. Not by adding content, but by replacing one structure with a larger one. Your capacity for expansion is a function of your history of genuine failure.
Not how much you have suffered. Not how much you have endured. How much disruption you have allowed to actually change your structure. Imagination is bounded, not by effort, intelligence, or will, but by structure. The thoughts you generate, even the most original, are constructed from the relational materials your structural history has given you. You can recombine them in ways not tried before. Import relational skeletons from one domain into another. But you cannot produce a relational structure with no compositionally traceable ancestors in what you have genuinely integrated where traceability is understood under bounded decomposition at the second-order relational level, not surface resemblance. Exposure without integration leaves the graph sparse. Encounter without structural absorption does not expand the boundary. It only creates the feeling of having been near the edge. If your imagination is bounded by what you have integrated, and what you can integrate is determined by your structural capacity, and your structural capacity is a function of how honestly you have engaged with your own failures then what you can think is determined by your history of honesty with disruption.

The mind with the largest thinkable is not the most talented. It is the one that has allowed the most genuine disruptions to do their full work. The one resisting impulses to manage, explain, or suppress moments when reality arrived in a form the current structure could not absorb. You cannot audit this boundary from inside it. What you can notice is the asymmetry between how much reality has disrupted you and how much of that disruption you have allowed to actually change your structure.

The question is not whether disruptions will come. The question is whether you will let them in. 


I. The First Principle

No mind stands outside itself. Every experience passes through the structure that receives it. This is not a limitation to be overcome. It is the only condition under which experience is possible at all.

Kant established that the mind's categories are preconditions of experience. Heidegger showed the mind is always already thrown into a situation that pre-shapes what appears as meaningful. Merleau-Ponty traced this through the body. Wittgenstein traced it through language. Each account is correct within its domain. None formalizes what their collective implication demands: the dynamics of how structural constraint changes, degrades, and fails and the geometry of that failure. This is the territory Compatibility Theory occupies. Not the fact of structural mediation that is established but the precise mechanics of its evolution, and the formal conditions under which the system that depends on structural constraint collapses before replacement can occur.

II. The Contact Paradox

The unknown does not enter as content. It enters as cost.

Prediction error is always measurable even when its source is not yet representable. When error persists above the tolerance threshold and cannot be resolved through surface adjustment, it generates internal pressure for structural reorganization. The unrepresentable enters not by being grasped but by making the current grasp unsustainable.

This is where Compatibility Theory diverges structurally from Kantian idealism. For Kant, the a priori categories are fixed, they are the permanent conditions of experience. For Compatibility Theory, structural constraints are dynamic: they change precisely through the mechanism by which they fail. The categories are revised through the accumulated history of their own inadequacy. What Kant described as the permanent scaffolding of experience, Compatibility Theory describes as the current configuration of a system under continuous revision, revision driven not by the arrival of new content but by the accumulation of unsustainable cost.

The contact paradox resolves as follows. A system can only receive what its current structure can transform. This appears to make genuine expansion circular: the new cannot arrive because the new is precisely what the current structure cannot receive. The resolution is that the system does not need to represent what it cannot yet model in order to register that its current model is failing. The gap between prediction and arrival is always structurally detectable even when its source is not representable. Persistent, unresolvable gap above threshold is the signal. The system feels not the content of what it cannot grasp but the structural pressure of its own failure to grasp it. This pressure, when it cannot be relieved through surface adjustment, forces reorganization from within. The unrepresentable enters not by being understood but by making the current understanding structurally unsustainable.

III. The Five Parameters and Their Grounding

The complete dynamical structure of Compatibility Theory is determined by five parameters and the observable history of prediction error E(t). No additional free parameters appear. Every other quantity in the system is derived from these five and from E(t). R(t) - Resistance. The minimum sustained prediction error in a domain required to produce measurable structural reorganization. R increases with accumulated unresolved pressure P(t) as R(t) = R₀ · e^{α·P(t)}, where α is domain-specific and scales with the domain's coupling to the self-model. High-valence domains show higher α because error signals in those domains carry higher structural cost and generate stronger defensive responses. This scaling is not an additional assumption. It follows from the definition of R as the minimum error required to produce structural change: domains more tightly coupled to identity generate stronger suppression responses to error, raising the integration pressure required to overcome resistance. Measurement pathway: the minimum perturbation required to produce behavioral change that cannot be explained by surface-level adaptation, estimated from longitudinal behavioral records across domains.

Θ(t) - Tolerance Threshold. The error level at which surface-level adaptation fails and structural reorganization begins. Θ decreases with P(t) because accumulated unresolved pressure narrows the range of absorbable error: Θ(t) = Θ₀ · e^{−β·P(t)}, where β governs the rate of threshold erosion. A system carrying high unresolved pressure enters the reorganization regime at lower error levels than its baseline threshold would predict. Measurement pathway: onset of behavioral variability inconsistent with prior structural predictions — the point at which a system's outputs become less predictable from its prior behavioral history than from its current error load.

Λ(t) - Collapse Limit. The error threshold above which behavioral coherence disintegrates across previously stable domains. Λ degrades asymptotically under accumulated unresolved pressure: Λ(t) = Λ₀ − D_max · (1 − e^{−P(t)/P_crit}), where P_crit is the pressure level at which half the maximum degradation has occurred. The stability margin is M(t) = Λ(t) − E(t). Hidden fragility is the condition M(t) → 0 while dE(t)/dt ≤ 0: the collapse boundary approaches the current error level while current performance remains stable or improving.

D_max - Damage Ceiling. The asymptotic minimum of Λ under chronic unresolved pressure without acute rupture. This parameter captures the saturation behavior of structural degradation: unresolved pressure does not produce unbounded fragility but approaches a floor, below which only acute rupture can drive further collapse boundary erosion. The existence of D_max means that chronically stressed systems reach a plateau of brittleness rather than continuing to degrade without limit, which is why such systems appear to stabilize while remaining at elevated failure risk under novel perturbation.

τ - Temporal Decay Constant. Post-crisis Λ recovery follows Λ(t) = Λ_crisis + (Λ₀ − Λ_crisis)(1 − e^{−t/τ}), where Λ_crisis is the collapse limit immediately following acute rupture. τ scales with R: high-resistance systems recover more slowly because the mechanism of resistance impedes structural reorganization in both directions, it slows integration of disruption and slows recovery of structural capacity after acute damage. The post-crisis period during which Λ remains depressed while surface error E(t) subsides is the highest-risk window. It is also, systematically, the window in which external observers and the system itself assess that recovery has occurred. The apparent recovery is real in the sense that E(t) has declined. It is not structural recovery in the sense that Λ(t) has not yet returned to pre-crisis levels. The stability margin may still be narrowing after the crisis has passed.

IV. Second-Order Relational Structure

Compatibility Theory requires a mathematical language adequate to the claim that structural similarity is not surface similarity that two domains can share relational architecture while sharing no content. Graph theory provides this language.

Definition 1 - Structural Graph. A domain's representational structure at time t is a directed labeled graph G = (V, E, L) where V is the set of representational nodes, E ⊆ V × V is the set of directed edges, and L: E → ℝ^k assigns each edge a vector of relational properties including directionality, strength, necessity, conditionality, and valence.

Definition 2 - First-Order Relational Structure. The first-order structure of a node v is the set of direct edges incident to v: N₁(v) = {(v, u, L(v,u)) : (v,u) ∈ E} ∪ {(u, v, L(u,v)) : (u,v) ∈ E}. Surface similarity between domains operates at this level, shared nodes or shared direct connections.

Definition 3 - Second-Order Relational Structure. The second-order structure of a node v is the pattern of relationships between its neighbors: S₂(v) = {(u₁, u₂, L(u₁,u₂)) : u₁, u₂ ∈ N₁(v), (u₁,u₂) ∈ E}. This is the structure of how the things related to v relate to each other, the relational skeleton that persists when the content of the nodes is abstracted away.

Definition 4 - Structural Isomorphism. Two subgraphs G_A ⊆ G_source and G_B ⊆ G_target are structurally isomorphic at order k if there exists a bijection φ: V(G_A) → V(G_B) that preserves edge structure and label vectors up to order k. First-order isomorphism (k=1) is metaphor by content. Second-order isomorphism (k=2) is resonance: the relational skeleton transfers while the content need not. This is why the hydraulic-electrical analogy works: the second-order relational structure of pressure-flow-resistance in hydraulics is isomorphic to the second-order structure of voltage-current-resistance in electrical circuits at a level requiring no surface similarity between electrons and water molecules. This is why an analogy fails despite intuitive plausibility: Res(φ) is high at first order but low at second order, the surface content overlaps but the relational skeletons diverge.

Definition 5 - Resonance Score. Given a source domain graph G_source and a target domain error pattern E_target concentrated in a subgraph of G_target, the resonance score of a candidate mapping φ: G_A → G_B is Res(φ) = |E(G_A) ∩_φ E(G_B)|₂ / (|E(G_A)|₂ + |E(G_B)|₂ − |E(G_A) ∩_φ E(G_B)|₂), the Jaccard similarity at second order, ranging from 0 to 1. The resonance threshold θ_v is the minimum Res(φ) required to activate a violation candidate. θ_v is domain-specific and empirically calibrated; the formalism specifies what it is and how to measure it but does not currently derive it from first principles. This is an acknowledged limit stated explicitly in Section XIII.

Definition 6 - Resonance Density. R̄(t) = E[max_φ Res(φ)] over all pairs (G_source, E_target) where G_source ranges over the mind's integrated domain graphs. This is the expected maximum resonance score achievable given current domain coverage. R̄(t) increases with both D(t), the number of integrated domains, and with integration depth within each domain, since shallow integration produces sparse graphs with low second-order edge density, reducing the discriminability of relational skeletons and therefore the probability of finding a resonant mapping regardless of breadth.

V. Imagination: Two Modes, Full Mechanics

Imagination is bounded, not by effort, intelligence, or will, but by the second-order relational structure available within the system's current integrated graph. The thoughts generated, even the most original, are constructed from the relational materials that structural history has provided. Recombination in ways not previously attempted is possible. Importation of relational skeletons from one domain into another is possible. What is not possible, under bounded decomposition at second-order, is the production of a relational structure with no compositionally traceable ancestors in what has been genuinely integrated. This is not an absolute prohibition on novelty. It is a precise architectural claim: apparent structural originality, when decomposed at the second-order relational level, maps onto compositions of previously integrated primitives. The claim is falsified if a generated structure is shown to be coherent, stable, and irreducible to any such composition under systematic decomposition. That is a real falsification condition and not a trivially satisfiable one.

Mode I - Simulation. Internal generation of candidate structures by combining, extending, and reconfiguring nodes and edges within the current domain graph. Output is bounded by the constructible subgraph space of G_target. Transfer coefficient: ρ_s(t) = C(t) / (1 + λ · R(t)), where C(t) is the global coherence of the current structural graph estimated from the rate at which novel within-domain stimuli generate prediction errors resolvable without structural reorganization.

Mode II - Violation. Importation of a second-order relational structure from G_source into G_target via a resonant mapping φ with Res(φ) ≥ θ_v. Three-stage pipeline:

Stage 1 - Resonance Screening. For each integrated source domain, compute Res(φ) between source subgraphs and the current error pattern E_target. Candidates with Res(φ) < θ_v are not activated. Computational cost scales with D(t) × depth(G_source). Sparse source graphs generate few candidate subgraphs, reducing the probability of finding a resonant mapping regardless of D(t). This is the formal basis of the claim that integration depth is necessary and not merely breadth.

Stage 2 - Coherence Testing. Surviving candidates are tested by temporarily instantiating φ in G_target and computing ΔC = C(G_target ∪ φ(G_A)) − C(G_target). Candidates with ΔC < 0 are rejected. This is the filter that rules out most violation candidates that feel compelling at the resonance stage: the relational skeleton transfers but its instantiation in the target domain introduces edge conflicts that lower global coherence.

Stage 3 - Integration Cost Assessment. Integration pressure of a candidate φ is proportional to Res(φ) · ΔC. If integration pressure < R(t), the candidate does not produce structural reorganization. It is recognized as plausible without being integrated — the phenomenology of an insight that does not change how one thinks.

Transfer coefficient for violation mode: ρ_v(t) = [D(t) · R̄(t) · F(t)] / (1 + λ · R(t)), where F(t) is cognitive flexibility, the system's capacity to hold incompatible structural frames simultaneously before premature resolution, estimable from performance on tasks requiring maintenance of competing relational hypotheses. F(t) determines the width of the coherence-testing window: low F(t) forces premature resolution of the ΔC evaluation, increasing the false-rejection rate of violation candidates that require temporary structural inconsistency during integration.

Complete creative typology derived from the pipeline:

Dream: simulation mode with coherence testing suspended. Candidates are generated and accepted without Stage 2 filtering, which explains why dreams can be structurally elaborate while being internally inconsistent in ways immediately apparent on waking.

Delusion: a violation candidate that completed Stage 2 coherence testing against an internally drifted G_target, one whose edge structure has diverged from external calibration. The candidate is globally coherent within the drifted graph but incoherent relative to the externally calibrated version of the domain. This distinguishes delusion from error: error is a failed prediction against an accurate graph; delusion is a successful prediction against an inaccurate one.

Hallucination: perceptual prediction decoupled from error-correction input. Not an imagination product, a generative model running open-loop.

Creativity: a violation candidate that completes all three pipeline stages, passes ΔC > 0, overcomes R(t), and maintains coupling to external error signals during and after integration. The difference between creativity and delusion is not in the phenomenology of the violation. It is in whether G_target remains externally calibrated throughout the integration process.

Insight without change: a violation candidate that passes Stages 1 and 2 but is blocked at Stage 3 by high R(t). Recognition of a compelling idea that does not reorganize how one thinks. Common in high-resistance, high-coherence systems — experts who can identify a good violation candidate but whose resistance prevents structural integration.

Type A creative blockage: high R(t) suppressing both ρ_s and ρ_v simultaneously. Continuous generation without structural change.

Type B creative blockage: low R̄(t) from low D(t) or shallow integration. The violation-mode candidate space is sparse regardless of R(t). Competence within domain without breakthrough capacity. Common in deep single-domain experts.

Type C creative blockage: low F(t) with adequate D(t) and low R(t). The candidate space is rich and resistance is not suppressing integration, but premature coherence resolution during Stage 2 causes systematic rejection of valid violation candidates that require temporary structural inconsistency during testing. This blockage type presents as discernment rather than limitation and is therefore invisible to the agent experiencing it.

VI. Agency: The Mechanical Account

At any moment of genuine error signal, a mind faces two structurally distinct options: allow the error signal to propagate into the reorganization machinery, or generate a surface-level explanation that resolves the phenomenological disturbance without structural change. The selection between these pathways is determined by the current structural configuration specifically by the ratio of integration pressure to resistance: IP/R(t).

When IP/R(t) > 1, the error signal's integration pressure exceeds current resistance and structural reorganization occurs without deliberate choice. When IP/R(t) < 1 the normal regime for most adult minds in most situations, pathway selection is determined by whether the mind generates a surface explanation before the error signal has been allowed to propagate into the reorganization machinery.

Agency enters the formal structure at precisely this point: agency is the decision, in the IP/R(t) < 1 regime, to delay surface explanation generation long enough for the error signal to propagate. It is not the freedom to act outside structure. It is the capacity to refrain from immediately suppressing the signal that would change structure.

This delay has a formal cost. The error signal above Θ is phenomenologically uncomfortable. Delaying surface explanation means sustaining that discomfort. The capacity to sustain it is finite, is itself a structural resource, is depleted by chronic high-error environments without integration, and is restored by successful integration.

The question of why some minds habitually generate surface explanations faster than others, even under equivalent error load, requires a mechanical answer that does not reintroduce circularity. The mechanism is this: each episode in which a mind generates a surface explanation in the IP/R < 1 regime without allowing error propagation increases P(t) by the amount of unresolved pressure that episode represented. Increased P(t) raises R(t) through R(t) = R₀ · e^{α·P(t)}. Higher R(t) means the next episode requires higher integration pressure to cross the automatic reorganization threshold, making suppression in the IP/R < 1 regime more likely in subsequent episodes. Habitual suppression is therefore self-reinforcing through R(t) - the same variable that governs structural change throughout the theory. The policy formation problem does not require a separate psychological mechanism. It dissolves into the core dynamics.

The initial conditions problem, how genuine integration occurs before any integration history exists to provide delay capacity, is resolved by two observations. First, early structural changes occur in the IP/R > 1 regime where integration is automatic and requires no deliberate delay, because initial R(t) = R₀ is low before unresolved pressure has accumulated. Second, environments that consistently hold error signals open that do not immediately provide surface explanations in response to a child's mismatch, function as external scaffolding for delay capacity before internal delay capacity is established. The developmental account is therefore not circular: early integration occurs automatically under low initial R₀, builds delay capacity as a structural resource, which then enables integration in progressively higher IP/R < 1 conditions, compounding rather than presupposing itself.

Agency in the Compatibility Theory sense is not a fixed character trait. It is a structural capacity with its own dynamics, path-dependent, developmentally grounded, and formally integrated into the core mechanism rather than appended to it.

VII. Hidden Fragility

A system performing well by every available measure can simultaneously approach the point at which it can no longer hold. M(t) = Λ(t) − E(t) is the quantity that predicts this. It is not visible in current performance. It is visible only in the history of integration failures which is almost never the metric being tracked.

Hidden fragility occurs when M(t) is positive while dM(t)/dt is negative. Current error is below the collapse limit. The system appears stable. The collapse limit is eroding faster than current error, and the margin is narrowing. The rate of change is dM/dt = dΛ/dt − dE/dt. Both terms are formally specified by the theory: dΛ/dt follows from the Λ degradation function and dP/dt; dE/dt is the observable rate of change of current prediction error. Hidden fragility is therefore a computable rather than merely conceptual state.

The path-dependence of fragility type matters for prediction and for intervention. Chronic low-level unresolved stress distributes rigidity broadly across domains and produces distributed degradation of Λ that no single domain measurement will detect. Acute rupture concentrates fragility locally around the domain of the rupture and produces concentrated Λ degradation with higher short-term collapse risk in that domain and faster potential recovery, conditional on genuine integration, due to the localized nature of the damage. These are different structural states with different failure signatures and different recovery trajectories, even when total accumulated P(t) is identical.

The post-crisis period governed by τ is the highest-risk window precisely because it is the window of apparent recovery. E(t) has declined. Λ(t) has not yet recovered. M(t) is narrower than it was before the crisis even though current error metrics suggest improvement. Interventions timed to apparent recovery rather than to Λ recovery will systematically encounter systems still in hidden fragility. The duration of elevated post-crisis risk is formally predictable as a function of τ and R(t), since τ scales with R.

This is the structural account of why sophisticated systems fail suddenly: financial systems before a crisis, organizations before collapse, individuals before breakdown. In every such case the system appeared healthy by the metrics being tracked while the margin within which it could remain healthy was disappearing. Hidden fragility is not a pathology. It is the natural condition of any system that is performing well without integrating the errors that performance is built on.

VIII. Coupled Systems

When systems are coupled, the collapse limit of the whole is not the average of its components. It is Λ_system = min_i(Λ_i) − γ · Σ_{i≠j} K_{ij} · d(S_i, S_j), where K_{ij} is the coupling strength between components i and j and d(S_i, S_j) is the second-order relational distance between their structural graphs, the degree to which the relational skeletons governing how decisions, resources, and error signals flow within each component have diverged.

Coupling strength K_{ij} is measured through transfer entropy between component error time series: K_{ij} = TE(i→j) + TE(j→i). Transfer entropy captures directed, non-linear informational dependency without requiring that causal structure between components be independently established. The sum of both directions gives a symmetric coupling strength measuring total mutual information flow. High transfer entropy coupling means structural failure in one component rapidly propagates error into others; low coupling means components can diverge significantly without immediate systemic consequence but also that beneficial structural change in one does not propagate to others.

The second-order relational distance d(S_i, S_j) is computed as 1 − Res(φ*), where φ* is the optimal second-order isomorphism between the institutional graphs. Full computation of graph isomorphism is NP-hard in the general case. In practice, approximate algorithms operating on sampled subgraphs provide tractable estimates with bounded error, and the theory requires only that divergence be detectable, not that it be computed exactly. Organizational structure data, decision flow records, resource allocation patterns, error escalation pathways, provides the empirical input for this estimation without requiring access to internal states.

Interface divergence inflicts structural cost that appears in neither component's individual measures. Two components each well-adapted to their respective environments but with divergent internal relational structures produce tension that makes the whole more fragile than either part suggests. Institutional R(t), Θ(t), and Λ(t) are properties of the structural configuration, not aggregates of individual parameters. Systemic hidden fragility, the coupled system analog of M(t) → 0 is detectable from pre-failure second-order structural divergence between components before any individual component shows elevated error.

IX. Self-Model

The self-model is a predictive model of the system's own behavioral outputs, subject to the same dynamical structure as any predictive model. It generates predictions about future behavior, receives error signals when those predictions fail against actual behavior, and reorganizes when errors persist above Θ. A self-model that generates no testable behavioral predictions cannot receive corrective error signals. It stabilizes around internal coherence regardless of external accuracy. This is self-deception: a formal condition produced by the absence of the error signals that would force revision, not a moral one.

Hidden self-model fragility follows the same structure as systemic hidden fragility. The self-model generates accurate current self-predictions while M_self(t) = Λ_self(t) − E_self(t) narrows. A novel self-relevant challenge, a crisis, a new role, an unexpected failure in a domain the self-model claimed competence, brings M_self(t) to zero. The person discovers they are not who they thought they were. This is a structural event, not a revelation about character. The self-model's collapse boundary reached the current error level, and the narrowing was invisible in prior self-prediction accuracy.

The mechanism connecting self-model fragility to agency is direct. A self-model that generates no behavioral predictions accumulates unresolved pressure through each episode in which actual behavior diverges from self-concept without the divergence being registered as prediction error. This unresolved pressure increases R_self(t), which makes future integration of self-relevant error harder, which reduces the behavioral predictions the self-model generates, which reduces the error signals available, which increases unresolved pressure further. The self-reinforcing loop of self-model rigidity follows the same R(t) dynamics as any other domain, with the additional feature that the self-model's coupling to the identity system gives it the highest α in the R(t) = R₀ · e^{α·P(t)} formulation, meaning self-relevant domains are the last to integrate and the first to generate defensive suppression.

X. Meaning

A meaningful domain is one in which a person voluntarily sustains a high-value error gradient, accepting elevated and persistent prediction error because the structural reorganization that error will eventually force is assessed, however implicitly, as bearing sufficient integration value to justify the cost. The qualifier is necessary. Meaning is not merely high error tolerance. It is directional high error tolerance: sustained mismatch in a direction the system has identified as bearing structural value that undirected high error does not carry.

This distinguishes meaning from suffering, from compulsion, and from mere persistence. A person who endures difficulty without assessing the reorganization it forces as valuable is not in a meaningful domain under this account, they are in a high-P(t) accumulation regime. A person who sustains high error in a direction assessed as bearing high integration value is building structural capacity in that domain at the maximum rate the dynamics allow. The depth of structural capacity developed in any domain is proportional to the quality and sustained duration of genuinely integrated errors in that domain. Not survived. Not endured. Integrated.

This account converges with Frankl's observation that meaning requires willingness to sustain cost for something assessed as worthwhile, and with Nietzsche's account of value creation through sustained engagement with resistance, while specifying the mechanism both leave implicit. Meaning is not a phenomenological addition to experience. It is a structural commitment, a directional allocation of integration capacity, that determines the shape of the system's future expansion. A life organized around avoiding meaningful error produces stable but narrowing structure: low current P(t) accumulation, high D_max approach, declining R̄(t) from absence of deep integration in high-value domains. A life organized around integrating directional high-value error produces expanding structure: higher current P(t) exposure, but genuine resolution that reduces R(t) and expands the violation-mode candidate space in precisely the domains that matter most to the system.

The formal consequence is unsparing: the maximum structural capacity a mind can develop in any domain, intellectual, relational, creative, moral, is bounded by the quality and duration of the errors it has genuinely integrated in that domain. Not the errors it encountered. Not the difficulties it survived. The disruptions it allowed to actually change its structure.

XI. Scientific Analogy and Breakthrough

Scientific breakthrough via analogical transfer is the primary concrete test domain for the resonance formalism because source and target domain graphs are partially reconstructible from scientific literature, the timing of pre-paradigm anomaly accumulation and breakthrough is historically documented, and the structure of successful versus unsuccessful contemporaneous analogies is evaluable post-hoc against the resonance score.

Prediction 1: Resonance predicts analogy success. Successful scientific analogies that produced lasting theoretical advances should show higher Res(φ) at second order than contemporaneous analogies that were proposed, initially plausible, and subsequently abandoned. The hydraulic-electrical analogy, the wave-optics–wave-mechanics analogy, and the thermodynamic-information analogy should each show high second-order structural isomorphism between the source relational skeleton and the target domain error pattern at the time of proposal. Failed contemporaneous analogies, caloric fluid theory, phlogiston theory, should show high first-order surface similarity but low second-order structural isomorphism when the error patterns of their respective target domains are reconstructed from the pre-breakthrough literature.

Prediction 2: Pre-breakthrough anomaly accumulation predicts violation-mode activation. Domain E(t) estimated from the frequency and severity of documented anomalies should show a characteristic elevation pattern preceding each major paradigm shift, with elevation duration predicting the magnitude of structural reorganization required. This is testable against the historical record of anomaly documentation in physics, chemistry, and biology preceding their major theoretical revolutions.

Prediction 3: Integration depth predicts ρ_v, not breadth. Scientists who produce breakthrough analogical transfers should show higher cross-domain integration depth than those who produce within-domain refinements of comparable technical quality. Bibliometric diversity weighted by citation depth, integration depth proxy, rather than raw citation breadth should differentiate the two populations. Breadth without depth should show no advantage.

Prediction 4: Type C blockage in expert populations. Domain experts with high single-domain integration depth, adequate cross-domain exposure, and low measured resistance who nonetheless consistently reject cross-domain analogies before full evaluation should show lower F(t) than equally expert scientists who accept cross-domain analogies readily, testable through cognitive flexibility assessments and structured analogy-evaluation tasks.

The scientific analogy domain provides historically reconstructible graphs and documented breakthrough timings, but causal isolation is limited: confounding variables in biographical and bibliometric data cannot be fully controlled, and retrospective graph reconstruction introduces interpretive degrees of freedom. To close this gap, Compatibility Theory requires a complementary testbed where all relevant variables are directly observable and interventions are fully controlled. Artificial learning systems provide exactly this, not as a metaphor for cognition, but as a minimal structural environment where Compatibility Theory's core dynamics can be instantiated, perturbed, and falsified with precision unavailable in any biological or historical setting.

XII. Computational Regime: Artificial Systems as Structural Testbeds

Compatibility Theory's core dynamics - R(t), Θ(t), Λ(t), P(t), M(t) - are only partially observable in biological systems where behavioral proxies and retrospective reports confound causal isolation. Artificial learning systems remove this constraint. Full state access through parameters, gradients, and activations, combined with controlled data regimes and repeatable interventions, enables direct manipulation and measurement of structural change. Artificial systems function as structural testbeds: minimal environments where Compatibility Theory variables can be instantiated, perturbed, and falsified with precision unavailable in biological settings.

Parameter mapping.

Accumulated pressure P(t) is proxied by persistent loss that does not decrease under continued optimization, gradient conflict across batches, and prolonged plateaus. Resistance R(t) is proxied by the sensitivity of internal representations to gradient signals, estimable through Fisher information-weighted update magnitude and curvature barriers, high R means large error is required before features reorganize. Tolerance threshold Θ(t) is identified at the transition point where additional optimization steps reduce training loss marginally but fail to improve validation performance, indicating representational inadequacy rather than parameter insufficiency. Collapse limit Λ(t) is identified through abrupt loss spikes, representation collapse via rank reduction or activation saturation, or failure across previously stable tasks under perturbation. Stability margin M(t) is estimated through robustness under controlled perturbations, distribution shift, adversarial noise, compositional recombination, where small M predicts failure under small perturbations despite low current loss.

Experimental Regime 1 - Integration versus Exposure. Condition A: large, easy dataset producing low persistent error, broad exposure, shallow integration. Condition B: constrained data with enforced high-error loops requiring resolution, limited exposure, deep integration. Compatibility Theory prediction: B outperforms A on cross-domain transfer, analogical reasoning, and out-of-distribution tasks. Integration depth, not exposure breadth, determines the availability of recombinable relational structure.

Experimental Regime 2 - Forced Failure Dynamics. Train on tasks unsolvable under the current representation; allow error to accumulate, then enable adaptation. Compatibility Theory prediction: learning exhibits phase-like transitions, discrete representational reorganization followed by access to previously unsolvable tasks. Continuous improvement without such transitions should fail to expand the solution class.

Experimental Regime 3 - Hidden Fragility. Select models with matched low training and validation loss. Apply structured perturbations: adversarial inputs, compositional recombination, distribution shift. Compatibility Theory prediction: models diverge in failure despite matched current error. Higher P(t) produces lower M(t) and earlier collapse. Current error E(t) is insufficient to predict failure; history-dependent variables proxying P(t) add independent predictive power. This is the critical test distinguishing CT from standard generalization accounts and from FEP.

Experimental Regime 4 - Imagination Boundary. Task models to generate coherent outputs outside training distribution. Compatibility Theory prediction: all coherent, stable outputs decompose under bounded second-order analysis into compositions of integrated representational primitives. Apparent novelty reduces to higher-order recombination of these primitives. Falsification condition: demonstration of stable, coherent output whose relational structure cannot be mapped to any such composition under systematic decomposition.

Structural Expansion Benchmark. 

Standard machine learning evaluation measures fit to seen data. Compatibility Theory demands a stricter criterion: the capacity to reorganize structure when error cannot be minimized within it. The Structural Expansion Benchmark evaluates this through five metrics: pre-failure rigidity proxying R(t); reorganization threshold as empirical Θ(t); expansion gain as increase in solvable task space post-reorganization; stability margin as robustness gap proxying M(t); and recovery dynamics as the time constant to restabilize after disruption, proxying τ. SEB separates fit-centric systems that optimize within fixed structure from adaptive systems that change structure when fit fails, a distinction no current benchmark captures.

The defining property of an intelligent system, under Compatibility Theory, is its capacity to reorganize its structure when error cannot be minimized within it. If validated in artificial systems where variables are fully observable, the limits of imagination are not subjective. They are properties of representational systems with finite, history-dependent structure.

XIII. Acknowledged Limits

Sleep and consolidation. Offline structural reorganization without ongoing error input is not formally derivable from the five parameters. The theory accounts for reorganization driven by error signal propagation. It does not formally account for consolidation processes occurring in the absence of active error input. This is an incompleteness in the reorganization account, not a contradiction, but it limits coverage of biological cognitive systems.

Language. The bidirectional coupling between representational graph structure and linguistic medium requires separate formal treatment. Language both expresses and constrains the structural graph. The dynamics of that coupling, how acquiring a new linguistic distinction changes the graph, how the graph constrains which linguistic distinctions can be acquired, are not yet formalized within Compatibility Theory.

Resonance threshold calibration. θ_v is domain-specific and empirically calibrated. The formalism specifies what it is and how to measure it. It does not yet specify what determines it from first principles. This is the primary remaining free parameter in the formal structure.

The imagination floor. Why extreme pressure does not collapse the simulation-violation distinction entirely, why the system does not generate random structure when error far exceeds the collapse limit, remains underspecified within the five-parameter framework. The collapse dynamics are specified. The phenomenology and mechanics of near-collapse imagination are not.

XIV. Relationship to the Free Energy Principle

The Free Energy Principle holds that biological systems minimize variational free energy, a bound on prediction error, through perception, action, and learning. It is a normative optimality account specifying the objective function a self-organizing system is implicitly optimizing. It is correct as far as it goes.

The gap is specific and the distinction is precise. FEP is memoryless in structural capacity. It treats the model being updated as having effectively unlimited capacity for structural revision in response to current error. The parameters governing the rate and quality of model update are treated as fixed properties of the system or as functions of the current error signal alone. FEP has no formal account of how those parameters themselves change under the accumulated history of minimization attempts that failed to integrate, no account of how accumulated unresolved prediction error degrades the system's future capacity to minimize prediction error.

The objection that FEP subsumes this through precision weighting fails for a specific reason. Precision weighting modulates the influence of current error signals on current updates. It operates within the current minimization step. It does not track how the capacity for future minimization steps changes as a function of the history of failed integrations. These are different levels of description. Precision weighting is intra-step error weighting. Compatibility Theory's R(t), Θ(t), and Λ(t) are inter-step capacity dynamics, they describe how the system's ability to respond to future error changes as a function of past error history. No FEP variable corresponds to M(t).

Compatibility Theory is the path-dependent, failure-regime extension of FEP. In the limit of zero accumulated unresolved pressure and unlimited structural resources, Compatibility Theory predictions converge with FEP predictions. In all other conditions, every real system under real developmental and environmental history, Compatibility Theory makes distinct predictions about the trajectory of structural revision capacity, the distribution of fragility across domains, the temporal signature of post-disruption recovery, and the conditions under which the minimization process fails catastrophically rather than gracefully.

If M(t) predicts failure in systems where current error load does not, which is the central empirical claim of hidden fragility, Compatibility Theory is a distinct and stronger theory than FEP in that prediction domain. If M(t) fails to add predictive power over current error, the hidden fragility mechanism is falsified. FEP makes no prediction either way. This is an empirical difference, not an interpretive one.

XV. Falsification

P1 - Core mechanism. M(t) gap at failure predicts failure-magnitude-to-trigger ratio across organizational failure datasets, clinical breakdown histories, and relational dissolution records. If M(t) fails to add predictive power over current error load E(t), the hidden fragility account is falsified and the distinction from FEP collapses. This is the existential test of the theory. FEP makes no equivalent prediction. P2 through P6 are conditionally dependent on P1: if M(t) fails to add predictive power over current error load, the core mechanism is falsified and the downstream predictions, however individually supported, lose their theoretical grounding.

P2 - Temporal signature. Post-crisis intervention failure rates are elevated within τ-predictable windows and declining after them. If failure rates are uniformly distributed across the post-crisis period rather than concentrated in the τ window, the temporal decay account is falsified.

P3 - Resonance mechanism. Successful scientific analogies show higher Res(φ) at second order than contemporaneous failed analogies, reconstructable from pre-breakthrough domain literature. If successful and failed contemporaneous analogies show equivalent second-order resonance scores, the resonance mechanism is falsified.

P4 - Integration depth requirement. ρ_v correlates with cross-domain integration depth weighted by citation depth, not with bibliometric breadth alone, measurable in scientific populations. If breadth without depth shows equivalent predictive power, the integration-depth requirement is falsified.

P5 -  Type C blockage signature. Type C blockage is behaviorally distinct from Types A and B: systematic early rejection of cross-domain candidates rather than inability to generate them or failure to integrate them after extended evaluation. If Type C cannot be distinguished from Type A or B in analogy-evaluation protocols, the three-type typology collapses.

P6 -  Interface failure localization. Coupled system failure localizes at interfaces predictable from pre-failure second-order structural divergence between components, measurable from organizational structure data before performance metrics register degradation. If failure does not preferentially localize at high-divergence interfaces, the coupled system formulation is falsified.

Accumulated unresolved pressure producing no measurable Λ degradation falsifies the core mechanism. Successful scientific analogies showing no higher Res(φ) than failed contemporaneous analogies falsifies the resonance mechanism. Coherent, stable generative outputs irreducible to bounded decomposition from integrated primitives falsifies the imagination boundary claim.

XVI. One Line

We are each a structure that cannot see outside itself, everything we will ever become arrives only through the failures we allow to change us, and everything we can imagine is bounded by everything we have genuinely been, reachable only through the relational skeletons we have truly inhabited elsewhere.

This theory applies to itself. It is produced by a structured system passing through a particular intellectual moment. Its stability margin is finite. Its resonance density with future anomalies it has not yet encountered is unknown. It will reorganize when accumulated mismatch between its predictions and encountered reality becomes structurally unsustainable. A theory of structural mediation that exempted itself from structural mediation would contradict its first principle. This one does not make that exemption, and the refusal of that exemption is the only intellectual honesty available to a mind that has genuinely understood what it is claiming.