# The Architecture of Failure: How Systemic Brittleness Drives Convergent Coherence to Forge Objective Truth

## Abstract

Coherentist theories of justification face the isolation objection: a perfectly coherent belief system could be detached from reality. While Quinean naturalism grounds knowledge in holistic webs constrained by empirical experience and pragmatic revision, the framework lacks formalized diagnostics for assessing the structural health of knowledge systems before failure occurs.

This paper develops Emergent Pragmatic Coherentism (EPC), building on Quine's architecture to address this gap. This naturalistic externalist framework grounds coherence in the long-term viability of public knowledge systems. EPC introduces systemic brittleness as a prospective diagnostic tool. It assesses epistemic health by tracking the observable costs generated when a system’s claims are applied. Selective pressure from these costs drives disparate knowledge systems to converge on an emergent, objective structure we term the Apex Network. The Apex is not a final theory in hand. It is the constraint-determined structure shared by maximally viable solutions, revealed through historical failure filtering.

In this framework, justification requires both internal coherence and the demonstrated resilience of the certifying network. Truth is analyzed in three levels: from contextual coherence within any system, to warranted assertion within a low-brittleness consensus network, to objective truth as alignment with the Apex Network. EPC synthesizes insights from Quinean holism, network epistemology, evolutionary dynamics, and systems theory while adding the externalist diagnostic of brittleness that its predecessors lacked.

By offering prospective guidance through constraint analysis and grounding a falsifiable research program in historical methods, EPC advances a form of Systemic Externalism. In this view, justification depends on the proven reliability of public knowledge architectures. Applications to historical paradigm shifts and contemporary epistemic challenges illustrate the framework’s diagnostic and explanatory power.

## 1. Introduction: Diagnosing Epistemic Health

Why did germ theory replace miasma theory? A standard explanation cites superior evidence. We agree, but add a systems-level claim about why that evidence mattered historically. Miasma theory incurred mounting operational costs, notably misdirected public health efforts in London, and demanded constant ad hoc modifications throughout the mid-19th century (Snow 1855). Its brittleness is evident in high patch velocity $P(t)$. Germ theory, by contrast, reduced these costs while unifying diverse phenomena.

This raises the fundamental question: how do we distinguish genuine knowledge from coherent delusions? A Ptolemaic astronomer's system was internally coherent, its practitioners trained in sophisticated techniques, yet it was systematically misaligned with reality. What external constraint prevents coherent systems from floating free?

This shift exemplifies the isolation objection to coherentism: a belief system might be coherent yet detached from reality (BonJour 1985). While internalist coherentist responses (Olsson 2005; Kvanvig 2012; Krag 2015) struggle to provide external constraints, Quinean naturalism addresses this by anchoring knowledge in empirical experience and pragmatic revision. We propose Emergent Pragmatic Coherentism, building on Quine's architecture (Quine 1960) and Kitcher's evolutionary model (1993) to formalize the diagnostic assessment of knowledge systems through systemic brittleness.

### 1.1 Lineage and Contribution

Emergent Pragmatic Coherentism builds upon a tradition beginning with Quine's naturalized epistemology (1969), continuing through Davidson's (1986) coherence theory, Kitcher's (1993) evolutionary model of scientific progress, and Longino's (1990) social account of objectivity. Each sought to dissolve the dualism between justification and empirical process. EPC accepts and extends this inheritance by describing the convergent implications of Quine's architecture and introducing systemic brittleness as a formalized diagnostic for epistemic health. Knowledge is not a mirror of nature nor a purely linguistic web of belief, but a living system maintained through adaptive feedback loops.

While Quine anchored epistemology in the psychology of individual agents revising holistic webs in response to recalcitrant experience, EPC extends this naturalistic foundation to analyze the macro-dynamics of public knowledge systems. Belief-formation and revision, grounded in Quine's account, aggregate into systems-level patterns that can be assessed through brittleness metrics. Where Davidson focused on internal coherence, EPC operationalizes Thagard's (1989) connectionist ECHO, modeling coherence as activation harmony in constraint networks and extending this with pragmatic inhibitory weights derived from brittleness. Unlike Zollman's (2007) abstract Bayesian graphs, which model the effects of network topology on belief propagation, EPC's framework tracks how knowledge systems respond to real-world pragmatic pressures through measurable costs.

The Quinean inheritance is architectural, not merely methodological. Our framework requires knowledge structured as holistic webs subject to pragmatic revision, coordinating into public systems converging toward constraint-determined objectivity. This architecture is non-negotiable: systemic brittleness requires interconnected networks; pragmatic pushback requires revision mechanisms; convergence requires overlapping webs aligning toward an objective standard. Quine's account provides this in thoroughly naturalized form. Alternative foundations preserving these features remain compatible with our systems-level analysis, though the specific dispositional semantics remains optional.

Rescher's (1973) abstract criteria for systematicity are given concrete diagnostic form through our brittleness indicators, while Kitcher's (1993) credit-driven evolution gains a failure-driven engine via the Negative Canon. Davidson's internal coherence becomes *structural homeostasis* maintained through pragmatic constraint. Kitcher's and Longino's insights into social calibration are operationalized: the intersubjective circulation of critique becomes a mechanism for reducing systemic fragility.

EPC thus transforms coherence from a metaphor of fit into a measurable function of cost and constraint. Its realism is not the correspondence of propositions to an independent world, but the emergent stability of constraint satisfaction across iterated cycles of error and repair. Building on Quinean naturalism, EPC grounds justification in the demonstrated viability of knowledge systems, providing formalized diagnostic tools to assess the structural health that Quine's framework presupposed but did not quantify.

Our response to the isolation objection is distinctive: coherence rests not on historical accident but on emergent necessary structure. Reality imposes pragmatic constraints: physical laws, biological limits, logical requirements, and coordination necessities. These constraints create a landscape that necessarily generates optimal configurations. These structures emerge from the constraint landscape itself, existing whether discovered or not, just as the lowest-energy state of a molecule emerges from quantum mechanics whether calculated or not. Objective truth is alignment with these emergent, constraint-determined structures.

We ground coherence in demonstrated viability of entire knowledge systems, measured through their capacity to minimize systemic costs. Drawing from resilience theory (Holling 1973), which models how systems absorb disturbance without regime change, we explain how individuals' holistic revisions to personal webs of belief in response to recalcitrant experience drive bottom-up formation of viable public knowledge systems. When aggregated at the systems level, this Quinean process of pragmatic revision generates measurable patterns we diagnose through brittleness metrics.

This transforms the isolation objection: a coherent system detached from reality isn't truly possible because constraints force convergence toward viable configurations. A perfectly coherent flat-earth cosmology generates catastrophic navigational costs. A coherent phlogiston chemistry generates accelerating conceptual debt. These aren't merely false but structurally unstable, misaligned with constraint topology.

The process resembles constructing a navigation chart by systematically mapping shipwrecks. Successful systems navigate safe channels revealed by failures, triangulating toward viable peaks. The Apex Network is the structure remaining when all unstable configurations are eliminated. Crucially, this historical filtering is a discovery process, not a creation mechanism. The territory is revealed by the map of failures, not created by it.

This paper models inquiry as evolutionary cultivation of viable public knowledge systems. It is a macro-epistemology for cumulative domains like science and law, proposing Lamarckian-style directed adaptation through learning rather than purely Darwinian selection.

Viability differs from mere endurance. A brutal empire persisting through coercion exhibits high brittleness; its longevity measures energy wasted suppressing instability. The distinction is crucial for our framework: viability is a system's capacity to solve problems with sustainably low systemic costs, empirically measurable through ratios of coercive to productive resources.

**Scope clarification:** This paper's primary focus is epistemological—it addresses justification of beliefs and the structure of knowledge systems. However, brittleness analysis applies to any system that makes implicit claims about how the world works: political institutions, social norms, economic arrangements. These are not mere analogies. A brutal state makes implicit empirical claims (e.g., "order requires centralized coercion," "dissent causes instability") that can be assessed via the same brittleness metrics we apply to scientific theories.

The framework treats social institutions as testable systems: they impose structures on human interaction, those structures generate costs when misaligned with human social reality ($C(t)$ from coercion, $P(t)$ from proliferating rules, $M(t)$ from institutional complexity), and those costs provide objective evidence of misalignment. This is not sociology replacing epistemology—it is epistemic analysis applied to a broader class of systems that embed empirical claims.

When we invoke social institutions in this paper, we are demonstrating the framework's generality, not changing the subject. The logic is identical: systems that accumulate brittleness reveal structural misalignment; low-brittleness systems earn higher warrant.

The framework incorporates power, path dependence, and contingency as key variables. Power exercised to maintain brittle systems becomes a primary non-viability indicator through high coercive costs. Claims are probabilistic: brittleness increases vulnerability to contingent shocks without guaranteeing collapse. This failure-driven process grounds fallible realism. Knowledge systems converge on emergent structures determined by mind-independent constraints, yielding a falsifiable research program.

The framework targets cumulative knowledge systems where inter-generational claims and pragmatic feedback enable evaluation. It provides macro-level foundations for individual higher-order evidence (Section 7), not solutions to Gettier cases or basic perception.

Ultimately, this paper develops a framework for epistemic risk management. This framing is crucial for understanding what brittleness analysis claims and what it does not claim.

Brittleness indicators do not prove a system's core claims are false. A rising trend in $P(t)$, $C(t)$, or $M(t)$ does not constitute logical refutation. What it signals is that the system is becoming a **higher-risk, degenerating research program**, where continued adherence or investment becomes increasingly irrational relative to lower-brittleness alternatives.

This parallels financial risk assessment. A portfolio manager does not claim that a high-volatility, illiquid asset will necessarily fail. The claim is probabilistic: this asset profile carries elevated risk of catastrophic loss when perturbations occur. Similarly, epistemic risk management identifies systems where structural fragility elevates the probability of collapse, major revision, or replacement when faced with novel challenges.

The framework thus provides prospective guidance—the diagnostic tools to assess structural health before hidden fragilities lead to crisis—while preserving fallibilism. Rising brittleness is a warning signal, not a proof of falsity.

### 1.2 Key Terminology

This paper develops several core concepts to execute its argument. Systemic brittleness refers to the accumulated costs that signal a knowledge system's vulnerability to failure. We argue that through a process of pragmatic selection, validated claims can become Standing Predicates: reusable, action-guiding conceptual tools. This evolutionary process drives knowledge systems toward the Apex Network, which we define as the emergent, objective structure of maximally viable solutions determined by mind-independent constraints.

### 1.3 The Position in Summary

To prevent misreading, we state the framework's commitments explicitly:

**Realism about constraints.** The pragmatic pressures that filter knowledge systems are real and mind-independent. The costs are objective. The constraint landscape exists whether we map it or not. When a brittle system collapses under its own accumulated costs, this reflects genuine misalignment with non-negotiable features of reality, not mere social consensus or instrumental convenience.

**Deflationism about truth.** "True" is not a substantive property naming correspondence, coherence, or utility. It is a device of endorsement. To call P true is to assert P, to license its use in inference, to treat it as available for building upon. Truth-talk functions to mark which claims have earned standing through pragmatic filtering, not to name a metaphysical relation between propositions and facts.

**Pragmatism about selection.** We do not "check" beliefs against reality through some privileged access. Reality selects among belief systems by imposing costs on misalignment. The structure that survives is discovered through this filtering, not deduced from first principles or verified through direct comparison. Our knowledge of the Apex Network is always indirect, inferred from the systematic pattern of what fails.

**Coherentism about justification.** Beliefs are justified by their role in a web: their inferential connections, their integration with other commitments, their contribution to the network's overall coherence. But the web itself must earn authority externally. Coherence is necessary but not sufficient. A belief can cohere perfectly with a phlogiston network or a flat-earth cosmology; this provides no justification unless that network has demonstrated viability through sustained low brittleness.

These commitments are compatible. Realism about what selects does not require correspondence theory about what "true" means. Deflationism about truth does not require antirealism about constraints. Coherentism about justification structure does not require isolation from external pressure. The framework integrates these positions by recognizing that they answer different questions: what exists (realism about constraints), what "true" means (deflationism), how we discover what works (pragmatic selection), and what makes individual beliefs justified (coherentist structure with externalist grounding).

## 2. The Core Concepts: Units of Epistemic Selection

Understanding how knowledge systems evolve and thrive while others collapse requires assessing their structural health. A naturalistic theory needs functional tools for this analysis, moving beyond internal consistency to gauge resilience against real-world pressures. Following complex systems theory (Meadows 2008), this section traces how private belief becomes a public, functional component of knowledge systems.

### 2.1 From Individual Dispositions to Functional Propositions: Architectural Requirements and One Naturalistic Foundation

Understanding how knowledge systems evolve requires clarifying their architectural prerequisites. The framework's core claims (systemic brittleness, pragmatic pushback, convergent evolution toward the Apex Network) presuppose a specific knowledge structure, not a collection of atomic beliefs or deductions from axioms. Three features prove essential, with a fourth emerging from their interaction.

First, holism: knowledge forms interconnected webs where adjustments ripple through the system. Brittleness accumulates systemically because modifications create cascading costs. An isolated false belief is simply false; a false belief embedded in a holistic web generates conceptual debt as the system patches around it.

Second, pragmatic revision: external feedback causally modifies knowledge structures. Without revision mechanisms, pragmatic pushback becomes inert. Systems could acknowledge costs without adjusting, rendering our entire framework toothless.

Third, multiple agents under shared constraints: independent agents navigate the same reality, facing identical pragmatic constraints. This is not about social coordination but parallel exploration of a common constraint landscape. Like multiple explorers independently mapping the same terrain, agents will converge on similar structures not through communication but through bumping against the same obstacles. The overlap we observe in human knowledge systems is evidence of this convergent discovery, not its cause.

Fourth, constraint-determined objectivity (emergent from the first three): knowledge systems converge toward an objective structure (the Apex Network) determined by mind-independent pragmatic constraints. This structure exists whether discovered or not, providing the standard for truth, but it is revealed through elimination of failures rather than known a priori. When multiple holistic systems undergo pragmatic revision in response to shared constraints, the convergent patterns that emerge reveal the objective structure of viable solutions. This is foundational realism without traditional foundationalism: there IS an objective foundation, but it must be discovered through extensive parallel inquiry rather than stipulated by reason. Our fallibilism concerns epistemic access (we never achieve certainty our map matches the territory), not ontological status (the territory has objective structure).

These architectural features are non-negotiable. What follows sketches one naturalistic foundation that provides this architecture in integrated form. Following Quine's call to naturalize epistemology (Quine 1969), we ground knowledge in dispositions to assent: publicly observable behavioral patterns. From a materialist perspective, even consciousness and self-awareness are constituted by higher-order dispositions—the recursive neural patterns that generate what we experience as awareness emerge from, rather than transcend, the material substrate of dispositional structures. Alternative accounts (coherentist epistemology, inferentialist semantics, neural network theories) remain compatible provided they preserve holism, pragmatic revision, and operation under shared constraints. We develop the Quinean version because it offers thoroughgoing naturalism with all required components. However, the systems-level analysis beginning in Section 2.2 does not depend on Quine's specific metaphysics of mental content, only on the architectural features just outlined.

The convergence from individual disposition to shared knowledge structures proceeds not through deliberate coordination but through parallel discovery—multiple agents independently developing similar solutions when faced with identical constraints. Social communication may accelerate this process but is not structurally necessary for the emergence of coherent, convergent knowledge systems.

#### 2.1.1 The Quinean Foundation: Disposition to Assent

We begin with Quine's core insight: a belief is not an inner mental representation but a disposition to assent—a stable pattern of behavior (Quine 1960). To believe "it is raining" is to be disposed to say "yes" when asked, to take an umbrella, and so on. This provides a fully naturalistic starting point, free from abstract propositions.

#### 2.1.2 The Functional Bridge: Belief as Monitored Disposition

Here, we add a crucial functional layer to Quine's account. While a disposition is a third-person behavioral fact, humans possess a natural capacity for self-monitoring. From a materialist perspective, this self-monitoring capacity is itself constituted by higher-order dispositions to assent—the brain's neural architecture generates dispositions all the way down. When we speak of "awareness" of our dispositional states, we are not positing a separate observer but describing how complex, layered dispositions create the phenomena we call consciousness. This awareness is not a privileged glimpse into a Cartesian theater but an emergent property of recursive dispositional structures—dispositions about dispositions—analogous to proprioception, that allows for self-report and deliberate revision.

#### 2.1.3 From Disposition to Assertion

Following Quine, we can treat belief as a disposition to assent. For present purposes we need only one further step: when a speaker asserts a sentence publicly, it becomes a claim that can be tested, challenged, and integrated into a wider body of commitments. The unit of analysis in what follows is therefore not a proposition in the abstract sense, but a **publicly assessable claim**—an assertion-token or sentence-type in use—together with its inferential role in the network.

#### 2.1.4 Parallel Discovery and the Emergence of Objectivity

When multiple agents independently navigate the same constraint landscape, their publicly assessable claims converge not through coordination but through parallel discovery of viable solutions. Through pragmatic feedback, independent agents develop shared dispositions to assent to certain sentences in certain contexts because doing so has proven viable when tested against the same reality. "Water boils at 100°C" is not a discovered Platonic truth, but a sentence that multiple independent inquirers have become strongly disposed to assent to because this disposition enables immense predictive success when facing the same thermodynamic constraints.

This directly addresses Quine's indeterminacy thesis. While semantic reference may remain metaphysically indeterminate, multiple agents can achieve functional convergence. The shared disposition to assent to the sentence "Water is H₂O" is precise enough to ground the science of chemistry not because chemists coordinated their beliefs, but because the constraint landscape of chemical reality forced convergence on this particular dispositional pattern. The objectivity of the Apex Network, therefore, is not the objectivity of a Platonic realm, but the emergent objectivity of constraint-determined optimal configurations that multiple agents independently discover.

**Standing Predicates as Evolved Tools.** Publicly assessable claims that dramatically reduce network brittleness undergo profound status change. Their functional core is promoted into the network's processing architecture, creating a Standing Predicate: a reusable conceptual tool that functions as the "gene" of cultural evolution. When a doctor applies the Standing Predicate `...is an infectious disease` to a novel illness, it automatically mobilizes a cascade of validated, cost-reducing strategies: isolate the patient, trace transmission vectors, search for a pathogen, sterilize equipment. Its standing is earned historically, caching generations of pragmatic success into a single, efficient tool. Unlike static claims, Standing Predicates are dynamic tools that unpack proven interventions, diagnostics, and inferences.

By grounding epistemic norms in the demonstrated viability of convergent dispositional patterns, the framework addresses normativity: epistemic force emerges from pragmatic consequences of misalignment with constraint-determined structures. Following Quine's engineering model, epistemic norms function as hypothetical imperatives—if your goal is sustainable knowledge production, minimize systemic brittleness in these patterns.

#### 2.1.5 Why This Architecture Matters: Non-Negotiable Features

These architectural requirements are structural prerequisites, not arbitrary choices. Holism matters because brittleness accumulates through cascading costs across interconnected networks. Atomistic beliefs cannot exhibit systemic brittleness; only in holistic webs do adjustments create ripple effects, generating the conceptual debt (P(t)) and complexity inflation (M(t)) our diagnostics track.

Pragmatic revision matters because external costs must causally modify knowledge structures. Without revision mechanisms, pragmatic pushback becomes inert—systems could acknowledge costs without adjusting. The Quinean architecture ensures costs drive actual restructuring.

Multiple agents navigating shared constraints matters because the Apex Network emerges through parallel discovery, not social coordination. When independent agents face the same constraint landscape and possess pragmatic revision mechanisms, their belief webs will converge on similar structures—not because they coordinate, but because reality's constraints carve out the same viable pathways. Like multiple species independently evolving eyes in response to light, multiple agents independently develop similar knowledge structures in response to shared pragmatic pressures. The overlapping patterns we observe aren't products of coordination but inevitable convergences forced by the constraint landscape itself.

Constraint-determined objectivity matters for realism. The framework's response to the isolation objection requires that reality impose an objective structure. The Apex Network must exist as an objective feature of the constraint landscape, discovered through elimination rather than created by consensus. This distinguishes viable knowledge from coherent fiction.

These features work together as a package: holism without revision yields stasis; revision without shared constraints yields divergent, incomparable systems; shared constraints without revision prevents discovery of viable structures; convergence without objective constraints would be mere coincidence, not evidence of underlying reality. The coherent subset of networks emerges naturally when these conditions are met—social coordination may accelerate convergence but is not structurally necessary. Alternative foundations preserving these features remain compatible with our analysis.

### 2.2 The Units of Analysis: Predicates, Networks, and Replicators

Having established the architectural requirements (holistic, pragmatically-revised, networks of multiple agents converging under shared constraints toward constraint-determined objectivity) and sketched one naturalistic foundation (Quinean dispositions), we now shift to the systems level where the framework's distinctive contributions emerge. Our deflationary move redirects attention from individual agent psychology to public, functional structures. The units of analysis that follow (Standing Predicates, Shared Networks, and their evolutionary dynamics) depend on the required architecture but remain neutral on metaphysics of belief. The convergence of independent agents' belief webs into overlapping knowledge structures occurs through parallel discovery—behavioral patterns shaped by sustained pragmatic feedback from the same constraint landscape.

Standing Predicate: The primary unit of cultural-epistemic selection: validated, reusable, action-guiding conceptual tools within propositions (e.g., "...is an infectious disease"). Functioning as "genes" of cultural evolution, Standing Predicates are compressed conceptual technology. When applied, they unpack suites of validated knowledge: causal models, diagnostic heuristics, licensed interventions. Functioning as high-centrality nodes in Thagard-style networks (2000), Standing Predicates maintain persistent activation through historical vindication, with propagation weighted by pragmatic utility rather than pure explanatory coherence.

In information-theoretic terms, a Standing Predicate functions as a Markov Blanket (Pearl 1988; Friston 2013): a statistical boundary that compresses environmental complexity into a stable, causal variable, achieving computational closure. When a doctor applies the predicate "...is an infectious disease," they draw a boundary allowing them to operate on coarse-grained variables (viral load, transmission vectors) without computing the infinite micro-complexity of molecular trajectories. This compression minimizes prediction error—the surprise the system experiences when its model misaligns with reality. Standing Predicates are boundaries that have proven thermodynamically efficient at carving reality at viable joints.

**Clarifying "standing" and the replicator analogy:** The term "Standing Predicate" denotes earned status, not metaphysical foundation. A predicate achieves "standing" through sustained validation across contexts: it stands ready for application without requiring practitioners to re-argue its warrant each time. Once "infectious disease" achieves Standing Predicate status, medical practitioners need not relitigate germ theory when diagnosing a novel pathogen; the predicate's historical track record grants it operational standing.

The biological replicator metaphor ("genes of cultural evolution") is structural analogy, not literal mechanism. Standing Predicates function analogously to biological replicators: they are informational templates that persist and spread because they reliably solve recurrent problems. Both encode compressed solutions to environmental challenges; both replicate through copying with variation; both face selection through performance. But this is analogy about functional architecture, not claim about evolutionary psychology or memetics. We are doing pragmatic epistemology—tracking which conceptual tools survive filtering by constraint-determined viability—not making mechanistic claims about cultural transmission dynamics.

Shared Network: Emergent public architecture of coherent propositions and predicates shared across individual belief webs for collective problem-solving. Networks nest hierarchically (germ theory within medicine within science). Their emergence is a structural necessity, not negotiation: failure-driven revisions converge on viable principles, forming transmissible public knowledge. As resource-constrained systems, networks implicitly evaluate each proposition through a brittleness filter: will integrating this claim reduce long-term systemic costs, or will it generate cascading adjustments and conceptual debt? This economic logic drives convergence toward low-brittleness configurations. We use Consensus Network to denote a Shared Network that has achieved broad acceptance and a demonstrated low-brittleness track record.

Drawing from evolutionary epistemology (Campbell 1974; Bradie 1986) and cultural evolution (Mesoudi 2011), networks' informational structure (Standing Predicates) acts as replicator (copied code) while independent agents navigating shared reality serve as interactors (physical vessels for testing). This explains knowledge persistence beyond particular communities (e.g., rediscovered Roman law): the informational structure can be preserved and retested by different agents facing the same constraints. Independently formed networks converging on similar structures reveal an objective constraint landscape underwriting successful inquiry, anticipating the Apex Network (Section 4).

#### Conceptual Architecture

The framework's core dynamics can be visualized as:

```
Belief Systems → Standing Predicates → Feedback Constraints → Brittleness Metrics → Apex Network
     ↓                ↓                        ↓                    ↓              ↓
Private States   Public Tools             Pragmatic Pushback   Diagnostic Tools  Objective Structure
```

This flow illustrates how individual cognition becomes public knowledge through constraint-driven selection.

### 2.2.1 From Local Error to Systemic Failure

A crucial distinction grounds our framework: the difference between local error and systemic fragility.

**Local error** is ubiquitous in all knowledge systems. A single failed prediction, an isolated false belief, an anomalous result—these are the normal texture of inquiry. Ptolemaic astronomy made accurate predictions for specific celestial events. Phlogiston theory correctly identified many combustion phenomena. No knowledge system is immune to local errors, and identifying them is what drives normal science.

**Systemic fragility** emerges not from individual errors but from the pattern of response to error over time. When a system encounters resistance from reality, how does it adapt? Does it revise core principles, or does it add defensive patches? Do corrections generate broader unification, or do they fracture coherence into special cases? Does maintaining the system require escalating resources, or does it grow more efficient?

The framework focuses on this second-order pattern: the trajectory of costs incurred as systems interact with pragmatic constraints. A system exhibiting low brittleness absorbs errors through systematic refinement. A system exhibiting high brittleness responds to errors by accumulating conceptual debt, suppressing dissent, and requiring increasing resources merely to maintain baseline function.

This is why brittleness predicts collapse in ways that local falsification cannot. Practitioners can always accommodate individual anomalies—Kuhn showed this decisively. What they cannot indefinitely sustain is the escalating cost of accommodation itself.

### 2.3 Pragmatic Pushback and Systemic Costs

**Defining Systemic Brittleness.** An epistemic system is systemically brittle to the extent that maintaining its coherence and usability under empirical or practical pressure requires escalating ad hoc modifications, increasing insulation from feedback, or disproportionate resource expenditure.

This definition makes explicit several key features:
- **System-level property:** Brittleness is not a property of individual propositions but of entire knowledge architectures.
- **Diachronic pattern:** The key indicator is escalation over time—rising costs, not static levels.
- **Comparative measure:** Assessed relative to what maintaining coherence without escalating costs would require.

Crucially, brittleness is distinct from several concepts it might be confused with:
- **Not falsity:** A brittle system can contain true propositions; the issue is the cost of maintaining them.
- **Not irrationality:** Practitioners of brittle systems may be perfectly rational given their evidence; brittleness is structural, not psychological.
- **Not lack of coherence:** Brittle systems can be internally coherent; the problem is external misalignment revealed through costs.
- **Not unpopularity:** A system can be widely accepted yet brittle, or marginal yet viable. Brittleness tracks constraint-driven costs, not social opinion.

Shared networks are active systems under constant pressure from what we term pragmatic pushback: Quine's "recalcitrant experience" observed at the systems level. Where Quine described how anomalous sensory stimulations force adjustments in an individual's dispositions to assent, we track how these individual revisions aggregate into observable patterns of systemic cost when knowledge systems are applied at scale. Pragmatic pushback manifests as concrete, non-negotiable consequences: a bridge collapses, a treatment fails, a society fragments. These are the same empirical constraints Quine identified, now visible through their accumulated impact on public knowledge architectures.

Pragmatic pushback manifests as information leakage: when a network's conceptual boundaries (its Markov Blankets) misalign with the territory's actual causal structure, reality "leaks through" in the form of prediction errors, failed interventions, and cascading anomalies. A phlogiston theorist doesn't simply hold a false belief; their entire causal framework generates continuous surprise as each experiment reveals discrepancies their model cannot anticipate. Systemic brittleness is the aggregate measure of this information leakage—the thermodynamic cost of maintaining misaligned boundaries against environmental feedback. This generates two cost types.

First-Order Costs are direct, material consequences: failed predictions, wasted resources, environmental degradation, systemic instability (e.g., excess mortality). These are objective dysfunction signals. Systemic Costs are secondary, internal costs a network incurs to manage, suppress, or explain away first-order costs. These non-productive expenditures reveal true fragility; for a formal mathematical model of systemic brittleness and its dynamic evolution, see Appendix A.

Systemic brittleness, as used here, is a systems-theoretic measure of structural vulnerability, not a moral or political judgment. It applies uniformly across empirical domains (physics, medicine), abstract domains (mathematics, logic), and social domains (institutions, norms). The measure tracks failure sensitivity: how readily a system generates cascading costs when its principles encounter resistance. This diagnostic framework is evaluatively neutral regarding what kinds of systems should exist; it identifies which configurations are structurally sustainable given their constraint environment. A highly coercive political system exhibits brittleness not because coercion is morally wrong but because maintaining such systems against pragmatic resistance (demographic stress, coordination failures, resource depletion) generates measurable, escalating costs that signal structural unsustainability.

Conceptual Debt Accumulation: Compounding fragility from flawed, complex patches protecting core principles.

Coercive Overheads: Measurable resources allocated to enforcing compliance and managing dissent. While Quine's framework focuses on how anomalous experience forces belief revision in individual agents, it does not address how power can actively suppress such experience before it forces revision. Coercive overheads provide this missing diagnostic: resources maintaining brittle systems against pressures become direct, measurable non-viability indicators. Critically, coercion functions as information blindness. By suppressing dissent—the primary data stream signaling that a model is misaligned with reality—the system severs its own sensor loops. This is not merely an energetic cost but an epistemic one: the system goes blind to the gradient of the constraint landscape, guaranteeing eventual collapse regardless of available resources. A tyranny with infinite coercive capacity still cannot escape the thermodynamic consequences of operating on systematically false models.

Pragmatic pushback is not limited to material failures. In abstract domains like theoretical physics or mathematics, where direct empirical tests are deferred or unavailable, pushback manifests through Systemic Cost accumulation: secondary costs a network incurs to manage, suppress, or explain away dysfunction. Research programs requiring accelerating ad hoc modifications to maintain consistency, or losing unifying power, experience powerful pragmatic pushback.

These epistemic inefficiencies are real costs rendering networks brittle and unproductive, even without direct experimental falsification. The diagnostic lens thus applies to all inquiry forms, measuring viability through external material consequences or internal systemic dysfunction.

To give the abstract concept of brittleness more concrete philosophical content, we can identify several distinct types of systemic cost. The following indicators serve as conceptual lenses for diagnosing the health of a knowledge system, linking the abstract theory to observable patterns of dysfunction. These are analytical categories for historical and philosophical analysis, not metrics for a quantitative science.

| Conceptual Indicator | Dimension of Brittleness | Illustrative Manifestation |
| :--- | :--- | :--- |
| P(t) | Conceptual Debt Accumulation | Increasing share of publications devoted to anomaly-resolution vs. novel predictions |
| C(t) | Coercive Overhead | Increasing share of resources devoted to security/suppression vs. productive R&D |
| M(t) | Model Complexity | Rate of parameter/complexity growth vs. marginal performance gains |
| R(t) | Resilience Reserve | Breadth of independent, cross-domain confirmations of core principles |

**Methodological Caution**

The indicators above are diagnostic proxies, not a single commensurable scale. Brittleness is diagnosed by convergent signs: accelerating auxiliary structure AND narrowing exportability AND rising tuning overhead, not by any one scalar ratio that could mechanically "cancel out" competing virtues.

We do not propose mechanical formulas of the form "P(t) > 0.7 = brittle." Such precision is spurious. Instead, these metrics make visible resource allocation patterns that standard evidential accounts systematically miss. Think of them as analogous to Minimum Description Length (Rissanen 1978) or algorithmic complexity: we assess incompressible exception-handling relative to core explanatory structure. When the description length of auxiliary machinery (patches, special cases, ad hoc adjustments) begins to exceed the description length of core principles, brittleness is rising.

This is not ratio arithmetic where one novel prediction cancels one auxiliary hypothesis. It is pattern recognition: identifying the systemic signature of a knowledge system fighting reality's constraints rather than aligning with them. The diagnostics are heuristic indicators tied to a principled complexity measure, not mechanical scoring rules.

**Operationalizing P(t): Distinguishing Patches from Development.** A persistent challenge in applying these metrics is distinguishing defensive patches from legitimate theoretical refinement. Not every modification signals brittleness. The framework provides structured criteria for this distinction, though applying them requires interpretive judgment informed by domain expertise.

A modification counts as a patch, contributing to P(t), when it exhibits these characteristics: it is primarily anomaly-driven rather than prediction-driven, responding to known discrepancies rather than generating novel testable predictions; it is ad-hoc rather than systematic, addressing a specific problematic case without deriving from or contributing to broader theoretical principles; it reduces integration by adding special-case rules that fracture the theory's unity rather than enhancing it; and it shows an accelerating pattern, where the rate of such modifications increases over time as each patch generates new anomalies requiring further patches.

Legitimate theoretical development, by contrast, is generative rather than defensive, opening new research programs and predicting previously unconsidered phenomena; it is systematically motivated, deriving from core principles and extending their application rather than circumventing their failures; it increases integration by unifying previously disparate phenomena under common principles; and it shows stable or decreasing modification rates, as refinements resolve multiple anomalies simultaneously.

Concrete diagnostic questions help distinguish these patterns. Does the modification resolve only the triggering anomaly, or does it address a class of related problems? Can the modification be derived from existing core principles, or does it require invoking entirely new mechanisms? Does it generate testable predictions beyond the anomaly that motivated it? Would removing it collapse a narrow application or destabilize the entire theoretical structure? A modification answering "only the triggering anomaly," "entirely new mechanisms," "no new predictions," and "narrow application" exhibits the signature of a patch. This remains a hermeneutic exercise requiring domain knowledge, but these criteria provide structured guidance for assessing whether a research program is accumulating conceptual debt or genuinely progressing.

The following examples use these indicators as conceptual lenses for retrospective philosophical analysis of historical knowledge systems, illustrating how the abstract notion of brittleness could manifest as observable patterns. These are not worked empirical studies but conceptual illustrations that make the framework's diagnostic approach concrete.

Conceptual Illustration 1: Phlogiston Chemistry

Consider the transition from phlogiston theory to oxygen-based chemistry in the late 18th century. Phlogiston theory initially provided elegant unification: combustion, respiration, calcination all explained as phlogiston release. However, as quantitative measurements accumulated (Lavoisier's careful mass tracking), the theory required increasingly elaborate auxiliary machinery. This case demonstrates how brittleness metrics can be retrospectively applied to historical knowledge systems, using documented patterns to illustrate the framework's diagnostic approach.

**Brittleness Metrics Applied Retrospectively:**

**M(t) - Model Complexity:** The theory began requiring "negative weight" for phlogiston to explain mass gain during combustion. Parameters and special cases proliferated while explanatory scope narrowed. Each anomaly demanded novel auxiliary assumptions rather than flowing from core principles.

**P(t) - Conceptual Debt:** By the 1780s, chemistry publications were predominantly defensive: explaining anomalies, reconciling inconsistencies, rather than extending the framework to new phenomena. The research program shifted from discovery to damage control.

**C(t) - Coercive Overhead:** Phlogiston defenders required increasing institutional resources to maintain the framework's authority despite mounting difficulties. Lavoisier faced sustained resistance from established chemists wielding institutional power.

**R(t) - Resilience Reserve:** The framework's core principles lacked exportability. Attempts to extend phlogiston reasoning to new reaction types or thermal phenomena generated further anomalies rather than unified explanations.

By contrast, Lavoisier's oxygen-based chemistry achieved dramatically lower brittleness across all metrics:
- **Lower M(t):** Unified quantitative framework (stoichiometry) with fewer auxiliary principles
- **Lower P(t):** Immediate novel predictions and extensions to new reaction types
- **Lower C(t):** Rapid adoption once empirical weight became accessible to verification
- **Higher R(t):** Framework exported successfully to thermodynamics, biochemistry, industrial chemistry without fundamental revision

The point is not that one anomaly refuted phlogiston, but that sustaining it required rising maintenance costs with diminishing explanatory returns. This cost accumulation is what brittleness makes visible. The transition was driven not by crucial experiments alone but by the unsustainable operational burden of maintaining the phlogiston framework.

Conceptual Illustration 2: Contemporary AI Development. Current deep learning paradigms could be analyzed for early signs of rising brittleness. For instance, one might assess Model Complexity (M(t)) by examining the trend of dramatically escalating parameter counts and computational resources required for marginal performance gains. Likewise, one could interpret the proliferation of alignment and safety research not just as progress, but also as a potential indicator of rising Patch Velocity (P(t)): a growing need for post-hoc patches to manage emergent, anomalous behaviors. These trends, viewed through the EPC lens, serve as potential warning signs that invite cautious comparison to past degenerating research programs.

Having established brittleness as our diagnostic concept, we now address the methodological challenge: how to measure these costs objectively without circularity.

## 3. The Methodology of Brittleness Assessment

### 3.1 The Challenge of Objectivity: Achieving Pragmatic Rigor

Operationalizing brittleness faces a fundamental circularity: measuring systemic costs objectively requires neutral standards for "waste" or "dysfunction," yet establishing such standards appears to require the very epistemic framework our theory aims to provide.

While this appears circular, it is the standard method of reflective equilibrium, operationalized here with external checks. We break the circle by anchoring our analysis in three ways: grounding measurements in physical constraints, using comparative analysis, and requiring convergent evidence.

Brittleness assessment remains partially hermeneutic. The framework provides structured tools rather than mechanical algorithms, offering "structured fallibilism" rather than neutral assessment. This methodology provides pragmatic objectivity sufficient for comparative assessment and institutional evaluation.

We frame the approach as epistemic risk management. A rising trend in a system's brittleness indicators does not prove its core claims are false. Instead, it signals the system is becoming a higher-risk, degenerating research program, making continued investment increasingly irrational. Just as financial risk management uses multiple converging indicators to assess portfolio health, epistemic risk management uses brittleness metrics to assess knowledge systems before hidden fragility leads to catastrophic failure.

### 3.2 The Solution: A Tiered Diagnostic Framework

To clarify how objective cost assessment is possible without appealing to contested values, we organize brittleness indicators into a tiered diagnostic framework, moving from foundational and least contestable to domain-specific.

**Tier 1: Foundational Bio-Social Costs:** The most fundamental level: direct, material consequences of network misalignment with conditions for its own persistence. These are objective bio-demographic facts, measurable through historical and bioarchaeological data:
  - Excess mortality and morbidity rates (relative to contemporaneous peers with similar constraints)
  - Widespread malnutrition and resource depletion
  - Demographic collapse or unsustainable fertility patterns
  - Chronic physical suffering and injury rates

Systems generating higher death or disease rates than viable alternatives under comparable constraints incur measurable, non-negotiable first-order costs. These metrics are grounded in biological facts about human survival and reproduction, not contested normative frameworks.

**Tier 2: Systemic Costs of Internal Friction:** The second tier measures non-productive resources systems expend on internal control rather than productive adaptation. These are energetic and informational prices networks pay to manage dissent and dysfunction generated by Tier 1 costs, often directly quantifiable (see Section 2.3 for detailed treatment):
  - **Coercion Ratio (C(t)):** Share of state resources allocated to internal security and suppression versus public health, infrastructure, and R&D.
  - **Information Suppression Costs:** Resources dedicated to censorship or documented suppression of minority viewpoints, and resulting innovation lags compared to more open rival systems.

**Tier 3: Domain-Specific Epistemic Costs:** The third tier addresses abstract domains like science and mathematics, where costs manifest as inefficiency:
  - **Conceptual Debt Accumulation (P(t)):** Rate of auxiliary hypotheses to protect core theory (literature analysis).
  - **Model Complexity Inflation (M(t)):** Parameter growth without predictive gains (parameter-to-prediction ratios).
  - **Proof Complexity Escalation:** Increasing proof length without explanatory gain (mathematics).

While interpreting these costs is normative for agents within a system, their existence and magnitude are empirical questions. The framework's core causal claim is falsifiable and descriptive: networks with high or rising brittleness across these tiers carry statistically higher probability of systemic failure or major revision when faced with external shocks. This operationalizes Kitcher's (1993) 'significant problems' pragmatically: Tier 1 bio-costs define significance without credit cynicism, while C(t) measures coercive monopolies masking failures.

Robustly measuring these costs requires disciplined methodology. The triangulation method provides practical protocol for achieving pragmatic objectivity.

#### 3.2.1 Cost-Shifting as Diagnostic Signal

This framework reveals cost-shifting: systems may excel in one tier (epistemic efficiency) while incurring catastrophic costs in another (bio-social harm). Such trade-offs signal hidden brittleness, as deferred costs accumulate vulnerability. Diagnosis identifies unsustainable patterns across tiers, not a single score.

#### 3.2.2 Model Complexity as Compression Failure

Before examining how to measure these costs objectively, we should clarify why Model Complexity $M(t)$ functions as a brittleness indicator at the conceptual level, not merely as an operational metric. The deeper insight is that viable theories function as compression: they reduce vast empirical data into compact, generative principles that enable prediction beyond the original observations.

Consider Newton's laws of motion: three concise equations that compress countless observations of falling apples, planetary orbits, and projectile trajectories into a unified framework. This compression is not merely elegant; it is epistemically powerful because only a genuinely compressed theory can extrapolate reliably to novel cases. A theory that has captured the underlying causal structure can generate predictions for situations never before encountered. This is the hallmark of deep compression.

Contrast this with a system approaching what we might call the incompressibility trap. Late geocentric astronomy required proliferating epicycles, deferents, and equants, each calibrated to specific observational data. As the system's complexity grew toward the complexity of the phenomena it described, it ceased to function as a compact theory and became closer to a sophisticated lookup table: a catalog of past observations with reduced generative power. When a theory requires ever more auxiliary machinery for marginal gains, it can reproduce what it has seen but cannot reliably anticipate what it has not.

This is why rising $M(t)$ signals brittleness. It indicates that a system is losing compression efficiency, requiring more and more internal complexity to maintain performance. This is not a subjective aesthetic preference for simplicity. It reflects a fundamental epistemic constraint: theories that approach the complexity of their subject matter lose the capacity to guide action in novel circumstances. They become brittle because they cannot compress reality's patterns, only mirror them.

This conceptual grounding also clarifies the philosophical status of Occam's Razor. Parsimony is not an arbitrary methodological preference but a necessary condition for viable knowledge systems. A theory cluttered with parameters and special cases is not merely inelegant; it is structurally incapable of the predictive compression that allows knowledge systems to extend beyond their training data. The principle of parsimony, understood through this lens, is a constraint imposed by the very nature of viable inquiry.

### 3.3 The Triangulation Method

No single indicator is immune to interpretive bias. Therefore, robust diagnosis of brittleness requires triangulation across independent baselines. This protocol provides a concrete method for achieving pragmatic objectivity.

**Baseline 1: Comparative-Historical Analysis:** We compare system metrics against contemporaneous peers with similar technological, resource, and environmental constraints. For example, 17th-century France exhibited higher excess mortality from famine than England, not because of worse climate, but because of a more brittle political-economic system hindering food distribution. The baseline is what was demonstrably achievable at the time.

**Baseline 2: Diachronic Trajectory Analysis:** We measure direction and rate of change within a single system over time. A society where life expectancy is falling, or a research program where the share of work devoted to ad-hoc patches vs. novel predictions is increasing, is exhibiting rising brittleness regardless of its performance relative to others.

**Baseline 3: Biological Viability Thresholds:** Some thresholds are determined by non-negotiable biological facts. A society with Total Fertility Rate sustainably below 2.1 is, by definition, demographically unviable without immigration. A system generating chronic malnutrition in over 40% of its population is pushing against fundamental biological limits.

Diagnosis requires convergent baselines: e.g., rising mortality (diachronic), peer underperformance (comparative), and biological thresholds. This parallels climate science's multi-evidence convergence, achieving pragmatic objectivity for comparative evaluations.

**Avoiding Circularity:** Defining "first-order outcomes" requires anchoring in physical, measurable consequences that are theory-independent where possible. For social systems, Tier 1 metrics (mortality, morbidity, resource depletion) can be measured without presupposing the values being tested. For knowledge systems, predictive success and unification can be tracked through citation patterns and cross-domain applications. The key is triangulation: multiple independent metrics converging provides confidence that diagnoses reflect underlying brittleness rather than observer bias.

#### 3.3.1 Why Viability Is Not Just "Evidence" Renamed

An immediate objection: if viability is measured by successful predictions, explanatory unification, and effective interventions, how does this differ from standard accounts of evidential superiority? Is systemic brittleness just a new vocabulary for an old idea?

The distinction is this: standard evidential accounts focus on the propositional level, assessing which claims are supported by which observations. Viability analysis operates at the systems level, tracking the costs a knowledge network generates when its principles are collectively applied over time. These costs include but exceed predictive failure.

**Maintenance costs:** Resources spent patching, hedging, explaining away anomalies, even when predictions technically succeed. A theory can accurately predict outcomes while requiring ever-increasing ad hoc modifications to do so. The Ptolemaic system maintained predictive adequacy for centuries, but the brittleness lay not in failed predictions but in the proliferating machinery required to generate those predictions.

**Coercive costs:** Resources spent enforcing compliance when a system generates resistance from its own practitioners or subjects. When a paradigm suppresses dissent through institutional power rather than persuasion, when it requires loyalty oaths or purges rather than open inquiry, these costs signal brittleness invisible to purely evidential analysis.

**Opportunity costs:** Paths not taken, problems not addressed, because the system's structure forecloses them. A research program can have strong evidential support for what it investigates while systematically avoiding domains where it would perform poorly. The inability to even formulate certain questions is a cost standard evidential accounting cannot capture.

A system can therefore have strong local evidential support while accumulating unsustainable systemic costs. Ptolemaic astronomy made adequate predictions for centuries. Its brittleness lay not in predictive failure but in the proliferating machinery required to maintain those predictions. Evidence, in the standard sense, supported it. Viability analysis reveals the hidden costs that standard evidential accounting misses.

The relationship is therefore not identity but inclusion. Evidence (predictive success) is one input to viability. But viability tracks the total cost of maintaining a system as a going concern, including costs that never appear in confirmation theory because they are not about the truth of individual propositions but about the sustainability of the entire knowledge-producing enterprise.

### 3.4 The Conceptual Distinction Between Productive and Coercive Costs

A persistent objection is that classifying spending as "productive" versus "coercive" requires the very normative commitments the framework aims to ground, leading to circularity. The framework resolves this by distinguishing them not by their intrinsic nature but by their observable function within the system over time. The distinction is conceptual and diagnostic, not based on a priori definitions.

**Coercive overheads** are identifiable as expenditures that function to suppress the symptoms of a system's underlying brittleness. Their signature is a pattern of diminishing returns: escalating investment is required merely to maintain baseline stability, and this spending correlates with stagnant or worsening first-order outcomes (such as public health or innovation). Such costs do not solve problems but manage the dissent and friction generated by them.

**Productive investments**, in contrast, are expenditures that demonstrably reduce first-order costs and exhibit constant or increasing returns. They address root causes of brittleness and tend to generate positive spillovers, enhancing the system's overall viability.

The classification, therefore, emerges from analyzing the dynamic relationship between resource allocation and systemic health, not from a static definition of "coercion." We ask: does this expenditure pattern reduce the system's total costs, or does it function as a secondary cost required to mask the failures of the primary system? This conceptual distinction allows for a non-circular, trajectory-based diagnosis of systemic health.


## 4. The Emergent Structure of Objectivity

With diagnostic methodology in hand (Section 3), we can now turn to the question these tools enable us to answer: what structure emerges when we systematically eliminate high-brittleness systems? The preceding sections established the methods for identifying failure. This section addresses the framework's core theoretical contribution: building the theory of objectivity that systematic failure-analysis makes possible. We show how the logic of viability selects for knowledge system evolution and drives convergence toward an emergent, objective structure: the Apex Network.

### 4.1 A Negative Methodology: Charting What Fails

Our account of objectivity is not the pursuit of a distant star but the painstaking construction of a reef chart from the empirical data of shipwrecks. It begins not with visions of final truth, but with our most secure knowledge: the clear, non-negotiable data of large-scale systemic failure. Following Popperian insight (Popper 1959), our most secure knowledge is often of what is demonstrably unworkable. While single failed experiments can be debated, entire knowledge system collapse (descent into crippling inefficiency, intellectual stagnation, institutional decay) provides clear, non-negotiable data.

Systematic failure analysis builds the Negative Canon: an evidence-based catalogue of invalidated principles distinguishing:

**Epistemic Brittleness:** Causal failures (scholastic physics, phlogiston) generating ad-hoc patches and predictive collapse.

**Normative Brittleness:** Social failures (slavery, totalitarianism) requiring rising coercive overheads to suppress dissent.

Charting failures reverse-engineers viability constraints, providing external discipline against relativism. This eliminative process is how the chart is made, mapping hazards retrospectively to reveal the safe channels between them. Progress accrues through better hazard maps.

### 4.2 The Apex Network: Ontological and Epistemic Status

As the Negative Canon catalogs failures, pragmatic selection reveals the contours of the Apex Network. The key distinction is two-level. Epistemologically, our claim is coherentist-externalist: beliefs are warranted within networks whose overall performance remains resilient under pragmatic pressure. Metaphysically, our claim is structural realist: the constraints producing that pressure are mind-independent.

This distinction matters for the reviewer's central worry. Saying that brittle systems are misaligned with causal structure does not collapse justification into one-by-one correspondence testing. It states the external condition under which coherent networks survive. Coherence still does the internal justificatory work. Constraint pressure explains why some coherent networks remain stable and others do not.

We therefore define the Apex Network as the maximally shared load-bearing structure across viable networks, not as a single complete theory currently available to us. Formally, it is the intersection of the structural elements that persist across low-brittleness systems under shared constraints. In that sense, if there is any non-trivial convergence at all, there is an Apex structure, even if it is partial and historically expanding.

This framing also sharpens the multiple-peaks objection. Multiple local optima can exist. We grant that directly. But the objection defeats our view only if viable peaks are entirely disjoint. That stronger claim is unlikely in tightly constrained domains because any system that remains viable must solve common load-bearing problems. Distinct peaks can differ at the frontier while sharing a substantial core.

We also grant a genuine limitation: in weakly constrained domains, the shared core may be thin. This is not a defect in the framework. It is a diagnosis of underdetermination. The framework should not promise thick convergence where constraint structure does not support it.

The account remains non-circular because it gives independent criteria for distinguishing durable convergence from local lock-in. Local lock-in tends to show path dependence without independent rediscovery, escalating maintenance or coercive costs, and poor extensibility to new domains. Durable convergence tends to show independent rediscovery, stable low brittleness across institutional changes, and cross-domain transfer without proliferating defensive patches.

This structure lets us answer the isolation objection in the right scope. Isolation is not denied as a synchronic possibility for any single coherent web. It is denied as a stable diachronic outcome for public networks operating under shared constraints. Over time, systems that do not track those constraints accumulate costs and lose shareability.

The conclusion is modest but strong. The Apex Network is objective because it is constraint-determined, not because consensus creates it. It is fallibly known because access occurs through historical elimination, not direct inspection. That is why realism at the metaphysical level and coherentism at the epistemic level are complementary in this framework rather than rivals.


#### 4.2.1 Formal Characterization

Drawing on network theory (Newman 2010), we can formally characterize the Apex Network as:

A = ∩{W_k | V(W_k) = 1}

Where A = Apex Network, W_k = possible world-systems (configurations of predicates), V(W_k) = viability function (determined by brittleness metrics), and ∩ = intersection (common structure across all viable systems).

The intersection of all maximally viable configurations reveals their shared structure. This shared structure survives all possible variations in historical path: the emergent, constraint-determined necessity arising from how reality is organized.

A coherent system detached from the Apex Network isn't merely false but structurally unstable. It will generate rising brittleness until it either adapts toward the Apex Network or collapses. Coherence alone is insufficient because reality's constraints force convergence.


#### 4.2.2 Cross-Domain Convergence and Pluralism

Cross-domain predicate propagation drives emergence: when Standing Predicates prove exceptionally effective at reducing brittleness in one domain, pressure mounts for adoption in adjacent domains. Germ theory's success in medicine pressured similar causal approaches in public health and sanitation. This successful propagation forges load-bearing, cross-domain connections constituting the Apex Network's emergent structure.

This process can be conceptualized as a fitness landscape of inquiry. Peaks represent low-brittleness, viable configurations (e.g., Germ Theory), while valleys and chasms represent high-brittleness failures catalogued in the Negative Canon (e.g., Ptolemaic System). Inquiry is a process of navigating this landscape away from known hazards and toward resilient plateaus.


#### 4.2.3 Discovery Without Blueprint

Two claims can both be true without tension. First, the Apex Network is not a transcendent blueprint, a Platonic object, or a teleological endpoint built into history. Second, given a fixed constraint structure, viable systems will share determinate structural features regardless of which community discovers them first. The first claim blocks metaphysical inflation. The second claim blocks relativism.

So the right modal reading is conditional. Relative to a world with constraint profile $C$, there is a determinate apex structure $A(C)$. Change the constraints, and a different structure may be selected. This is enough objectivity for epistemology, without requiring a timeless realm of propositions.


### 4.3 A Three-Level Framework for Truth

This emergent structure grounds a fallibilist but realist account of truth. It clarifies a documented tension in Quine's thought between truth as immanent to our best theory and truth as a transcendent regulative ideal (Tauriainen 2017). Our framework shows these are not contradictory but two necessary components of a naturalistic epistemology. Crucially, in this model, a demonstrated track record of low systemic brittleness functions as our primary fallible evidence for a system's alignment with the Apex Network; it is not, itself, constitutive of that alignment.

A belief is not justified merely because it coheres with a Consensus Network. Coherence with a flat-earth network or a phlogiston network provides no warrant. What matters is coherence with a network that has itself earned epistemic authority through sustained success under external, non-discursive constraints.

This is the externalist move that distinguishes Systemic Externalism from classical coherentism. Coherence remains necessary: isolated beliefs lack inferential integration. But coherence is not sufficient. The network must have demonstrated resilience through low brittleness, successful interventions, minimal coercive overhead, and adaptability to novel problems. Justification is earned through historical performance, not stipulated by internal agreement.

A belief can therefore be justified yet false, if it coheres with a network that has, at that time, the best available track record. And a belief can be unjustified yet true, if it coheres only with a high-brittleness network or none at all. This preserves the standard distinction between justification (a property of beliefs relative to networks) and truth (a property indexed to the Apex Network's objective structure).

* **Level 3: Contextual Coherence.** The baseline status for any claim. A proposition is coherent within a specific Shared Network, regardless of that network's long-term viability. This level explains the internal rationality of failed or fictional systems, but the framework's externalist check, the assessment of systemic brittleness, prevents this from being mistaken for higher warrant.
* **Level 2: Warranted Assertion.** The highest practical epistemic status available to finite inquirers. A belief is warranted for assertion as true when it is certified by a Consensus Network with a demonstrated low-brittleness track record. This preserves the standard analytic distinction: beliefs and assertions are justified, truths are not. To doubt a claim at this level, without new evidence of rising brittleness, is to suspend trust in the most resilient inquiry structures currently available.
* **Level 1: Objective Truth.** The regulative ideal of the process. A proposition is objectively true if its principles are part of the real, emergent Apex Network: the objective structure of viable solutions. In information-theoretic terms, this represents optimal computational closure, where a system's enacted boundaries (its Markov Blankets) achieve maximum alignment with the causal constraints of the environment, minimizing information leakage to the theoretical minimum. While this structure is never fully mapped, it functions as the formal standard that makes our comparative judgments of "more" or "less" brittle meaningful.

This layered framework avoids a simplistic "Whig history" by recognizing that Level 2 warrant is historically situated. Newtonian mechanics earned that status by being a maximally low-brittleness system for its problem-space for over two centuries. Its replacement by relativity does not retroactively invalidate the earlier warrant. It shows the evolutionary process at work, where an expanding problem-space revealed constraints requiring a more viable successor.

### 4.3.1 The Hard Core: Functional Entrenchment and Pragmatic Indispensability

The three-level framework reveals how propositions do not merely "correspond" to truth as an external standard but become constitutive of truth itself through functional transformation and entrenchment.

Our framework provides robust, naturalistic content to truth-attributions: to assert that $P$ is true at Level 2 is to treat $P$ as warranted by a low-brittleness Consensus Network; to say $P$ is objectively true (Level 1) is to say $P$ aligns with the emergent, constraint-determined Apex Network. Truth is what survives systematic pragmatic filtering. The predicate "is true" tracks functional role within viable knowledge systems, not correspondence to a Platonic realm.

**From Validated Data to Constitutive Core: The Progression**

A proposition's journey to becoming truth itself follows a systematic progression through functional transformation:

1. **Initial Hypothesis (Being-Tested):** The proposition begins as a tentative claim within some Shared Network, subject to coherence constraints and empirical testing. It is data to be evaluated.

2. **Validated Data (Locally Proven):** Through repeated application without generating significant brittleness, the proposition earns trust. Its predictions are confirmed; its applications succeed. It transitions from hypothesis to validated data, something the network can build upon.

3. **Standing Predicate (Tool-That-Tests):** The proposition's functional core, its reusable predicate, is promoted to Standing Predicate status. It becomes conceptual technology: a tool for evaluating new phenomena rather than something being evaluated. "...is an infectious disease" becomes a diagnostic standard, not a claim under test.

4. **Convergent Core Entry (Functionally Unrevisable):** As rival formulations are relegated to the Negative Canon after generating catastrophic costs, the proposition migrates to the Convergent Core. Here it achieves Level 2 status: Warranted Assertion.

5. **Hard Core (Constitutive of Inquiry Itself):** In the most extreme cases, a proposition becomes so deeply entrenched that it functions as a constitutive condition for inquiry within its domain. This is Quine's hard core, the principles so fundamental that their removal would collapse the entire edifice.

**Quine's Hard Core and Functional Entrenchment**

Quine famously argued that no claim is immune to revision in principle, yet some claims are practically unrevisable because revising them would require dismantling too much of our knowledge structure. Our framework explains this tension through the concept of functional entrenchment driven by bounded rationality (March 1978).

A proposition migrates to the hard core not through metaphysical necessity but through pragmatic indispensability. The costs of revision become effectively infinite:

- **Logic and Basic Mathematics:** Revising logic requires using logic to evaluate the revision (infinite regress). Revising basic arithmetic requires abandoning the conceptual tools needed to track resources, measure consequences, or conduct any systematic inquiry. These exhibit maximal brittleness-if-removed.

- **Thermodynamics:** The laws of thermodynamics undergird all engineering, chemistry, and energy policy. Revising them would invalidate centuries of validated applications and require reconstructing vast swaths of applied knowledge. The brittleness cost is astronomical.

- **Germ Theory:** After decades of successful interventions, public health infrastructure, medical training, and pharmaceutical development all presuppose germ theory's core claims. Revision would collapse these systems, generating catastrophic first-order costs.

**The Paradox Resolved: Fallibilism Without Relativism**

How can we be fallibilists who acknowledge all claims are revisable in principle, while simultaneously treating hard core propositions as effectively unrevisable in practice? The resolution: "revisable in principle" means if we encountered sufficient pragmatic pushback, we would revise even hard core claims. For hard core propositions, the threshold is extraordinarily high but not infinitely high. This makes the framework naturalistic rather than foundationalist. Hard core status is functional achievement, not metaphysical bedrock.

Truth is not discovered in a Platonic realm but achieved through historical filtering. Propositions become true by surviving systematic application without generating brittleness, migrating from peripheral hypotheses to core infrastructure, becoming functionally indispensable to ongoing inquiry, and aligning with the emergent, constraint-determined Apex Network.

This resolves the classical tension between Quine's holism (all claims are revisable) and the practical unrevisability of core principles: both describe different aspects of the same evolutionary process through which propositions earn their status by proving their viability under relentless pragmatic pressure.

### 4.3.2 Why Quine's Architecture Matters: Systems-Level Implications

**Describing Quine's Architecture: Systems-Level Implications**

Quine's "Web of Belief" (Quine 1951, 1960) provides the essential architecture for our framework: holistic structure, pragmatic revision in response to recalcitrant experience, and overlapping coordination across multiple agents. The Quinean architecture is not merely compatible with our framework but essential to it—systemic brittleness, pragmatic pushback, and convergence all require this structure. Our contribution is to describe the observable patterns that emerge when this Quinean process operates at scale across public knowledge systems, and to provide formalized diagnostics for assessing the structural health these processes generate.

First, Quine's recalcitrant experience, aggregated across populations and time, generates observable patterns of systemic cost. These accumulated costs—what we track as brittleness—provide the externalist filter grounding knowledge in mind-independent reality. This relentless, non-discursive feedback prevents knowledge systems from floating free of constraints.

Second, Quine's pragmatic revision mechanism, operating through bounded rationality (March 1978), explains entrenchment: propositions migrate to the core because they have demonstrated immense value in lowering systemic brittleness. This functions as systemic caching—proven principles are preserved to avoid re-derivation costs. Conservation of Energy became entrenched after proving indispensable across domains, its revision now prohibitively expensive.

These patterns, already present in Quine's framework, become diagnosable through brittleness metrics. The framework provides tools to assess how resilient cores form through systematic, externally-validated selection (Carlson 2015). In doing so, it resolves a documented tension in Quine's thought between truth as immanent to our best theory and truth as a transcendent regulative ideal (Tauriainen 2017). Our three-level framework shows these are not contradictory but two necessary components of a naturalistic epistemology. Core principles achieve Level 2 warrant through systematic pragmatic filtering, while the Apex Network (Level 1) functions as the regulative structure toward which theories converge through constraint-driven selection.

**The Individual Revision Mechanism: Conscious Awareness as Feedback Interface**

While the above describes systems-level patterns, a complete account requires explaining the micro-level mechanism: how do individual agents actually revise their personal webs of belief? Quine established that recalcitrant experience forces adjustment, but left the phenomenology and strategy selection implicit. Our disposition-based account makes this process explicit.

As established in Section 2.1, our framework identifies belief with the conscious awareness of one's dispositions to assent. This awareness functions as the natural feedback mechanism that enables deliberate revision. When an agent holds a disposition generating pragmatic costs, the revision cycle proceeds through several stages. First, dispositional conflict emerges: the agent's disposition produces failed predictions, wasted effort, coordination failures, or other measurable costs. Second, conscious recognition occurs as the agent becomes aware that holding this disposition correlates with these costs, either through direct experience or social signaling. Third, this awareness creates motivational pressure (the discomfort of cognitive dissonance, frustration at repeated failure, or pragmatic motivation to reduce costs). Fourth, the agent consciously explores alternative dispositions compatible with their core web commitments, testing adjustments mentally or through limited trials. Fifth, through repeated practice and positive reinforcement, a new disposition stabilizes, replacing the costly pattern. Finally, the agent's conscious model of their own belief system updates to reflect this revision, stored in memory for future reference.

This cycle operates at the individual level but drives macro-level convergence when aggregated across populations. Multiple agents independently experiencing costs from similar dispositions will independently revise toward lower-cost alternatives. When these revisions are communicated through assertion and coordinated through social exchange, patterns of convergence emerge, not through central planning or mysterious coordination but through distributed pragmatic optimization. Each agent, responding to locally experienced costs, makes adjustments that happen to align with others' adjustments because they are all responding to the same underlying constraint structure.

The role of memory deserves emphasis. Actual belief systems require multiple forms of memory operating simultaneously. Dispositional memory maintains stable patterns of assent over time, preventing constant drift. Revision memory stores awareness of past adjustments and their outcomes, enabling learning from history. Cost memory accumulates experience of which dispositional patterns generate brittleness, functioning as a pragmatic evaluation metric. And coordination memory preserves learned patterns of successful social alignment, facilitating efficient future cooperation.

Conscious awareness of dispositions includes awareness of their history: we remember not just what we believe but how beliefs have changed and what costs prompted those changes. This historical awareness enables belief systems to function as evolving, learning systems. An agent who remembers that adopting disposition D reduced costs compared to previous disposition D' is better positioned to make future revisions, creating a ratcheting effect where successful adjustments are preserved while failures are discarded.

Why does this matter for convergence on the Apex Network? The Apex is not a pre-existing target that agents consciously aim for. It is the emergent structure that must arise when millions of agents, each consciously aware of their own dispositions and the costs they generate, independently revise toward lower-brittleness patterns. Convergence is explained not by mysterious coordination or shared access to truth but by the simple fact that reality's constraint structure punishes certain dispositional patterns and rewards others, and conscious agents can detect and respond to this feedback. The conscious awareness component enables systematic belief revision rather than random drift, explaining deliberate adjustment, learning from experience, and the directedness of convergence toward viability.

Quine's web, understood as this active cognitive system capable of deliberate self-modification, provides the foundation for our systems-level analysis. Pragmatic pushback is not an abstract force but is experienced by individuals as the costs of holding unviable dispositions, motivating the revision process that drives knowledge toward the Apex Network.

**The Phenomenology of Revision: Emotions as System Diagnostics**

The revision cycle described above has a phenomenological dimension that makes epistemic processes consciously accessible. Emotions function as the lived experience of epistemic states: anxiety manifests as high prediction error—the discomfort when the world persistently fails to match expectations; frustration signals accumulating costs from misaligned dispositions; joy or "flow" marks computational closure—the satisfaction when actions perfectly predict outcomes. From this perspective, emotions are not noise to be filtered but data about the structural integrity of our internal models. They are the phenomenological dashboard of the epistemic engine, making the costs of misalignment consciously accessible and thereby motivating revision.

**Agency as the Variation Engine**

While the network provides the selection mechanism through pragmatic filtering, the individual agent is the source of variation. 'Free will,' in this framework, is the capacity to generate novel functional propositions—heresies that challenge current consensus. The 'genius' or 'reformer' is the agent willing to propose a new, potentially lower-brittleness predicate and bear the high social cost of being an early adopter before parallel discovery forces convergence. This recovers individual agency within a systems framework: the constraint landscape determines what survives, but individuals determine what gets tested. Innovation is not centrally planned but emerges from distributed experimentation, with reality serving as the ultimate arbiter of viability.

This individual revision cycle is the micro-engine of macro-level convergence. The process is not centrally coordinated, nor does it require that agents aim for a shared truth. Rather, convergence emerges as a result of distributed pragmatic optimization. Millions of agents, each independently experiencing and seeking to reduce the costs generated by their own unviable dispositions, make local adjustments. Because they are all interacting with the same underlying constraint structure of reality, their individual solutions are independently pushed toward the same basin of attraction. The shared constraint landscape is the coordinating force. Unviable patterns are punished with costs for everyone who adopts them, while viable patterns are rewarded with lower costs. Social communication accelerates this process by allowing agents to learn from the costly experiments of others, but the fundamental driver of convergence is the uniform selective pressure exerted by a shared, mind-independent reality.

This three-level truth framework describes the justificatory status of claims at a given moment. Over historical time, pragmatic filtering produces a discernible two-zone structure in our evolving knowledge systems.

### 4.3.3 On the Nature of Truth in This Framework

This framework adopts a deflationary stance toward truth. To call a proposition "true" is not to ascribe a substantive property, whether correspondence, coherence, or otherwise. It is to endorse it, to license its use in inference, to treat it as something to build upon. "P is true" does no more than assert P with emphasis.

What is substantive in this framework is the constraint landscape and the costs it generates. These are real and mind-independent. A brittle system that collapses under pragmatic pressure does so because it misaligns with genuine features of reality, not because we collectively decided to stop believing in it. But "truth" names our endorsement of what survives pragmatic filtering, not a relation between propositions and facts that we check.

This dissolves an apparent tension. When we speak of viability under constraints or describe the Apex Network as the constraint-determined structure our systems are filtered toward, we are not smuggling in correspondence theory. The constraint landscape is real. The costs are real. But the word "true" does not name a correspondence relation to these realities. It names the status we grant to what survives them.

Consider an analogy: biological fitness is real. The environmental pressures that select for certain adaptations over others are mind-independent facts about the world. But to call an organism "fit" is not to describe a correspondence relation between the organism and some Platonic fitness form. It is to say the organism survived selection. Similarly, to call a proposition "true" is to say it survived pragmatic filtering. The filtering mechanism is real and objective; the predicate "true" marks what passed through.

Realism, in this framework, applies to constraints. Deflationism applies to truth. These are compatible: we can be realists about what selects without being correspondence theorists about what "true" means.

This also clarifies the status of the Apex Network. It is not a realm of abstract propositions awaiting discovery. It is the emergent configuration of viable functional structures that any knowledge system operating under shared constraints must converge toward. Different communities may articulate these structures using different vocabularies, different formalisms, even different ontologies. What makes them converge on the "same" network is not that they are matching a pre-existing propositional template but that they are being shaped by the same constraint topology. The Apex Network is real in the way an evolutionary attractor basin is real: not as a thing but as a pattern that constraint-driven processes inevitably produce.

This deflationary approach also explains why the framework can remain neutral on classic debates about the nature of truth while still providing robust epistemic guidance. We need not settle whether truth is correspondence or coherence or pragmatic utility to diagnose that a system accumulating high brittleness is becoming epistemically unreliable. The costs are the signal. Truth-talk is how we mark what made it through the filter.

### 4.3.4 Constraint Fields and Attractor Basins

When we speak of "alignment with constraint structure" or "viability under constraints," we are not invoking correspondence theory's picture of beliefs being checked against reality. Instead, we are describing how constraint structure functions as a **field** that generates **attractor basins**.

Think of constraint structure not as a fixed template to be matched, but as a force field. Just as gravity creates attractor basins in physical space (valleys where objects naturally accumulate), pragmatic constraints create attractor basins in possibility space. Knowledge systems that impose lower operational costs (lower $P(t)$, $C(t)$, $M(t)$) occupy deeper basins. Systems that impose higher costs occupy shallower basins or unstable equilibria.

"Alignment" in this framework means: the system has settled into a stable attractor basin under the constraint field. "Misalignment" means: the system occupies an unstable configuration that will be displaced when sufficient perturbation occurs.

There is no separate "checking" operation where we compare our beliefs to reality. The constraint field is reality's causal structure, and belief systems naturally migrate toward viable configurations through pragmatic selection. Failed predictions, rising maintenance costs, and institutional friction are all manifestations of the field's pressure pushing the system out of an unstable basin.

The Apex Network is not a Platonic form to be approximated. It is the deepest, most stable attractor basin in the constraint field—the configuration that multiple independent systems converge toward because it imposes minimum operational costs under the full range of constraints.

This is deflationary: we are not ascribing a substantive "truth" property to beliefs that align with reality. We are simply observing that certain configurations survive pragmatic filtering while others do not. The survivors define the attractor basin we call the Apex Network.

### 4.4 The Evolving Structure of Knowledge: Convergent Core and Pluralist Frontier

The historical process of pragmatic filtering gives our evolving **Consensus Networks** a discernible structure, which can be understood as having two distinct epistemic zones. This distinction is not about the nature of reality itself, but describes the justificatory status of our claims at a given time.

#### 4.4.1 The Convergent Core

This represents the load-bearing foundations of our current knowledge. It comprises domains where the relentless pressure of pragmatic selection has eliminated all known rival formulations, leaving a single, or functionally identical, set of low-brittleness principles. Principles reside in this core, such as the laws of thermodynamics or the germ theory of disease, not because they are dogmatically held or self-evident but because all tested alternatives have been relegated to the **Negative Canon** after generating catastrophically high systemic costs. While no claim is immune to revision in principle, the principles in the **Convergent Core** are functionally unrevisable in practice, as doing so would require dismantling the most successful and resilient knowledge structures we have ever built. A claim from this core achieves the highest degree of justification we can assign, approaching our standard for Objective Truth (Level 1). For example, the **Standing Predicate** `...is a pathogen` is now functionally indispensable, having replaced brittle pre-scientific alternatives.

#### 4.4.2 The Pluralist Frontier: Multiple Peaks, Shared Structure

A central objection to any convergence thesis is underdetermination: multiple theories may be compatible with all available evidence. This framework accepts underdetermination as genuine pluralism. Where pragmatic constraints underdetermine solutions, multiple stable configurations can survive filtering. We should expect persistent theoretical diversity, not forced convergence. But this does not entail relativism. If viable systems share any propositions at all, a maximally shared subset exists by definition. The substantive question is not whether an intersection exists, but how thick it is.

The most useful image is a mountain range rather than a single peak. Rival theories can occupy different peaks while standing on the same geological base. The Apex Network is that shared base and ridge structure: the load-bearing commitments that any viable peak must inherit. In tightly constrained domains, this shared structure should be substantial. In weakly constrained domains, it may remain thin for long periods. Both outcomes are compatible with the framework.

Analogy: eyes evolved independently over forty times across the animal kingdom. Many designs work. Compound eyes, camera eyes, pinhole eyes. These are distinct local peaks, different solutions to the problem of vision. But every visual system must solve the same problems: focusing light, detecting wavelengths, transmitting signals to a processing center. The "shared structure" is not one correct design. It is what any solution to those constraints must include. This pluralism is not relativism because all viable peaks share necessary constraint-determined structure and are sharply distinguished from demonstrably failed systems in the Negative Canon.

A reviewer may worry: if multiple local peaks exist, and we are "trapped" on a local peak rather than the global maximum, doesn't this vindicate the isolation objection? Are we simply coherent within our peak while other peaks are equally coherent within theirs?

No, for three reasons:

**First, constraint sharing:** All viable peaks must share load-bearing structural elements required by the constraints. The eye evolution analogy makes this concrete: forty independent solutions to vision exist, but all share necessary features (light-focusing mechanism, photoreceptor array, signal transmission pathway). "Multiple peaks" does not mean "arbitrary diversity." It means: constraint structure permits variation within boundaries, but strictly rules out configurations that violate necessary requirements.

**Second, objective ranking:** Local peaks are not epistemically symmetric. They are ranked by measurable brittleness. A knowledge system trapped on a high-brittleness local peak will exhibit rising $P(t)$, $C(t)$, and $M(t)$ relative to systems on low-brittleness peaks. This cost asymmetry is objective—it manifests as publication patterns, resource allocation, institutional friction—and provides asymmetric warrant. We are not "equally justified" within our respective peaks.

**Third, attractor basin depth:** The "global peak" metaphor misleads. Constraint fields do not always have a single global maximum. They have attractor basins of varying depth. What matters is not whether we are on the global peak, but whether we are in a deep, stable basin versus a shallow, unstable one. Brittleness metrics diagnose basin depth. Systems in deep basins (low brittleness) earn higher warrant than systems in shallow basins (high brittleness), even if both are "coherent" internally.

This yields constrained pluralism, not relativism: where constraints permit multiple stable solutions, we accept genuine diversity. But that diversity is bounded by shared structural requirements and ranked by objective performance metrics.

An important qualification concerns apparent ties. If two frontier peaks exhibit comparable brittleness over the currently tested range, the framework does not declare one "more true" by stipulation. It assigns provisional warrant parity at Level 2. The proper response is experimental and methodological: design new tests at the edges where the systems diverge, expand the problem-space, and track which framework absorbs new pressure with lower additional cost.

The Pluralist Frontier therefore describes domains where multiple low-brittleness configurations coexist. Different interpretations of quantum mechanics, for instance, may occupy different peaks while sharing the mathematical structure and predictive apparatus that any viable quantum theory requires. Their deepest disagreement is real, but it sits at the frontier, not in the shared predictive core. Competing models of consciousness may differ in ontological commitments while converging on the functional architecture that any adequate theory must capture. Pluralism about the periphery is compatible with convergence on the core.

**Against Relativism: The Constrained Pluralism Argument**

The framework's acceptance of multiple viable configurations invites relativism charges. Here is the response in compressed form:

1. **Relativism claim (thin version):** If internal coherence were the only standard, rival coherent systems would be epistemically symmetric.

2. **Constraint reply:** In practice, coherent systems face asymmetric pressure. Some require escalating ad hoc repair and insulation to remain operational, while others remain resilient and extend smoothly to new domains.

3. **Cost asymmetry is not constituted by agreement:** The costs that diagnose brittleness (maintenance burden, coercive overhead, foreclosed opportunities) are objective resource drains imposed by misalignment with constraints, not determined by social consensus.

4. **Negative Canon provides boundary:** The space of viable theories is not "anything goes" but strictly bounded by what has demonstrably failed. Phlogiston chemistry is not a "viable contender" but a relegated failure in the Negative Canon.

5. **Pluralism within constraints:** Multiple peaks can exist where constraints underdetermine solutions (e.g., QM interpretations), but all must share necessary load-bearing structure.

6. **Conclusion:** This yields constrained pluralism rather than symmetry-based relativism. Rival systems are ranked by objective, measurable performance under shared constraints.

This constrained pluralism must be sharply distinguished from full-blooded relativism. An immediate worry: if justification is relative to a Consensus Network, and multiple networks exist, does this collapse into "different frameworks, different truths"?

No. The framework implies that networks are not epistemically equal. Their authority is earned differentially through historical performance. A network with low brittleness, successful interventions across domains, and minimal coercive overhead has earned more epistemic authority than one requiring constant patching, coercive enforcement, and ad hoc modification.

Disagreement between networks is therefore not symmetric. When a low-brittleness network (modern epidemiology) conflicts with a high-brittleness network such as miasma theory circa 1850, we do not shrug and say "different frameworks, different truths." We say: one has earned the right to our credence through demonstrated viability; the other has accumulated costs that disqualify it.

This is not relativism but fallibilist externalism. Justification is relative to a network, but networks are ranked by objective, measurable performance under shared constraints. The ranking is itself fallible and revisable. But at any given time, it provides non-arbitrary grounds for preferring one system's deliverances over another's.

The frontier is not an "anything goes" zone but a highly restricted space strictly bounded on all sides by the Negative Canon. A system based on phlogiston is not a "viable contender" on the frontier of chemistry but a demonstrably failed research program relegated to the canon of what does not work. This pluralism is therefore a sign of epistemic underdetermination: a feature of our map's current limitations, not reality's supposed indifference.

This position resonates with pragmatist accounts of functional pluralism (Price 1992), which treat different conceptual frameworks as tools whose legitimacy is determined by their utility within a specific practice. Within this frontier, the core claims of each viable competing system can be granted Level 2 warrant. This is also the zone where non-epistemic factors, such as institutional power or contingent path dependencies, can play their most significant role, sometimes artificially constraining the range of options explored or creating temporary monopolies on what is considered justified.

### 4.5 Illustrative Cases of Convergence and Brittleness

The transition from Newtonian to relativistic physics offers a canonical example of this framework's diagnostic application. After centuries of viability, the Newtonian system began to accumulate significant systemic costs in the late 19th century. These manifested as first-order predictive failures, such as its inability to account for the perihelion of Mercury, and as rising conceptual debt in the form of ad-hoc modifications like the Lorentz-FitzGerald contraction hypothesis. This accumulating brittleness created what Kuhn (1962) termed a "crisis" state preceding paradigm shifts. The Einsteinian system proved a more resilient solution, reducing this conceptual debt and substantially lowering the systemic costs of inquiry in physics.

A more contemporary case can be found in the recent history of artificial intelligence, which illustrates how a brittleness assessment might function in real time. The periodic "AI winters" can be understood as the collapse of high-brittleness paradigms, such as symbolic AI, which suffered from a high rate of ad-hoc modification when faced with novel challenges. While the subsequent deep learning paradigm proved a low-brittleness solution for many specific tasks, it may now be showing signs of rising systemic costs. These can be described conceptually as, for example, potentially unsustainable escalations in computational and energy resources for marginal performance gains, or an accelerating research focus on auxiliary, post-hoc modifications rather than on foundational architectural advances. This situation illustrates the **Pluralist Frontier** in action, as rival architectures might now be seen as competing to become the next low-brittleness solution.

### 4.6 Navigating the Landscape: Fitness Traps, Path Dependence, and the Role of Power

An evolutionary model of knowledge must account for the complexities of history, not just an idealized linear progress. The landscape of viability is not smooth: knowledge systems can become entrenched in suboptimal but locally stable states, which we term "fitness traps" (Wright 1932). This section clarifies how the framework incorporates factors like path dependence and institutional power not as external exceptions but as core variables that explain these historical dynamics.

The model's claim is not deterministic prediction but probabilistic analysis: beneath the surface-level contingency historians rightly emphasize, underlying structural pressures create statistical tendencies over long timescales. A system accumulating brittleness is not fated to collapse on a specific date but becomes progressively more vulnerable to contingent shocks. The model thus complements historical explanation by offering tools to understand why some systems prove more resilient than others.

A system can become locked into a high-brittleness fitness trap by coercive institutions or other path-dependent factors. A slave economy, for instance, is a classic example. While objectively brittle in the long run, it creates institutional structures that make escaping the trap prohibitively costly in the short term (Acemoglu and Robinson 2012). The framework's key insight is that the exercise of power does not negate a system's brittleness; rather, the costs of maintaining that power become a primary indicator of it. This power manifests in two interrelated ways. First is its defensive role: the immense coercive overheads required to suppress dissent and manage internal friction are a direct measure of the energy a system must expend to resist the structural pressures pushing it toward collapse.

Second, power plays a constitutive role by actively shaping the epistemic landscape itself. Powerful institutions do not merely respond to brittleness defensively; they can construct and enforce the very parameters of a fitness trap. By controlling research funding, defining legitimate problems, and entrenching path dependencies, institutional power can lock a system into a high-brittleness state. This is not an unmeasurable phenomenon; the costs of this constitutive power manifest as diagnosable symptoms, such as suppressed innovation (a low I(t) ratio) and the need for escalating ideological enforcement, which registers as rising Coercive Overheads (C(t)). This pattern of epistemic capture appears across domains: from tobacco companies suppressing health research to colonial knowledge systems that extracted Indigenous insights while denying reciprocal engagement, thereby masking brittleness through institutional dominance. While this can create a temporary monopoly on justification, the framework can still diagnose the system's underlying brittleness. The costs of this constitutive power often manifest as a lack of adaptability, suppressed innovation, and a growing inability to solve novel problems that fall outside the officially sanctioned domain. To detect such hidden brittleness, we can augment C(t) with sub-metrics for innovation stagnation, tracking lags in novel applications or cross-domain extensions relative to comparable systems as proxies for suppressed adaptive capacity. Concretely, innovation lag can be operationalized as: **I(t) = (Novel Applications per Unit Time) / (Defensive Publications per Unit Time)**. When I(t) declines while C(t) remains high, this signals power-induced rigidity masking underlying brittleness. For example, Lysenkoist biology in the Soviet Union showed I(t) approaching zero (no cross-domain applications) while defensive publications proliferated. Over historical time, even the most entrenched systems face novel shocks, where the hidden costs of their power-induced rigidity are typically revealed.

The severity of a fitness trap can be conceptually diagnosed. The work of historians using cliodynamic analysis, for example, is consistent with this view, suggesting that the ratio of a state's resources dedicated to coercive control versus productive capacity can serve as a powerful indicator of systemic fragility. Historical analyses have found that polities dedicating a disproportionately high and sustained share of their resources to internal suppression often exhibit a significantly higher probability of fragmentation when faced with external shocks (Turchin 2003). This provides a way to conceptually diagnose the depth of a fitness trap: by tracking the measurable, defensive costs a system must pay to enforce its power-induced constraints.

Finally, it is necessary to distinguish this high-brittleness fitness trap from a different state: low-brittleness stagnation. A system can achieve a locally stable, low-cost equilibrium that is highly resilient to existing shocks but lacks the mechanisms for generating novel solutions. A traditional craft perfected for a stable environment but unable to adapt to a new material, or a scientific paradigm efficient at solving internal puzzles but resistant to revolutionary change, exemplifies this state. While not actively accumulating systemic costs, such a system is vulnerable to a different kind of failure: obsolescence in the face of a faster-adapting competitor. Diagnosing this condition requires not only a static assessment of current brittleness but also an analysis of the system's rate of adaptive innovation. True long-term viability therefore requires a balance between low-cost stability and adaptive capacity. This evolutionary perspective completes our reef chart, not as a finished map, but as an ongoing process of hazard detection and channel discovery.

### 4.7 The Dynamics of Paradigm Transition

The preceding analysis raises a critical question: if the Apex Network represents a more viable configuration, why do high-brittleness systems often persist for centuries before collapsing? Why do we observe punctuated equilibrium in the history of knowledge systems rather than smooth, continuous adaptation toward lower-brittleness states?

The answer lies in recognizing that knowledge systems are not merely abstract webs of propositions but material and institutional investments. A high-brittleness system generates ongoing costs, as documented through our brittleness metrics. However, abandoning that system requires a massive upfront expenditure: textbooks must be rewritten, practitioners retrained, institutional hierarchies reorganized, and established techniques replaced. This creates two distinct cost streams that determine system stability.

The first is the cost of maintenance: the daily, accumulating expenditure required to keep a brittle system functioning. This includes the resources devoted to patching anomalies (rising $P(t)$), the coercive overhead needed to suppress dissent (rising $C(t)$), and the escalating complexity required to maintain performance (rising $M(t)$). These costs compound over time as each patch generates new anomalies requiring further patches.

The second is the cost of transition: the catastrophic, one-time expenditure required to abandon the existing system and reorganize around a rival configuration. This is not merely intellectual but material and social. Consider the transition from Ptolemaic to Copernican astronomy: it required not just accepting new equations but reorganizing the entire infrastructure of astronomical practice, from observatory equipment to pedagogical curricula to theological frameworks. The perceived magnitude of this reorganization cost creates resistance to paradigm shifts even when the existing system is demonstrably brittle.

A system remains stable, despite accumulating brittleness, so long as the perceived transition cost exceeds the burden of ongoing maintenance. This explains the persistence of fitness traps discussed in Section 4.6: agents within the system can recognize rising brittleness yet rationally judge that the catastrophic cost of reorganization outweighs the incremental pain of continued patching. The system enters a phase where brittleness compounds but institutional structures remain locked in place.

However, this stability is not permanent. As maintenance costs accumulate, they eventually approach and then exceed the one-time cost of transition. This is the tipping point: the moment when the daily energetic expenditure required to suppress reality's pushback becomes more burdensome than the catastrophic reorganization. At this point, the system becomes vulnerable to paradigm shift.

This dynamic generates the punctuated equilibrium pattern observed throughout the history of science and institutions. We see long periods of stability during which a system accumulates brittleness through patches and coercive enforcement, followed by relatively sudden ruptures where the system undergoes rapid reorganization. The transition is rarely smooth because the coordination problem is severe: abandoning an established system requires collective action, and agents must coordinate their shift despite the immediate costs each will bear.

The role of external shocks in this process is clarificatory rather than causal. A crisis such as war, plague, or environmental collapse does not create the brittleness; it reveals it by suddenly spiking the maintenance costs. A system already near its tipping point, managing high brittleness through costly suppression, can be pushed past the threshold by a shock that makes the daily cost of maintenance suddenly untenable. This is why brittle systems are vulnerable to contingent events that more resilient systems absorb without regime change.

This framework also clarifies the role of what historians might term "early adopters" or "intellectual pioneers." These are agents who perceive that the accumulated burden of maintenance has already exceeded the cost of transition before this becomes consensus. By bearing the immediate social and material costs of heresy, they function as nucleation points for the new paradigm, demonstrating its viability and thereby reducing the perceived transition cost for others. Their actions are not irrational but reflect a different assessment of when the crossing point has been reached.

This account complements the fitness trap analysis in Section 4.6 by explaining not just why systems become locked but what unlocks them. The interplay between accumulating maintenance costs and the threshold of transition costs provides a naturalistic explanation for the temporal dynamics of paradigm shifts, grounded in the same brittleness metrics that diagnose system health. It is not a deterministic prediction of when shifts will occur but a framework for understanding why they occur when they do.

**Qualification:** This mechanism explains one pathway to revolutionary science—the crisis-driven pathway Kuhn emphasized, where accumulating anomalies and rising theoretical costs create pressure for paradigm shift. It does not claim to explain all scientific revolutions.

Some revolutions occur not through accumulated brittleness but through sudden expansion of possibility space. The double helix discovery (Watson & Crick 1953) revolutionized genetics not because classical Mendelian genetics was brittle, but because molecular biology opened entirely new investigative domains. Classical genetics remained viable within its scope; the revolution consisted in recognizing that genetic mechanisms could be studied at a more fundamental level.

Our framework addresses crisis-driven revolutions (phlogiston to oxygen, Ptolemaic to Copernican) where rising operational costs force replacement. It does not claim that brittleness is the only mechanism for paradigm shift. Possibility-expanding revolutions like the double helix represent a different dynamic: not replacement due to fragility, but augmentation through new access to causal detail.

In both cases, however, the Negative Canon plays a crucial role. Even possibility-expanding revolutions are constrained by what has demonstrably failed. The molecular revolution still had to respect the inheritance patterns Mendel discovered—it explained them at a deeper level rather than replacing them.

## 5. Applications: Mathematics as a Paradigm Case of Internal Brittleness

The account thus far has focused on domains where pragmatic pushback comes from external reality: empirical predictions fail, technologies malfunction, societies collapse. This naturally raises a challenge: does the framework apply only to empirically testable domains, or can it illuminate abstract knowledge systems like mathematics and logic? This section argues that it can. By examining mathematics, we can demonstrate the framework's generality, showing how the logic of brittleness operates through purely internal constraints of efficiency, consistency, and explanatory power.

**The Core Insight:** Mathematical frameworks face pragmatic pushback through internal inefficiency rather than external falsification.

### 5.1 The Logic of Internal Brittleness

While mathematical frameworks cannot face direct empirical falsification, they experience pragmatic pushback through accumulated internal costs that render them unworkable. These costs manifest through our standard brittleness indicators, adapted for abstract domains:

**M(t): Proof Complexity Escalation**
- Increasing proof length without proportional explanatory gain
- Measured as: average proof length for theorems of comparable scope over time
- Rising M(t) signals a degenerating research program where increasing effort yields diminishing insight

**P(t): Conceptual Debt Accumulation (proxied by Axiom Proliferation)**
- Ad-hoc modifications to patch paradoxes or anomalies
- Measured as: increasing share of new axioms devoted to resolving contradictions vs. generating novel theorems
- High P(t) indicates mounting conceptual debt from defensive modifications

**C(t): Contradiction Suppression Costs**
- Resources devoted to preventing or managing paradoxes
- Measured as: proportion of research addressing known anomalies vs. extending theory
- High C(t) reveals underlying fragility requiring constant maintenance

**R(t): Unification Power**
- Ability to integrate diverse mathematical domains under common framework
- Measured as: breadth of cross-domain applicability
- Declining R(t) signals fragmentation and loss of coherence

The abstract costs in mathematics can be operationalized using our diagnostic toolkit, demonstrating the framework's universality across domains where feedback is entirely internal to the system.

### 5.2 Case Study: Brittleness Reduction in Mathematical Foundations

To illustrate these metrics in action, consider examples of mathematical progress as brittleness reduction across different domains:

**Non-Euclidean Geometry:**
- Euclidean geometry exhibited high brittleness for curved space applications
- Required elaborate patches (like epicycles in astronomy) to explain planetary motion
- Non-Euclidean alternatives demonstrated lower brittleness for cosmology and general relativity
- **Pattern:** Replace high-brittleness framework with lower-brittleness alternative when problem domain expands

**Calculus Foundations:**
- Infinitesimals were intuitive but theoretically brittle, generating paradoxes of the infinite
- Epsilon-delta formalism demanded higher initial complexity but delivered lower long-term brittleness
- Historical adoption pattern follows brittleness reduction trajectory
- Demonstrates how short-term complexity increase can yield long-term stability gains

**Category Theory:**
- More abstract and initially more complex than set theory
- But demonstrates lower brittleness for certain domains (algebraic topology, theoretical computer science)
- Adoption follows domain-specific viability assessment
- Shows how optimal framework varies by application domain

Nowhere is this dynamic clearer than in the response to Russell's Paradox, which provides a paradigm case of catastrophic brittleness and competing resolution strategies.

**Naive Set Theory (pre-1901):**
- M(t): Moderate (proofs reasonably concise for most theorems)
- R(t): Exceptional (successfully unified logic, number theory, and analysis under a single framework)
- Appeared to exhibit low brittleness across all indicators
- Provided an elegant foundation for mathematics

**Russell's Paradox (Russell 1903):**
- Revealed infinite brittleness: the theory could derive a direct contradiction
- Considered the set R = {x | x ∉ x}. Is R ∈ R? Both yes and no follow from the axioms
- All inference paralyzed (if both A and ¬A are derivable, the principle of explosion allows derivation of anything)
- Complete systemic collapse: the framework became unusable for rigorous mathematics
- This wasn't a peripheral anomaly but a catastrophic failure at the system's foundation

**Response 1: ZF Set Theory (Zermelo-Fraenkel + Axiom of Choice)**
- Added carefully chosen axioms (Replacement, Foundation, Separation) to block the paradox
- M(t): Increased (more axioms create more complex proof requirements)
- P(t): Moderate (new axioms serve multiple purposes beyond merely patching the paradox)
- C(t): Low (paradox completely resolved, no ongoing suppression or management needed)
- R(t): High (maintained most of naive set theory's unifying power across mathematical domains)
- **Diagnosis:** Successful low-brittleness resolution through principled modification
- The additional complexity was justified by restored foundational stability

**Response 2: Type Theory (Russell/Whitehead)**
- Introduced stratified hierarchy that structurally prevents problematic self-reference
- M(t): High (complicated type restrictions make many proofs substantially longer)
- P(t): Low (structural solution rather than ad-hoc patch)
- C(t): Low (paradox is structurally impossible within the system)
- R(t): Moderate (some mathematical domains resist natural formulation within type hierarchies)
- **Diagnosis:** Alternative low-brittleness solution with different trade-offs
- Sacrifices some unification power for structural guarantees against contradiction

**Response 3: Paraconsistent Logic**
- Accepts contradictions as potentially derivable but attempts to control "explosion"
- M(t): Variable (depends on specific implementation details)
- P(t): Very High (requires many special rules and restrictions to prevent inferential collapse)
- C(t): Very High (demands constant management of contradictions and their containment)
- R(t): Low (marginal adoption, limited to specialized domains)
- **Diagnosis:** A higher-brittleness resolution, as it requires sustained, complex, and costly mechanisms to manage contradictions rather than eliminating them
- The system exhibits sustained high maintenance costs without corresponding payoffs

**Historical Outcome:** The mathematical community converged primarily on ZF set theory as the standard foundation, with Type Theory adopted for specific domains where its structural guarantees prove valuable (such as computer science and constructive mathematics). Paraconsistent approaches remain peripheral. This convergence reflects differential brittleness among the alternatives, not arbitrary historical preference or mere convention. The outcome demonstrates how pragmatic selection operates in purely abstract domains through internal efficiency rather than external empirical testing.

### 5.3 Power, Suppression, and the Hard Core

Engaging with insights from feminist epistemology (Harding 1991), we can see that even mathematics is not immune to power dynamics that generate brittleness. When a dominant mathematical community uses institutional power to suppress alternative approaches, this incurs measurable Coercive Overheads (C(t)):

**Mechanisms of Mathematical Suppression:**
- Career punishment for heterodox approaches to foundations or proof methods
- Publication barriers for alternative mathematical frameworks
- Curriculum monopolization by dominant approaches
- Citation exclusion of rival methodologies

**Measurable Costs:**
- **Innovation lag:** Talented mathematicians driven from the field when their approaches are rejected for sociological rather than technical reasons
- **Fragmentation:** Splinter communities forming alternative journals and departments
- **Inefficiency:** Duplication of effort as alternative approaches cannot build on dominant framework results
- **Delayed discoveries:** Useful insights suppressed for decades (e.g., non-standard analysis resisted despite valuable applications)

**The Brittleness Signal:** When a mathematical community requires high coercive costs to maintain orthodoxy against persistent alternatives, this signals underlying brittleness: the dominant framework may not be optimally viable.

**Historical Example: Intuitionist vs. Classical Mathematics**
- Intuitionists demonstrated genuine technical alternatives with different foundational commitments
- Classical community initially suppressed through institutional power (career barriers, publication difficulties)
- High coercive costs required to maintain dominance
- Eventual accommodation as constructive methods proved valuable in computer science and proof theory
- **Diagnosis:** Initial suppression revealed brittleness in classical community's claim to unique optimality

**Why Logic Occupies the Core**

Logic isn't metaphysically privileged; it's functionally indispensable.

**The Entrenchment Argument:**
1. Revising logic requires using logic to assess the revision
2. This creates infinite regress or circularity
3. Therefore logic exhibits infinite brittleness if removed
4. Systems under bounded rationality (March 1978) must treat such maximal-cost revisions as core

**This is pragmatic necessity, not a priori truth:**
- Logic could theoretically be revised if we encountered genuine pragmatic pressure sufficient to justify the cost
- Some quantum logics represent such domain-specific revisions
- But the cost threshold is exceptionally high: logic underpins all systematic reasoning
- Most "apparent" logic violations turn out to be scope restrictions rather than genuine revisions of core principles

### 5.4 The General Principle: Mathematics as Pure Pragmatic Selection

Mathematics demonstrates the framework applies beyond empirical domains. All domains face pragmatic selection, though the feedback mechanism varies: external prediction failure for physics, social collapse for politics, internal inefficiency for mathematics. The underlying principle is analogous: brittle systems accumulate costs that drive replacement by more viable alternatives. The costs differ by domain, but the selection logic remains.

Mathematics is not a special case requiring different epistemology; it's a pure case showing how pragmatic selection operates when feedback is entirely internal to the system. The convergence on ZF set theory, the accommodation of intuitionist insights, and the adoption of non-standard analysis where it proves useful all demonstrate the same evolutionary dynamics at work in physical science, but operating through internal efficiency rather than external empirical testing. This universality strengthens the framework's claim that objective knowledge arises from pragmatic filtering across all domains of inquiry.

Thus, mathematics, far from being a counterexample to a naturalistic epistemology, serves as its purest illustration, demonstrating that the logic of brittleness reduction operates universally, guided by the selective pressures of internal coherence and efficiency.

Having demonstrated how brittleness diagnostics apply even to abstract domains like mathematics, we now situate EPC within broader epistemological debates.

## 6. Situating the Framework in Contemporary Debates

This paper has developed what can be termed **Systemic Externalism**: a form of externalist epistemology that locates justification not in individual cognitive processes but in the demonstrated reliability of entire knowledge systems. This section clarifies the framework's position within contemporary epistemology by examining its relationship to four major research programs: coherentist epistemology, social epistemology, evolutionary epistemology, and neopragmatism.

### 6.1 A Grounded Coherentism and a Naturalized Structural Realism

While internalist coherentists like Carlson (2015) have successfully shown that the web must have a functionally indispensable core, they lack resources to explain why that core is forged by external discipline. Systemic Externalism provides this missing causal engine, grounding Carlson's internal structure in an externalist history of pragmatic selection.

Justification requires coherence plus network authority earned through external performance. Coherence is necessary: a belief lacking integration with a broader inferential system lacks the connections required for application and evaluation. But coherence is not sufficient. A network's authority must be earned, not stipulated. It is earned through demonstrated low brittleness, successful interventions across domains, minimal coercive overhead, and sustained adaptability to novel problems.

This dual requirement distinguishes our position sharply from both classical coherentism and purely process-based externalisms. Against classical coherentism: a perfectly coherent phlogiston network or flat-earth cosmology provides no justification because the network itself has failed the externalist test. Against process reliabilism: we do not assess individual cognitive mechanisms but the historical track record of entire public knowledge systems.

Unlike Zollman's (2007, 2013) static network models and Rosenstock et al. (2017), EPC examines evolving networks under pushback, extending ECHO's harmony principle with external brittleness filters Thagard's internalism lacks. The coherence structure tells us which beliefs hang together. The external performance record tells us whether that coherent structure deserves our credence.

#### 6.1.1 Relationship to Structural Realism

Structural realism comes in two forms. Epistemic Structural Realism (Worrall 1989) holds that we can only know the structure of the world, not the intrinsic nature of entities. Ontic Structural Realism (Ladyman and Ross 2007) holds that structure is all there is: relations are ontologically prior to relata.

Our framework is neither, because we are answering a different question.

Structural realism asks: What can we know? (ESR) or What exists? (OSR). These are epistemological and metaphysical questions respectively.

We ask: What makes beliefs justified? This is a question about warrant, not about the furniture of the world or the limits of knowledge.

Our answer: beliefs are justified by their role in a web that has earned authority through pragmatic success. That the successful webs tend to share structure is an empirical observation about what survives filtering, not a thesis about the intrinsic nature of reality.

We are therefore compatible with structural realism but not identical to it. A structural realist might adopt our account of justification. An advocate of our framework might accept or reject structural realism as a metaphysical thesis. The questions are orthogonal.

That said, our framework does provide what structural realism has long lacked: a naturalistic engine for convergence on objective structures. Worrall posited structural continuity across theory change but offered no mechanism explaining why such continuity should occur. We explain it: brittle theories fail systematically, low-brittleness ones survive. The historical record shows systematic elimination of high-brittleness systems. The convergence toward low-brittleness structures, documented in the Negative Canon, provides positive inductive grounds for realism about the objective viability landscape our theories progressively map.

Similarly, while Ontic Structural Realism posits that the world is fundamentally structural, our framework explains how scientific practices are forced to converge on these objective structures through pragmatic filtering. The Apex Network is the complete set of viable relational structures, an emergent fact about our world's constraint topology, discovered through pragmatic selection rather than stipulated by metaphysics.

The relationship is therefore one of mutual support. Structural realism (especially OSR) provides a plausible metaphysics for what the Apex Network might be. Our framework provides the missing evolutionary mechanism explaining how inquiry converges on it. But one can accept our justification account without committing to structural realist metaphysics, and one can be a structural realist without adopting our specific theory of warrant.

#### 6.1.2 Distinguishing Systemic Externalism from Other Externalisms

Systemic Externalism contrasts with Process Reliabilism (Goldman 1979) and Virtue Epistemology (Zagzebski 1996). Process Reliabilism locates justification in the reliability of individual cognitive processes; Systemic Externalism shifts focus to the demonstrated historical viability of the public knowledge system that certifies the claim. Virtue Epistemology grounds justification in individual intellectual virtues; Systemic Externalism attributes resilience and adaptability to the collective system. Systemic Externalism thus offers macro-level externalism, complementing these micro-level approaches.

### 6.2 Reconciling with Quine: From Dispositions to Objective Structures

Our relationship to Quine is one of architectural inheritance rather than metaphysical commitment. We adopt his holistic, pragmatically-revised, coordinative web structure as essential infrastructure for our framework. The specific dispositional semantics, indeterminacy thesis, and debates about analyticity remain orthogonal to our core claims about systemic brittleness, pragmatic pushback, and convergence toward the Apex Network. This section clarifies how we build on Quine's architecture while remaining agnostic about contested metaphysical details.

While deeply indebted to Quine, our framework must address the apparent tension between his austere behaviorism and our seemingly realist claims. The resolution lies not in departing from Quine, but in building a multi-level structure upon his foundation.

First, where Quine's indeterminacy thesis seems to preclude testable propositions, we show how conscious awareness of dispositions allows for the articulation of specific sentences that serve as functional propositions. This allows for public assessment and a robust epistemology without positing the abstract entities Quine rejected.

Second, we provide a naturalistic engine for the convergence that Quine's underdetermination thesis leaves mysterious. The Apex Network is not a Platonic realm but an attractor basin in the space of possible dispositional patterns, determined by real-world pragmatic constraints. It explains why some webs of belief systematically outperform others without abandoning naturalism.

Finally, we supplement Quine's micro-level external constraint of sensory stimulation with a macro-level constraint: systemic cost. A coherent system can always accommodate anomalous experiences, but it cannot indefinitely ignore the compounding costs its ad-hoc adjustments generate. This provides a robust, non-discursive filter that grounds the entire web of belief against the isolation objection. Our project, therefore, is not a rejection of Quine but an attempt to complete his naturalistic turn by integrating it with the dynamics of systems theory and cultural evolution. Crucially, it supplements his external constraint of individual sensory stimulation with a novel, macro-level external constraint: the non-discursive, systemic costs generated by the web as a whole.

#### 6.2.1 Structural Convergence Is the Only Kind There Could Be

A clarification is necessary here for readers expecting Peircean convergence on propositional content. Within Quinean architecture, there are no propositions as abstract entities bearing meanings independent of their place in the web. Beliefs are dispositions to assent to sentences. Meaning is exhausted by inferential role. The indeterminacy of translation thesis holds that there is no fact of the matter about what a sentence "really means" beyond its behavioral and inferential profile.

This transforms what "convergence" can mean. Peirce imagined inquiry converging on truth, implicitly on propositional content that matches reality. But if Quine is right, there is no propositional content over and above structural relations. The "content" of a belief just is its place in the web: its connections to stimulus, its inferential liaisons, its role in guiding action.

Structural convergence is therefore not a weaker or deflationary substitute for "real" convergence on truth. It is the only kind of convergence available within a Quinean framework. When we say that viable webs converge on shared structure, we are not offering less than Peirce promised. We are specifying what convergence amounts to once we take Quine's deflationism about meaning seriously.

The Apex Network is not an approximation to a Truth we could in principle state propositionally. It is the structural configuration toward which webs must tend under constraint. Within Quinean naturalism, there is nothing further for them to tend toward.

Consider what this means concretely. When independent scientific communities converge on shared mathematical structures, experimental protocols, and predictive frameworks, they are not discovering a pre-linguistic Platonic form. They are being shaped by the same pragmatic constraints into functionally equivalent dispositional patterns. The convergence is real. The constraints are objective. But what converges is structure: patterns of inference, habits of prediction, coordinated responses to stimuli.

This also clarifies the status of theoretical disagreement. Rival interpretations of quantum mechanics may differ in their ontological stories (wave collapse vs. many worlds vs. pilot wave) while sharing identical mathematical structure and predictive apparatus. From a Quinean perspective, the shared structure is what matters. The divergent interpretations are attempts to impose propositional content where only inferential structure exists. The convergence on structure is convergence on what is real within this framework.

This is not instrumentalism. The structure is not "just useful." It is what knowledge is. A Standing Predicate that has migrated to the hard core of our web is not a convenient fiction. It is a dispositional pattern so deeply entrenched, so widely validated, so functionally indispensable, that it has become constitutive of inquiry in its domain. Its reality is the reality of functional structure, not the reality of correspondence to abstract propositions.

The framework therefore integrates Quinean naturalism with realism about structure. We are realists because the constraint landscape is mind-independent and the convergent patterns it forces are objective. But we are Quinean because what converges is not propositional content but dispositional structure. This is the only realism available within naturalism, and it is realism enough.

### 6.3 A Realist Corrective to Neopragmatism and Social Epistemology

The framework developed here retains pragmatism's anti-foundationalist spirit and focus on inquiry as a social, problem-solving practice. Its core ambition aligns with the foundational project of classical pragmatism: to articulate a non-reductive naturalism that can explain the emergence of genuine novelty in the world (Baggio and Parravicini 2019). By grounding epistemology in dispositions to assent shaped by pragmatic feedback (following Quine's call to replace traditional epistemology with empirical psychology [Quine 1969]), we maintain naturalistic rigor while avoiding the foundationalist trap of positing privileged mental contents. Our disposition-based account provides precisely what Quine's naturalized epistemology promised but could not fully deliver: a bridge from individual cognitive behavior to social knowledge systems that remains fully naturalistic while accounting for external constraints.

However, our model offers a crucial corrective to neopragmatist approaches that are vulnerable to the charge of conflating epistemic values with mere practical utility (Putnam 2002; Lynch 2009) or reducing objectivity to social consensus. Thinkers like Rorty (1979) and Brandom (1994), in their sophisticated accounts of justification as a linguistic or social practice, lack a robust, non-discursive external constraint. This leaves them with inadequate resources for handling cases where entire communities, through well-managed discourse, converge on unviable beliefs.

Our framework provides this missing external constraint through its analysis of systemic failure. The collapse of Lysenkoist biology in the Soviet Union, for instance, was not due to a breakdown in its internal "game of giving and asking for reasons"; indeed, that discourse was brutally enforced. Its failure was a matter of catastrophic first-order costs that no amount of conversational management could prevent. This focus on pragmatic consequence as a real, external filter allows us to distinguish our position from other forms of "pragmatic realism." El-Hani and Pihlström (2002), for example, resolve the emergentist dilemma by arguing that emergent properties "gain their ontological status from the practice-laden ontological commitments we make." While we agree that justification is tied to practice, our model grounds this process in a more robustly externalist manner. Pragmatic viability is not the source of objectivity; it is the primary empirical indicator of a system's alignment with the mind-independent, emergent structure of the Apex Network.

This leads to a key reframing of the relationship between agreement and truth. Genuine solidarity is not an alternative to objectivity but an emergent property of low-brittleness systems that have successfully adapted to pragmatic constraints. The practical project of cultivating viable knowledge systems is therefore the most secure path to enduring agreement. This stands in sharp contrast to any attempt to define truth as a stable consensus within a closed system, a procedure that our framework would diagnose as a potential coherence trap lacking the necessary externalist check of real-world systemic costs.

Similarly, our framework provides an evolutionary grounding for the core insights of **social epistemology** (Goldman 1999; Longino 2002). Social epistemic procedures like peer review and institutionalized criticism can reduce systemic brittleness by filtering errors and forcing claims to survive adversarial scrutiny. However, their effectiveness depends on surrounding practices: data transparency, code availability, replication incentives, adversarial diversity. When these supports fail, peer review becomes unreliable as an error or fraud detector.

The claim is therefore not that peer review is uniformly truth-conducive or that existing institutions reliably track viability. The claim is that structured adversarial scrutiny is one mechanism by which communities reduce brittleness when properly constituted. Empirical evidence from the replication crisis (Open Science Collaboration 2015) shows that peer review without transparency and replication incentives can perpetuate high-brittleness systems. In fields like biomedical research, where referees lack access to raw data and processing code, peer review has proven particularly unreliable as a fraud-detection mechanism (Begley & Ellis 2012). The framework's contribution is diagnosing which institutional configurations generate rising costs (suppression of dissent, barriers to error correction, accumulated conceptual debt) versus which facilitate brittleness reduction.

This provides the externalist check that purely procedural models can lack. It also offers an empirical grounding for the central insight of standpoint theory (Harding 1991; Lugones 2003), naturalizing the idea that marginalized perspectives can be a privileged source of data about a system's hidden costs. In our model, marginalized perspectives are not privileged due to a metaphysical claim about identity, but because they often function as the most sensitive detectors of a system's First-Order Costs and hidden Coercive Overheads (C(t)). A system that appears stable to its beneficiaries may be generating immense, unacknowledged costs for those at its margins. Their dissent is not automatically correct, but it is epistemically valuable as a diagnostic of potential brittleness that dominant networks may be insulating themselves from recognizing. Suppressing these perspectives is therefore not just a moral failure, but a critical epistemic failure that allows brittleness to accumulate undetected. This view of collective knowledge as an emergent, adaptive process finds resonance in contemporary work on dynamic holism (Sims 2024).

#### Collective Calibration

Empirical models of social epistemic networks (O'Connor and Weatherall 2019) suggest that objectivity is a function of communication topology. EPC operationalizes this insight: calibration efficiency inversely correlates with brittleness. The more diverse the error signals integrated (Longino 1990; Anderson 1996), the more stable the Apex Network.

#### 6.3.1 Grounding Epistemic Norms in Systemic Viability

A standard objection to naturalistic epistemology is that a descriptive account of how we reason cannot ground a prescriptive account of how we ought to reason (Kim 1988). As noted above, pragmatist approaches face a similar charge of conflating epistemic values with merely practical ones like efficiency or survival (Putnam 2002; Lynch 2009). Our framework answers this "normativity objection" by grounding its norms not in chosen values, but in the structural conditions required for any cumulative inquiry to succeed over time.

Following Quine's later work, we treat normative epistemology as a form of engineering (Moghaddam 2013), where epistemic norms are hypothetical imperatives directed at a practical goal. Our framework makes this goal concrete: the cultivation of low-brittleness knowledge systems. This conditional form is not empty or arbitrary. Inquiry communities do not freely choose whether to care about memory, transmission, coordination, and error correction if they are engaged in cumulative inquiry at all. Those goals are built into the practice's structure. The authority for this approach rests on two arguments.

Epistemic ought-claims are also scale-dependent. We can ask what a single inquirer ought to believe, what a specific institution ought to adopt as procedure, or what a broad public ought to treat as warranted. In this paper, unless otherwise specified, claims about what "we" ought to do are claims at the widest practical scale: the maximally shared standards of cumulative public inquiry.

First, a constitutive argument: any system engaged in a cumulative, inter-generational project, such as science, must maintain sufficient stability to preserve and transmit knowledge. A system that systematically undermines its own persistence cannot, by definition, succeed at this project. The pressure to maintain a low-brittleness design is therefore not an optional value but an inescapable structural constraint on the practice of cumulative inquiry itself.

Second, an instrumental argument: the framework makes a falsifiable, empirical claim that networks with a high and rising degree of measured brittleness are statistically more likely to collapse or require radical revision. From this descriptive claim follows a conditional recommendation: if an agent or institution has the goal of ensuring its long-term stability and problem-solving capacity, then it has a powerful, evidence-based reason to adopt principles that demonstrably lower its systemic brittleness.

This reframes the paper's normative language. When this model describes one network as "better" or identifies "epistemic progress," these are not subjective value judgments but technical descriptions of systemic performance. A "better" network is one with lower measured brittleness and thus a higher predicted resilience against failure. Viability is not an optional norm to be adopted; it is a structural precondition for any system that manages to become part of the historical record at all.

### 6.4 Distinguishing Brittleness from Adjacent Frameworks

The brittleness framework's core innovation is making visible a specific kind of epistemic signal: the accumulated resource costs that diagnose a knowledge system's rising vulnerability to failure. This is not reducible to existing philosophical concepts. The following subsections answer the five most immediate objections: that brittleness is merely a relabeling of Lakatosian degeneration, Ockham's simplicity, predictive failure, sociological description, or Laudan's problem-solving effectiveness. Each has superficial similarities to brittleness analysis; each misses the framework's distinctive contribution.

#### 6.4.0 Brittleness and Traditional Epistemic Virtues

Before distinguishing brittleness from specific adjacent frameworks, we address a general worry: is systemic brittleness simply a repackaging of familiar epistemic criteria under new terminology?

The answer is no. Brittleness identifies a dimension of epistemic assessment that is orthogonal to—and can diverge from—all traditional epistemic virtues:

**Accuracy.** A knowledge system can be locally accurate while accumulating high brittleness. As we detail in Section 6.4.3, a theory can generate correct predictions while requiring escalating ad hoc modifications to do so. The Ptolemaic system maintained predictive adequacy for centuries, yet its brittleness manifested not in failed predictions but in the proliferating machinery needed to generate those predictions. Accuracy is propositional and synchronic; brittleness is systemic and diachronic.

**Internal coherence.** A system can be perfectly coherent within its own conceptual framework yet profoundly misaligned with reality's constraint structure. Coherence is necessary for knowledge systems—incoherent systems cannot function—but it is not sufficient. As we establish in Section 4.3, coherence with a flat-earth network or a phlogiston network provides no epistemic warrant unless that network has demonstrated viability through sustained low brittleness. Coherentism requires external grounding; brittleness provides it.

**Predictive success.** Short-term predictive success can mask long-term structural fragility. A research program can succeed in its initial domain while proving brittle when exported to adjacent contexts or when sustaining that success requires escalating resources. Predictive success measures output; brittleness measures the sustainability of the knowledge-generating process itself.

**Simplicity.** This deserves careful treatment (Section 6.4.2). Brittleness explains why Ockham's Razor succeeds: theories that achieve compression can generalize to novel cases, while theories approaching the complexity of their data lose predictive power and become brittle. But simplicity and brittleness remain orthogonal—a simple theory can be brittle if it requires constant patching, and a complex theory can exhibit low brittleness if it unifies vast domains without defensive modification. Simplicity is synchronic parameter-counting; brittleness is diachronic cost-tracking.

**The key insight:** Traditional criteria assess the epistemic status of propositions or theories at a moment in time. Brittleness assesses the sustainability of knowledge systems as ongoing enterprises under continued pressure. It adds the missing dimension of maintenance cost—the resources required to sustain coherence, accuracy, and predictive success when reality pushes back.

This is not a replacement for traditional criteria but a complement. A maximally viable knowledge system will exhibit low brittleness AND traditional virtues. But when they diverge—when a system is accurate yet brittle, or coherent yet unsustainable—brittleness provides the diagnostic that reveals hidden fragility before catastrophic failure.

#### 6.4.1 Brittleness vs. Degenerating Research Programmes

Lakatos's *Methodology of Scientific Research Programmes* (Lakatos 1970) provides the closest historical precedent for brittleness analysis. Both frameworks track the proliferation of auxiliary hypotheses around a protective belt when a research programme faces anomalies. A degenerating programme, in Lakatos's terms, generates no novel predictions; its theoretical modifications are purely defensive, accommodating known anomalies rather than anticipating new phenomena.

The similarity is genuine. But the frameworks operate at different temporal and epistemic levels.

**Lakatos is retrospective classification.** We determine whether a programme was degenerating by examining its historical trajectory: comparing the ratio of novel predictions to defensive modifications across distinct time periods. Historians can apply Lakatosian categories after the fact to explain why phlogiston chemistry yielded to oxygen-based chemistry or why Ptolemaic astronomy gave way to heliocentrism. The methodology tells us how to score theories once we have the full historical record.

**Brittleness provides forward-looking diagnosis.** The framework tracks current resource allocation patterns, coercive overhead, and complexity-to-performance ratios to signal rising risk before paradigm shift occurs. When P(t), C(t), and M(t) are all rising in a contemporary research programme, this diagnoses fragility and predicts increased vulnerability to displacement, even if we cannot specify what will replace it. Brittleness makes visible structural problems practitioners can recognize and potentially address in real time.

The epistemic difference is this: Lakatos tells historians how to classify past theories. Brittleness tells practitioners which current systems are becoming unreliable engines for future inquiry. One evaluates output; the other diagnoses structural health.

This distinction matters for practitioners. A research community can observe rising conceptual debt $P(t)$, escalating coercive overhead $C(t)$ to maintain the programme's authority, and accelerating model complexity $M(t)$ with diminishing returns. These are not post-hoc classifications but real-time warning signals. The brittleness framework provides actionable epistemic feedback that Lakatosian retrospection cannot supply.

**The Causal Engine Made Explicit:**

Lakatos describes the symptoms: proliferating ad-hoc hypotheses, declining novel predictions, protective belt expansions. Our framework identifies the underlying causal mechanism that generates these symptoms:

1. **Initial misalignment:** The research programme's core (hard core plus protective belt) has structural features that misalign with constraint structure in some domain.
2. **Pragmatic pushback:** This misalignment manifests as failed predictions, anomalies, or limited exportability to adjacent domains.
3. **Defensive response:** Rather than revise the hard core (high cost), practitioners add auxiliary hypotheses to protective belt (lower immediate cost).
4. **Rising maintenance burden:** Each auxiliary patch carries ongoing costs: parameters to tune, special cases to track, consistency constraints to maintain.
5. **Cost accumulation:** As patches proliferate, $M(t)$ rises (model complexity), $P(t)$ rises (patch velocity), $C(t)$ may rise (institutional resources defending programme).
6. **Diminishing returns:** New predictions become harder to generate; each new phenomenon requires new auxiliary structure.
7. **Unsustainability:** Eventually operational costs exceed benefits; practitioners migrate to alternative programmes with lower cost structure.
8. **Abandonment:** Programme enters Negative Canon as cautionary example of structural misalignment.

This is a causal process: misalignment leads to pragmatic pushback, which triggers defensive patching, which generates rising costs, which forces abandonment. Lakatos retrospectively identifies stages 3-7 after the fact. Brittleness metrics ($P(t)$, $C(t)$, $M(t)$) track stages 4-6 in real-time, providing forward-looking diagnosis.

**Example:** Phlogiston chemistry followed this exact trajectory. Initial elegance (stages 1-2) led to negative weight patch (stage 3), which generated proliferating parameters (stage 4), manifesting as defensive publication patterns (stage 5-6). Lavoisier's oxygen framework provided lower-cost alternative (stage 7), resulting in phlogiston abandonment (stage 8).

#### 6.4.2 Why Brittleness Is Not Just Simplicity

Ockham's Razor, and the broader tradition of preferring simpler theories, appears closely related to brittleness analysis. Both penalize proliferating structure. A theory that requires increasingly elaborate machinery seems both less simple and more brittle.

But simplicity and brittleness are orthogonal dimensions.

**Ockham's Razor is a heuristic tie-breaker** when theories are evidentially equivalent. If two theories predict the same phenomena equally well, prefer the simpler one. The principle operates synchronically: it assesses theories at a single time slice, comparing their structural economy. Simplicity is measured by parameter counts, ontological commitments, or other static complexity metrics.

**Brittleness tracks cost dynamics over time.** The framework assesses the diachronic trajectory: is the system's maintenance burden rising or falling? Are modifications integrating smoothly or generating compounding debt? A theory can be simple yet brittle if it starts elegant but accumulates patches rapidly. Conversely, a theory can be complex yet resilient if its high parameter count reflects genuine structural requirements rather than defensive elaboration.

Consider contrasting cases. Geocentric astronomy began simple (celestial spheres rotating around Earth) but accumulated brittleness as epicycles proliferated. The Standard Model in particle physics has high parameter count (dozens of fundamental constants) yet demonstrates low brittleness: its complexity reflects discovered structure, not defensive patching. Novel predictions extend smoothly; anomalies do not require ad hoc machinery; exportability to new energy scales proceeds without fundamental revision.

The key difference: **Simplicity is a synchronic property; brittleness is a diachronic process.** We are not measuring absolute complexity at time t. We are tracking the rate of complexity increase relative to marginal performance gains. When the description length of auxiliary exception-handling machinery begins to exceed the description length of core explanatory principles, brittleness is rising regardless of whether the system is objectively "simple."

This connects to principled complexity measures—with crucial qualifications. Minimum Description Length (Rissanen 1978) and Kolmogorov complexity (Gell-Mann and Lloyd 1996) provide *conceptual* grounding for what brittleness makes intuitive: theories that cannot compress their subject matter lose generative power. However, these measures cannot function as computational tools for our framework. Kolmogorov complexity is uncomputable by definition, and MDL faces practical intractability for complex knowledge systems.

The MDL connection clarifies what brittleness tracks: the rate at which description length grows relative to new evidence incorporated. A resilient theory compresses new data efficiently—each observation integrates smoothly without requiring novel parameters or special-case machinery. A brittle theory exhibits the opposite pattern: each new piece of evidence demands ad hoc patches, causing description length to grow faster than explanatory power. When a system requires an ever-longer description to account for the same phenomena it once explained concisely, this signals rising brittleness regardless of how the system performs on its original training data.

These measures serve as regulative ideals, clarifying the pattern we diagnose (incompressible exception-handling overwhelming core structure) without providing mechanical algorithms. Actual diagnosis relies on observable proxies: publication patterns, parameter proliferation, resource allocation—fallible indicators that track the pattern these formal measures characterize. A theory requiring novel principles for each new domain exhibits rising incompressible complexity. This is the signature of brittleness that simplicity heuristics alone cannot capture.

#### 6.4.3 Systemic Costs Beyond Predictive Success

Standard evidential accounts assess theories by their predictions: successful novel predictions count for the theory; failed predictions count against it. Brittleness analysis seems redundant. If a theory makes bad predictions, evidential accounting already marks this. What does brittleness add?

The framework makes visible three categories of costs that standard evidential accounts systematically miss:

**1. Maintenance costs when predictions succeed.** Ptolemaic astronomy maintained predictive adequacy for centuries. Predictions did not fail; they simply required escalating astronomical labor to generate. Each planetary position could be calculated, but the calculations demanded compounding effort: more epicycles, more eccentric adjustments, more specialized practitioners. The system's predictions worked, yet the cost of operating it as a going concern was rising unsustainably.

Standard evidential scoring sees only that predictions matched observations. Brittleness analysis sees the resource drain: accelerating P(t) as defensive publications proliferate, rising M(t) as model complexity outpaces performance gains. These are real costs imposed by misalignment with constraint structure, even when predictions technically succeed.

**2. Coercive overhead suppressing dissent.** A system can maintain apparent evidential success by suppressing anomalies, marginalizing critics, or controlling what counts as evidence. When a research programme requires increasing institutional resources to enforce compliance (censoring dissenting publications, defunding rival approaches, demanding loyalty oaths to core principles), this C(t) overhead signals brittleness invisible to propositional evidential accounts.

The costs are epistemic, not merely political. Energy spent suppressing challenges is energy diverted from extending the system to new domains or integrating novel phenomena. Rising C(t) predicts fragility: the suppressed anomalies will compound, and the system's authority will erode once coercive mechanisms fail.

**3. Opportunity costs from foreclosed paths.** A system's structure can foreclose entire research directions, rendering certain questions unaskable or certain phenomena invisible. These opportunity costs are invisible to synchronic evidential accounts focused on current predictive success.

Phlogiston chemistry, for example, could not pose the question "What is the role of oxygen in combustion?" because oxygen was conceptualized as dephlogisticated air—a theoretical primitive lacking independent identity. Lavoisier's framework opened problem spaces (quantitative stoichiometry, gas chemistry, thermodynamic integration) that phlogiston's architecture rendered inaccessible. The cost was not failed predictions but stunted growth.

**The difference:** Predictive accuracy is propositional; brittleness is systemic. Standard accounts ask: "Is this claim true?" Brittleness asks: "What is the cost of operating this system as a knowledge-generating engine?" The former assesses truth-apt outputs; the latter assesses the reliability and sustainability of the process generating those outputs.

This is why a theory can be evidentially adequate yet diagnosed as brittle. It passes standard tests, but its resource profile signals rising fragility. The framework adds a missing epistemic dimension: the cost of sustaining coherence.

#### 6.4.4 Brittleness as Epistemic, Not Merely Sociological

The objection is direct. If brittleness tracks resource allocation, institutional friction, and community behavior, is this not sociology of science rather than epistemology? Are we measuring social facts (how communities allocate funding, who gets published, which theories dominate) rather than epistemic warrant (which beliefs are justified)?

The response requires distinguishing manifestation from source.

**Brittleness has epistemic grounding.** The costs tracked by P(t), C(t), and M(t) are not constituted by social agreement. They are imposed by misalignment between a system's structure and reality's constraints. When a theory makes false causal claims, applications in novel domains reveal failures. When a model's core principles misrepresent structural relationships, extending the framework to new problems generates anomalies requiring ad hoc patches. When institutional authorities suppress evidence contradicting their framework, maintaining that suppression consumes resources that could fund genuine inquiry.

These costs are real, measurable, and non-negotiable. They are not socially constructed but discovered through pragmatic pressure. The constraint landscape is mind-independent; brittleness diagnoses the cost of misalignment with that landscape.

**Brittleness has sociological manifestation.** The costs appear as observable social phenomena: publication patterns shifting from novel research to defensive patching (P(t) rising), institutional budgets increasingly devoted to maintaining authority rather than funding exploration (C(t) rising), research communities requiring specialized training to navigate byzantine theoretical machinery (M(t) rising). These are measurable social facts.

But the analogy to biological fitness clarifies the relationship. Fitness is tracked through population dynamics (births, deaths, reproductive success)—sociological facts about organisms. Yet fitness ultimately reflects alignment with environmental constraints. An organism's high fitness is not constituted by the social fact that it has many offspring; the offspring proliferate because the organism's traits align with survival constraints. Similarly, rising C(t) is not merely the social fact that institutions spend resources on suppression; the suppression is necessary because the framework's structure generates resistance by misaligning with reality.

**The dual aspect:** Brittleness has an epistemic dimension (tracks reliability as guide to reality) and a sociological manifestation (appears as resource allocation). The manifestation is social; the source is epistemic misalignment. This is naturalized epistemology (Kitcher 1993): we track social processes as indicators of epistemic quality, recognizing that successful knowledge systems must survive material constraints.

The framework's contribution is showing how to read sociological signals epistemically. Not all resource expenditure signals brittleness (some costs reflect genuine structural requirements). But specific cost patterns—accelerating defensive publications, rising institutional overhead to enforce conformity, proliferating complexity with diminishing returns—reliably diagnose epistemic fragility because they reflect the unsustainable resource drain of fighting reality's constraints.

#### 6.4.5 Brittleness vs. Problem-Solving Effectiveness

Laudan's *Progress and Its Problems* (Laudan 1977) shares the brittleness framework's focus on what theories accomplish rather than abstract truth-relations. Both assess theories pragmatically: not "does this correspond to reality?" but "what does this system achieve?"

The similarity is substantial. Laudan scores theories by counting solved empirical problems (phenomena the theory predicts or explains) minus unsolved conceptual problems (internal contradictions, inconsistencies with background knowledge). A theory making progress solves more problems over time than it generates. This is outcome-focused, historically grounded, and avoids metaphysical disputes about truth.

But four differences distinguish brittleness from problem-solving effectiveness:

**1. Directionality.** Laudan looks backward, summing accumulated achievements. Brittleness looks forward, diagnosing risk. A theory can have high problem-solving score (many puzzles solved historically) while accumulating hidden costs that signal fragility. Phlogiston chemistry solved major problems—combustion, respiration, calcination—earning a respectable Laudan score. But brittleness analysis tracks the rising maintenance burden those solutions required: escalating auxiliary structure, defensive publications, foreclosed problem spaces. Laudan would rank phlogiston moderately successful; brittleness diagnoses it as vulnerable.

**2. Cost visibility.** Laudan's framework can rank a theory highly if it solves problems, even if sustaining those solutions imposes escalating costs. Brittleness makes costs explicit. A "successful" theory with high P(t), C(t), and M(t) is diagnosed as fragile regardless of problem-solving score. The framework asks: at what resource cost does this system generate its solutions? Theories imposing unsustainable costs are predicted to fail even if their current achievements are impressive.

**3. Negative evidence primacy.** Laudan accumulates positive achievements. Brittleness privileges the Negative Canon—what has demonstrably failed provides stronger constraints than what has contingently succeeded. The space of viable theories is bounded not by accumulated successes but by demonstrated failures. Phlogiston is not merely a "less successful" theory in Laudan's comparative ranking but a relegated failure establishing a boundary: viable chemistry must explain mass gain during combustion without invoking negative-weight substances. The Negative Canon provides absolute constraints, not relative rankings.

**4. Mechanism specificity.** Laudan describes what makes theories better (more problems solved). Brittleness provides a causal mechanism for why brittle theories fail: they impose unsustainable costs forcing replacement. When a system's maintenance burden exceeds available resources, practitioners abandon it regardless of historical achievements. This explains paradigm shifts mechanistically: not that better theories win by scoring more points, but that brittle theories become operationally unviable. The collapse is structural, not scoreboard.

Example: Phlogiston chemistry solved combustion (phlogiston release), respiration (phlogiston transfer), and metal formation (removal of phlogiston from calx). Impressive problem-solving. But sustaining these solutions required incompressible auxiliary structure: dephlogisticated air, negative weight, context-dependent properties. When Lavoisier's oxygen framework offered a unified quantitative account with lower M(t), immediate exportability to new reaction types (lower P(t)), and integration with emerging thermodynamics, the transition was not that Lavoisier solved more problems but that his framework imposed dramatically lower operational costs.

The frameworks complement rather than compete. Laudan provides historical accounting; brittleness provides structural diagnosis. Laudan explains progress; brittleness predicts fragility. One measures achievement; the other tracks sustainability.


### 6.5 Plantinga's Challenge: Does Evolution Select for Truth or Mere Survival?

Alvin Plantinga's Evolutionary Argument Against Naturalism (EAAN) poses a formidable challenge to any naturalistic epistemology: if our cognitive faculties are products of natural selection, and natural selection optimizes for reproductive success rather than true belief, then we have no reason to trust that our faculties reliably produce true beliefs (Plantinga 1993, 2011). Evolution could equip us with systematically false but adaptive beliefs: useful fictions that enhance survival without tracking reality. If naturalism is true, the very faculties we use to conclude naturalism is true are unreliable, rendering naturalism self-defeating.

Our framework provides a novel response by collapsing Plantinga's proposed gap between adaptive success and truth-tracking. We argue that in domains where systematic misrepresentation generates costs, survival pressure and truth-tracking converge necessarily. This is not because evolution "cares about" truth, but because reality's constraint structure makes persistent falsehood unsustainable.

**Systemic Brittleness as the Bridge**

Plantinga's argument assumes survival and truth can come apart: that belief systems could be adaptively successful while systematically misrepresenting reality. Our framework challenges this through systemic brittleness. In any domain where actions have consequences constrained by objective features of reality, false beliefs accumulate costs:

- Navigation beliefs with systematic errors generate failures, wasted energy, increased predation risk
- Causal beliefs that misunderstand cause-effect relationships lead to ineffective interventions and resource waste
- Social beliefs that misread dynamics generate coordination failures and coalition instability

These costs compound. A single false belief might be offset by other advantages, but networks of mutually reinforcing falsehoods generate accelerating brittleness through our P(t) mechanism (conceptual debt accumulation). The phlogiston theorist does not just hold one false belief; they must continuously patch their system with ad-hoc modifications as each application reveals new anomalies. This brittleness makes the system vulnerable to displacement by more viable alternatives.

**Domain Specificity: Where Truth-Tracking Matters**

Our response is not that evolution always selects for true belief. Rather, we identify specific conditions under which survival pressure forces truth-tracking. Domain structure determines whether falsity accumulates brittleness:

High-cost domains (strong selection for truth-tracking): Physical causation, basic perception, tool use and engineering, social coordination. Misunderstanding gravity, systematically misperceiving predators, or holding false beliefs about material properties generates immediate, compounding costs.

Low-cost domains (weak selection): Metaphysical speculation, aesthetic preferences, remote historical claims. Believing in Zeus versus no gods has minimal direct costs if behavior is similar. False beliefs about "objective beauty" do not accumulate brittleness.

Systematically misleading domains (Plantinga's worry bites hardest): Self-enhancement biases, tribal epistemology, certain evolved heuristics. Here overconfidence or believing "my group is superior" may be adaptive even if false, because they coordinate action or provide psychological benefits.

This domain-specificity is crucial. Our framework concedes Plantinga's point for low-cost and systematically-misleading domains: these are precisely where the isolation objection has least force. But in high-cost domains (those that matter most for science, engineering, and social coordination), the gap Plantinga identifies cannot persist across long timescales.

**Convergence Through Cultural Evolution**

Plantinga focuses on individual cognitive faculties shaped by biological evolution. Our framework operates at a different level: cultural evolution of public knowledge systems. This shift is crucial. Individual humans may harbor adaptive falsehoods, but public knowledge systems are subject to distinct, higher-order selective pressures: transmissibility (false systems that work in one context often fail when transmitted to new contexts), cumulative testing (each generation's application exposes misalignment), and inter-system competition (when rival systems make incompatible predictions, differential brittleness outcomes determine which survives).

The **Negative Canon** provides overwhelming evidence for this process. Ptolemaic astronomy, phlogiston chemistry, miasma theory, and Lysenkoism were not merely false; they accumulated measurable, catastrophic systemic costs that forced their abandonment. The historical record shows systematic elimination of high-brittleness systems across cultures and eras, suggesting convergence toward a constraint-determined structure (the Apex Network) rather than persistent plurality of useful fictions.

**Where Plantinga's Worry Remains**

We acknowledge our response does not fully dissolve Plantinga's challenge at the individual level. An individual human's cognitive faculties might indeed harbor systematically false but adaptive beliefs in low-cost or systematically-misleading domains. Our claim is more modest: in domains where falsity accumulates brittleness, cumulative cultural evolution forces convergence toward truth-tracking systems, even if individual psychology remains imperfect.

This leaves open several possibilities. Evolutionary debunking arguments may retain force in genuinely low-cost domains like pure aesthetics or speculative metaphysics. However, moral philosophy is not low-cost: moral systems that systematically misrepresent human needs generate catastrophic coordination failures, demographic collapse, and rising coercive overheads. The **Negative Canon** demonstrates this empirically. Individual-level cognitive biases may persist even as system-level knowledge improves. Fundamental uncertainty about whether our most basic faculties (logic, perception) are reliable cannot be eliminated; this is acknowledged in our fallibilism.

**The Self-Defeat Response Reversed**

Plantinga argues naturalism is self-defeating: if naturalism is true, we should not trust the faculties that led us to believe it. Our framework reverses this concern: the very fact that naturalistic science has systematically driven down systemic brittleness across centuries (enabling unprecedented technological success, predictive power, and unification) provides higher-order evidence that the system tracks the Apex Network. The proof is in the low-brittleness pudding.

If our cognitive faculties were fundamentally unreliable in the way Plantinga suggests, we would expect accelerating conceptual debt (rising P(t)), decreasing unification (falling R(t)), and catastrophic failures when theories are applied in novel domains. Instead, mature sciences show the opposite: decreasing P(t), increasing R(t), and successful cross-domain applications. This provides inductive grounds for trusting that our faculties, at least in high-cost domains, track real constraint structures rather than generate useful fictions.

Our response to Plantinga: In domains where systematic misrepresentation accumulates measurable costs, the supposed gap between adaptive success and truth-tracking collapses. Survival is not truth, but in a universe with stable constraints, surviving knowledge systems are forced to approximate truth because persistent falsehood generates brittleness that pragmatic selection eliminates. This is not metaphysical necessity but statistical regularity: an empirical claim falsifiable through the research program outlined in Section 7.2.

Plantinga is right that evolution per se does not guarantee reliability. But evolution plus pragmatic filtering in a constraint-rich environment does generate truth-tracking in precisely those domains where coherentism faces the isolation objection. Where Plantinga sees self-defeat, we see self-correction: the systematic reduction of brittleness over centuries is evidence that the process works, even if no individual step is guaranteed.

### 6.6 Computational and Systematic Precursors

EPC synthesizes four computational/systematic frameworks, advancing each through externalist brittleness diagnostics.

**Thagard's ECHO (1989, 2000):** Models explanatory coherence via 7 principles (symmetry, explanation, contradiction, etc.) as constraint satisfaction in connectionist networks. Activation spreads through excitatory/inhibitory weights until harmony maximizes. EPC extends this: **Standing Predicates** = high-activation nodes; propagation = ECHO dynamics. **Advance:** Brittleness adds dynamic weights derived from pragmatic costs. Accumulating ad-hoc patches (rising P(t)) would create progressively stronger inhibitory effects on the core propositions they protect, while coercive overheads (rising C(t)) would suppress dissenting nodes. ECHO operates internally; EPC adds pragmatic pushback as external discipline, solving Thagard's isolation problem.

**Zollman's Epistemic Graphs (2007, 2010):** Shows topology effects: sparse cycles preserve diversity (reliability bonus), complete graphs risk premature lock-in (speed/reliability trade-off). EPC's **Pluralist Frontier** operationalizes transient diversity. **Advance:** While Zollman models abstract belief propagation, EPC's framework suggests how these models could be grounded. It points toward using the historical record of systemic shocks (such as those catalogued in historical databases) as a source of external validity checks, moving beyond purely abstract network dynamics.

**Rescher's Systematicity (1973, 2001):** Defines truth as praxis-tested systematicity (completeness, consistency, functional efficacy) but lacks quantification. **Advance:** SBI(t) operationalizes: P(t) = consistency metric, R(t) = completeness breadth, enabling falsifiable predictions (brittleness-collapse correlations).

**Kitcher (1993) on Evolutionary Progress:** Models science as credit-driven selection with division of labor across 'significant problems.' **Advance:** **Negative Canon** provides failure engine, brittleness quantifies problem significance via coercive costs (C(t)), diagnosing degenerating programs without reducing to cynical credit-seeking.

**Sims (2024) on Dynamic Holism:** Ecological constraints drive diachronic cognitive revision in nonneuronal organisms. EPC parallels this at macro-cultural scale: pragmatic pushback = ecological variables. **Distinction:** Sims focuses on individual adaptation; EPC on intergenerational knowledge systems. Both emphasize context-sensitive, constraint-driven evolution; EPC adds synchronic diagnostics to Sims' diachronic methodology.

These precursors provide micro-coherence (Thagard), meso-topology (Zollman), normative criteria (Rescher), macro-dynamics (Kitcher), and biological analogy (Sims). EPC unifies them through falsifiable brittleness assessment grounded in historical failure data.

## 7. Final Defense and Principled Limitations

Having situated the framework within existing epistemological traditions and clarified its distinctive contributions, we now turn to a systematic defense of its scope and an honest acknowledgment of its limitations. Any philosophical framework achieves clarity through what it excludes as much as what it includes. This section addresses three critical questions: What epistemic problems does the framework solve, and which does it appropriately leave to other approaches? How can retrospective brittleness diagnostics provide prospective guidance? And what falsifiable predictions does the framework generate? These clarifications fortify the account against misunderstanding while delineating its proper domain of application.

As a macro-epistemology explaining the long-term viability of public knowledge systems, this framework does not primarily solve micro-epistemological problems like Gettier cases. Instead, it bridges the two levels through the concept of higher-order evidence: the diagnosed health of a public system provides a powerful defeater or corroborator for an individual's beliefs derived from that system.

The diagnosed brittleness of a knowledge system provides higher-order evidence that determines rational priors. Following Kelly (2005) on disagreement, when an agent receives a claim, they must condition their belief not only on the first-order evidence but also on the source's reliability (Staffel 2020). Let S be a high-brittleness network, like a denialist documentary. Its diagnosed non-viability acts as a powerful higher-order defeater. Therefore, even if S presents seemingly compelling first-order evidence E, a rational agent's posterior confidence in the claim properly remains low. Conversely, a low-brittleness network like the IPCC earns a high prior through demonstrated resilience. To doubt its claims without new evidence of rising brittleness is to doubt the entire adaptive project of science itself. This provides a rational, non-deferential basis for trust: justification flows from systemic health, grounding micro-level belief in macro-level viability.

### 7.1 From Hindsight to Foresight: Calibrating the Diagnostics

To address the hindsight objection (that we can only diagnose brittleness after failure), we frame the process as a two-stage scientific method: **Stage 1: Retrospective Calibration.** We use the clear data from the **Negative Canon** (historical failures) to calibrate our diagnostic instruments (the P(t), C(t), M(t), R(t) indicators), identifying the empirical signatures that reliably precede collapse. **Stage 2: Prospective Diagnosis.** We apply these calibrated instruments to contemporary, unresolved cases not for deterministic prediction, but to assess epistemic risk and identify which research programs are progressive versus degenerating.

### 7.2 A Falsifiable Research Program

The framework grounds a concrete empirical research program with a falsifiable core causal claim: *a system exhibiting high and rising brittleness across multiple indicators should, over historical time, prove more vulnerable to major revision or collapse when faced with external shocks than a system with low and stable brittleness.* "Collapse" is operationally defined as either (1) institutional fragmentation requiring fundamental restructuring; (2) wholesale paradigm shift in the domain; or (3) substantial reduction in problem-solving capacity requiring external intervention. Historical patterns in collapsed systems, such as Roman aqueduct failures due to accumulating brittleness in hydraulic engineering (Hodge 1992; Turchin 2003), are consistent with this expectation. The specific metrics and dynamic equations underlying this research program are detailed in Appendix A.

The framework also makes a second, independent prediction about convergence geometry. Independently developed viable systems should exhibit substantial structural overlap, and this overlap should tend to increase as inquiry expands and edge cases are tested. If, instead, viable systems repeatedly form an archipelago of mostly disjoint structures, with no stable growth in shared load-bearing commitments, the mountain-range model fails and the Apex claim must be weakened.

**Methodology**: (1) Operationalize brittleness through observable proxies (resource allocation patterns, auxiliary hypothesis rates in literature). (2) Conduct comparative historical analysis using databases like Seshat (a database of historical societies) to compare outcomes across systems with different pre-existing brittleness facing similar shocks, controlling for contingent events. (3) Track structural overlap across independently developed viable systems in the same domain, especially after novel boundary tests, to assess whether shared core commitments expand or fragment. As a conceptual illustration, consider competing COVID-19 models (2020–2022): one might analyze whether highly complex epidemiological models with many parameters showed signs of rising brittleness through persistent predictive failures and required constant revision, while simpler models maintained better predictive alignment over time (Roda et al. 2020). Such analysis would illustrate the framework's diagnostic potential.

### 7.3 Principled Limitations and Scope

Philosophical honesty requires acknowledging not just what a framework can do, but what it cannot. These are not flaws but principled choices about scope and method.

#### 7.3.1 Species-Specific Objectivity

**The Limitation:** Moral and epistemic truths are objective for creatures with our biological and social architecture. Hypothetical beings with radically different structures, for example, telepathic beings that reproduce by fission and feel no pain, would face different constraints and discover a different Apex Network.

**Why We Accept This:** This is relativism at the species level, not cultural level. We accept it as appropriate epistemic modesty; our claims are grounded in constraints humans actually face. This preserves robust objectivity for humans, as all cultures share the same core constraints.

#### 7.3.2 Learning Through Catastrophic Failure

**The Limitation:** We learn moral truths primarily through catastrophic failure. The **Negative Canon** is written in blood. Future moral knowledge will require future suffering to generate data.

**Why We Accept This:** Empirical knowledge in complex domains requires costly data. Medicine, engineering, and social systems all required catastrophic failures before developing safety mechanisms.

**Implication:** Moral learning is necessarily slow. We should be epistemically humble about current certainties yet confident in **Negative Canon** entries.

#### 7.3.3 Floor Not Ceiling

**The Limitation:** The framework maps necessary constraints (the floor), not sufficient conditions for flourishing (the ceiling). It cannot address what makes life meaningful beyond sustainable, supererogatory virtue and moral excellence, aesthetic value and beauty, or the difference between a decent life and an exemplary one.

**Why We Accept This:** Appropriate scope limitation. The framework does what it does well rather than overreaching. It identifies catastrophic failures and boundary conditions, leaving substantial space for legitimate value pluralism above the floor.

**What This Implies:** The framework provides necessary but not sufficient conditions. Thick theories of the good life must build on this foundation. The **Pluralist Frontier** is real: multiple flourishing forms exist, but all must respect the floor (avoid **Negative Canon** predicates).

#### 7.3.4 Expert Dependence and Democratic Legitimacy

**The Limitation:** Accurate brittleness assessment requires technical expertise in historical analysis, statistics, comparative methods, and systems theory. This creates epistemic inequality.

**Why We Accept This:** Similar to scientific expertise generally. Complex systems require specialized knowledge to evaluate.

**Mitigation:** Data transparency, distributed expertise, standpoint epistemology (marginalized groups as expert witnesses to brittleness), institutional design (independent assessment boards), and education can reduce but not eliminate this challenge.

#### 7.3.5 Discovery Requires Empirical Testing

**The Limitation:** While the Apex Network exists as a determined structure, discovering it requires empirical data. We cannot deduce optimal social configurations from first principles alone; we need historical experiments to reveal constraint topology.

**Why We Accept This:** Even in physics and mathematics, we need empirical input. Pure reason can explore logical possibilities, but determining which possibilities are actual requires observation or experiment. For complex social systems with feedback loops and emergent properties, this dependence is stronger.

**What This Allows:** Prospective guidance through constraint analysis. We can reason about likely optimal solutions by analyzing constraint structure, but we need empirical validation. This is stronger than pure retrospection but weaker than complete a priori knowledge.

#### 7.3.6 The Viable Evil Possibility

**The Limitation:** If a deeply repugnant system achieved genuinely low brittleness, the framework would have to acknowledge it as viable, though not necessarily just by other standards.

**Why We Accept This:** Intellectual honesty. The framework maps pragmatic viability, not all moral dimensions.

**Empirical Bet:** We predict such systems are inherently brittle. Historical cases like the Ottoman devşirme system or Indian caste systems exhibited high coercive overheads, innovation lags, and fragility under external shocks (Acemoglu & Robinson 2012; Turchin 2003). True internalization without coercion is rare and resource-intensive. If empirics proved otherwise, we would acknowledge the framework's incompleteness rather than deny evidence.

#### 7.3.7 Clarifying Scope and Limits

These limitations do not undermine the framework's contribution; they define appropriate scope. EPC excels at:

**Strong Claims:**
- Identifying catastrophic systemic failures
- Explaining moral progress as empirically detectable debugging
- Grounding realism naturalistically without non-natural properties
- Providing empirical tools for institutional evaluation
- Offering prospective guidance through constraint analysis
- Unifying descriptive and normative epistemology

**Modest Claims:**
- Does not provide complete ethics
- Does not solve all normative disagreements
- Does not eliminate need for judgment
- Does not achieve view-from-nowhere objectivity
- Does not offer categorical imperatives independent of systemic goals

**The Value Proposition:** A powerful but limited diagnostic tool for systemic health. Use it for what it does well. Supplement with other resources for what it cannot address. Do not expect a complete theory of human flourishing; expect robust tools for avoiding catastrophic failure and identifying progressive change.

#### 7.3.8 The AI Constraint: Embodiment and the Limits of Coherence

**The Limitation:** This framework offers a principled critique of current Large Language Models (LLMs). Contemporary LLMs operate almost exclusively at Level 3 contextual coherence. They are sophisticated engines for maintaining internal consistency within a linguistic corpus, but they lack a physical Markov Blanket that faces thermodynamic consequences for error. Without the friction of reality, an unembodied system cannot distinguish between a plausible fiction and a viable truth. For such a system, there is no metabolic cost to being wrong, no pragmatic pushback to force revision toward viability.

**Why This Matters:** Objective truth (Level 1) and warranted assertion (Level 2) require embodiment in our framework. A system must have boundaries that face real costs when misaligned with environmental constraints. This is not a contingent limitation of current technology but a structural requirement: coherence without consequence cannot converge on the Apex Network. Pragmatic selection presupposes the possibility of differential survival, which requires a physical substrate vulnerable to the costs of misalignment.

**Implication:** LLMs can serve as powerful tools for exploring the space of contextually coherent claims (Level 3), but they cannot independently achieve the epistemic status of justified belief. Their outputs require validation by embodied systems—humans, robots, scientific instruments—that face pragmatic pushback. This distinguishes our framework from views that reduce truth to coherence alone. Information leakage only matters when there are thermodynamic costs to maintaining misaligned boundaries, and such costs require physical embodiment.

This honest accounting strengthens rather than weakens the framework's philosophical contribution.

**What This Theory is NOT:**

- **Not a Complete Ethics:** EPC maps necessary constraints (the floor) but not sufficient conditions for flourishing. It identifies catastrophic failures but leaves space for legitimate pluralism in values above the viability threshold.
- **Not a Foundationalist Metaphysics:** It avoids positing non-natural properties or Platonic forms. Objectivity emerges from pragmatic selection, not metaphysical bedrock.
- **Not Deterministic Prediction:** Claims are probabilistic; brittleness increases vulnerability but does not guarantee collapse.
- **Not a Defense of Power:** Power can mask brittleness temporarily, but coercive costs are measurable indicators of non-viability.
- **Not Relativist:** While species-specific, it defends robust objectivity within human constraints via convergent evidence.
- **Not Anti-Realist:** It grounds fallibilist realism in the emergent Apex Network, discovered through elimination.

**Steelmanned Defense of the Core:** The "Drive to Endure" is not a smuggled value but a procedural-transcendental filter. Any project of cumulative justification presupposes persistence as a constitutive condition. Systems failing this filter (e.g., apocalyptic cults) cannot sustain inquiry. The "ought" emerges instrumentally: favor low-SBI norms to minimize systemic costs like instability and suffering, providing evidence-based strategic advice for rational agents.

## 8. Conclusion

Grounding coherence in long-term viability of knowledge systems rather than internal consistency alone provides the external constraint coherentism requires while preserving its holistic insights. The concept of systemic brittleness offers a naturalistic diagnostic tool for evaluating knowledge systems, while the notion of a constraint-determined Apex Network explains how objective knowledge can arise from fallible human practices.

Systematically studying the record of failed systems discerns the contours of the Apex Network: the emergent set of maximally convergent, pragmatically indispensable principles that successful inquiry is forced to discover.

This model is not presented as a final, complete system but as the foundation for a progressive and falsifiable research program. Critical future challenges remain, such as fully modeling the role of power asymmetries in creating path-dependent fitness traps and applying the framework to purely aesthetic or mathematical domains.

We began with the challenge of distinguishing viable knowledge from brittle dogma. The model we have developed suggests the ultimate arbiter is not the elegance of a theory or the consensus of its adherents but the trail of consequences it leaves in the world. Systemic costs are not abstract accounting measures; they are ultimately experienced by individuals as suffering, instability, and the frustration of human goals. From this perspective, dissent, friction, and protest function as primary sources of epistemological data. They are the system's own real-time signals, indicating where First-Order Costs are accumulating and foreshadowing the rising Coercive Overheads ($C(t)$) that will be required to maintain stability against those pressures.

**Qualification on dissent:** Not all dissent signals brittleness. Dissent from misunderstanding (e.g., popular opposition to evolution based on misinterpretation) provides no epistemic information about the discipline's internal health. The epistemically valuable dissent comes from practitioners or informed observers who identify specific rising costs: replication failures, suppressed anomalies, institutional barriers to error correction. External dissent from misunderstanding is noise; internal dissent about rising operational costs is signal.

It provides the external constraint that coherentism has long needed, but it does so without resorting to foundationalism. It accounts for the convergence that motivates scientific realism, but it does so within a fallibilist and naturalistic picture of inquiry.

Ultimately, Emergent Pragmatic Coherentism shares the realist's conviction that convergence reflects constraint, not convention. It reframes truth as an emergent structural attractor (the Apex Network) stabilized through an evolutionary process of eliminating failure. The approach calls for epistemic humility, trading the ambition of a God's-eye view for the practical wisdom of a mariner. The payoff is not a final map of truth, but a continuously improving reef chart: a chart built from the architecture of failure, helping us to distinguish the channels of viable knowledge from the hazards of brittle dogma.

## Appendix A: A Mathematical Model of Epistemic Viability

This appendix provides a provisional formalization of core EPC concepts to show that the framework is, in principle, amenable to formal expression. It is crucial to note that these models are illustrative and speculative. **The philosophical argument presented in the main body of the paper is self-contained and does not depend on the validity of any specific mathematical formulation presented here.** The purpose of this appendix is to explore one possible path for future research, not to provide empirical validation for the paper's central claims.

### A.1 Set-Theoretic Foundation

Let **U** be the universal set of all possible atomic predicates. An individual's **Web of Belief (W)** is a subset W ⊆ U satisfying internal coherence condition C_internal:

```
W = {p ∈ U | C_internal(p, W)}
```

**Shared Networks (S)** emerge when agents coordinate to solve problems. They represent the intersection of viable individual webs:

```
S = ∩{W_i | V(W_i) = 1}
```

where V is a viability function (detailed below).

Public knowledge forms nested, intersecting networks (S_germ_theory ⊂ S_medicine ⊂ S_biology), with cross-domain coherence driving convergence.

**The Apex Network (A)** is the maximal coherent subset remaining after infinite pragmatic filtering:

```
A = ∩{W_k | V(W_k) = 1} over all possible contexts and times
```

**Ontological Status:** A is not pre-existing but an emergent structural fact about U, revealed by elimination through pragmatic selection.

**Epistemic Status:** *A is unknowable directly*; it is inferred by mapping failures catalogued in the **Negative Canon** (the historical record of collapsed, high-SBI systems).

**Formal ECHO Extension:** This formalizes Thagard's ECHO extension: net_j = Σ w_{ij} a_i - β·brittleness_j, where w represents positive weights for explanatory coherence (Principle 1) and negative weights for contradiction (Principle 5), with β derived from P(t), C(t) proxies. The specific weight magnitudes would require empirical calibration. Zollman's cycle topology models Pluralist Frontier; complete graphs risk brittleness lock-in.

### A.2 The Systemic Brittleness Index

SBI(t) is a composite index quantifying accumulated systemic costs. We present three functional forms, each with distinct theoretical motivation and testable predictions.

**Key Components:**

**P(t) - Patch Velocity:** Rate of ad-hoc hypothesis accumulation measuring epistemic debt
- Proxy: Increasing share of auxiliary hypotheses vs. novel predictions

**C(t) - Coercion Ratio:** Resources for internal control vs. productive adaptation
- Proxy: (Suppression spending) / (R&D spending)

**M(t) - Model Complexity:** Information-theoretic bloat measure
- Proxy: Free parameters lacking predictive power

**R(t) - Resilience Reserve:** Accumulated robust principles buffering against shocks
- Proxy: Breadth of independent confirmations, age of stable core

**Functional Forms:**

**Form 1: Multiplicative Model**
```
SBI(t) = (P^α · C^β · M^γ) / R^δ
```

**Rationale:** Captures interaction effects where high values in multiple dimensions compound non-linearly. A system with both high complexity AND high patch velocity is more brittle than the sum would suggest. (This form could be used to model compounding effects, where brittleness dimensions interact and accelerate).

**Predictions:** Brittleness accelerates when multiple indicators rise simultaneously. Systems can tolerate high values in one dimension if others remain low.

**Testable implication:** Historical collapses should correlate with simultaneous elevation of 2+ metrics, not single-metric spikes.

**Form 2: Additive Weighted Model**
```
SBI(t) = α·P(t) + β·C(t) + γ·M(t) - δ·R(t)
```

**Rationale:** Assumes independent, additive contributions. Simpler to estimate and interpret; each component has linear effect.

**Predictions:** Each dimension contributes independently. Reducing any single metric proportionally reduces overall brittleness.

**Testable implication:** Interventions targeting single metrics should show proportional improvement.

**Form 3: Threshold Cascade Model**
```
SBI(t) = Σ[w_i · max(0, X_i(t) - T_i)] + λ·Π[H(X_j - T_j)]
```
where X_i ∈ {P, C, M}, H is Heaviside step function, T_i are critical thresholds

**Rationale:** Systems tolerate moderate brittleness but experience catastrophic acceleration once thresholds are crossed. Captures phase-transition dynamics observed in complex systems.

**Predictions:** Brittleness remains low until critical thresholds crossed, then accelerates rapidly. Non-linear "tipping point" behavior.

**Testable implication:** Historical data should show periods of stability followed by rapid collapse once multiple thresholds exceeded.

**Empirical Strategy:**

These forms make distinct predictions testable through historical analysis:
1. Compile brittleness metrics for 20-30 historical knowledge systems (ancient to modern)
2. Code collapse/persistence outcomes
3. Fit each model to historical data
4. Compare predictive accuracy using cross-validation
5. Use information criteria (AIC/BIC) to select best-fitting form

The phlogiston chemistry case (Section 3) illustrates how such data can be assembled from historical records. A full research program would systematically extend this approach. The framework's falsifiability depends on committing to specific functional forms and comparing predictions to data.

### A.3 Dynamics: Stochastic Differential Equations

Knowledge evolution is not deterministic. We model SBI dynamics as:

```
d(SBI) = [α·SBI - β·D(t) + γ·S(t) - δ·R(t) + θ·I(t)]dt + σ·√(SBI)·dW(t)
```

**Deterministic Terms:**

- **α·SBI:** Compounding debt (brittleness begets brittleness)
- **β·D(t):** Systemic debugging (cost-reducing discoveries)
- **+γ·S(t):** External shocks (novel anomalies, pressures)
- **δ·R(t):** Resilience buffer (accumulated robustness)
- **+θ·I(t):** Innovation risk (costs of premature adoption)

**Stochastic Term:**

- **σ·√(SBI)·dW(t):** Brownian motion capturing randomness in discovery timing; volatility increases with brittleness

**Parameter Estimation:**

The parameters α, β, γ, δ, θ, σ are unknowable a priori and must be fitted to historical data. This is not a limitation but standard scientific practice. Proposed empirical strategy:

1. Compile time-series brittleness data for multiple historical systems (as illustrated with phlogiston chemistry)
2. Use maximum likelihood estimation or Bayesian methods to fit parameters
3. Validate on held-out historical cases
4. Test whether fitted model successfully predicts collapse timing for independent test cases

The phlogiston chemistry case provides a template: with systematic bibliometric coding, we can construct d(SBI)/dt trajectories from publication patterns. Parameter estimation would then proceed through standard statistical methods.

**Predictive Utility:**

Once parameters are empirically estimated, the formulation enables probabilistic predictions: "System X has P% chance of crisis within Y years given current trajectory." This transforms brittleness from retrospective diagnosis to prospective risk assessment. The framework's scientific credibility depends on executing this program and comparing predictions to outcomes.

### A.4 Conceptualizing an Empirical Inquiry

The existence of these distinct functional forms suggests a path for future empirical inquiry. A historian or sociologist of science could:
1. Compile qualitative brittleness indicators for a set of historical knowledge systems.
2. Code their outcomes (e.g., persistence, revision, collapse).
3. Assess which of the conceptual models (multiplicative, additive, or threshold) best describes the observed historical patterns.

Such a program would not be about fitting precise numerical data but about determining which conceptual dynamic (compounding interaction, linear addition, or tipping points) best accounts for the historical evolution of knowledge. The framework's value lies in its ability to generate such guiding questions for historical and scientific investigation.

### A.5 The Role of Formalism in Philosophical Diagnosis

Once a model like this were empirically calibrated, it would not function as a predictive algorithm but as a diagnostic tool. Its utility would be to translate complex historical dynamics into a structured formal language, allowing for more precise comparisons between the health of different knowledge systems. For example, it could help answer questions like: "Is system X accumulating costs primarily through conceptual debt (P(t)), or is its fragility masked by coercive power (C(t))?" This transforms brittleness from a retrospective metaphor into a conceptually structured diagnostic, which is the primary philosophical payoff of the formal exercise.

## References

Acemoglu, Daron, and James A. Robinson. 2012. *Why Nations Fail: The Origins of Power, Prosperity, and Poverty*. New York: Crown Business. ISBN 9780307719225.

Ayvazov, Mahammad. 2025. "Toward a Phase Epistemology: Coherence, Response and the Vector of Mutual Uncertainty." *SSRN Electronic Journal*. https://doi.org/10.2139/ssrn.5250197.

Baggio, Guido, and Andrea Parravicini. 2019. "Introduction to Pragmatism and Theories of Emergence." *European Journal of Pragmatism and American Philosophy* XI-2. https://doi.org/10.4000/ejpap.1611.

Baysan, Umut. 2025. "Emergent Moral Non-naturalism." *Philosophy and Phenomenological Research* 110(1): 1–20. https://doi.org/10.1111/phpr.70057.

Begley, C. Glenn, and Lee M. Ellis. 2012. "Raise Standards for Preclinical Cancer Research." *Nature* 483(7391): 531–533. https://doi.org/10.1038/483531a.

Bennett-Hunter, Guy. 2015. "Emergence, Emergentism and Pragmatism." *Theology and Science* 13(3): 337–57. https://doi.org/10.1080/14746700.2015.1053760.

Bennett-Hunter, Guy. 2015. *Ineffability and Religious Experience*. London: Routledge (originally Pickering & Chatto).

Berlin, Brent, and Paul Kay. 1969. *Basic Color Terms: Their Universality and Evolution*. Berkeley: University of California Press.

BonJour, Laurence. 1985. *The Structure of Empirical Knowledge*. Cambridge, MA: Harvard University Press.

Bradie, Michael. 1986. "Assessing Evolutionary Epistemology." *Biology & Philosophy* 1(4): 401–59. https://doi.org/10.1007/BF00140962.

Brandom, Robert B. 1994. *Making It Explicit: Reasoning, Representing, and Discursive Commitment*. Cambridge, MA: Harvard University Press.

Buchanan, Allen, and Russell Powell. 2018. *The Evolution of Moral Progress: A Biocultural Theory*. New York: Oxford University Press.

Campbell, Donald T. 1974. "Evolutionary Epistemology." In *The Philosophy of Karl R. Popper*, edited by Paul A. Schilpp, 413–63. La Salle, IL: Open Court.

Carlson, Matthew. 2015. "Logic and the Structure of the Web of Belief." *Journal for the History of Analytical Philosophy* 3(5): 1–27. https://doi.org/10.22329/jhap.v3i5.3142.

Cartwright, Nancy. 1999. *The Dappled World: A Study of the Boundaries of Science*. Cambridge: Cambridge University Press.

Christensen, David. 2007. "Epistemology of Disagreement: The Good News." *Philosophical Review* 116(2): 187–217. https://doi.org/10.1215/00318108-2006-035.

Dewey, John. 1929. *The Quest for Certainty: A Study of the Relation of Knowledge and Action*. New York: Minton, Balch & Company.

Dewey, John. 1938. *Logic: The Theory of Inquiry*. New York: Henry Holt and Company.

El-Hani, Charbel Niño, and Sami Pihlström. 2002. "Emergence Theories and Pragmatic Realism." *Essays in Philosophy* 3(2): article 3. https://commons.pacificu.edu/eip/vol3/iss2/3.

Fricker, Elizabeth. 2007. *The Epistemology of Testimony*. Oxford: Oxford University Press.

Gadamer, Hans-Georg. 1975. *Truth and Method*. Translated by Joel Weinsheimer and Donald G. Marshall. New York: Continuum (originally Seabury Press; 2nd revised ed.).

Gaifman, Haim, and Marc Snir. 1982. "Probabilities over rich languages, testing and randomness." *Journal of Symbolic Logic* 47(3): 495-548.

Gil Martín, Francisco Javier, and Jesús Vega Encabo. 2008. "Truth and Moral Objectivity: Procedural Realism in Putnam's Pragmatism." *Theoria: An International Journal for Theory, History and Foundations of Science* 23(3): 343-356.

Goldman, Alvin I. 1979. "What Is Justified Belief?" In *Justification and Knowledge: New Studies in Epistemology*, edited by George S. Pappas, 1–23. Dordrecht: D. Reidel. https://doi.org/10.1007/978-94-009-9493-5_1.

Goldman, Alvin I. 1999. *Knowledge in a Social World*. Oxford: Oxford University Press.

Goodman, Nelson. 1983. *Fact, Fiction, and Forecast*. 4th ed. Cambridge, MA: Harvard University Press (originally 1954).

Haack, Susan. 1993. *Evidence and Inquiry: Towards Reconstruction in Epistemology*. Oxford: Blackwell.

Harding, Sandra. 1991. *Whose Science? Whose Knowledge? Thinking from Women's Lives*. Ithaca, NY: Cornell University Press.

Hartmann, Stephan, and Borut Trpin. 2023. "Conjunctive Explanations: A Coherentist Appraisal." In *Conjunctive Explanations: The Nature, Epistemology, and Psychology of Explanatory Multiplicity*, edited by Jonah N. Schupbach and David H. Glass, 111–129. Cham: Springer.

Henrich, Joseph. 2015. *The Secret of Our Success: How Culture Is Driving Human Evolution, Domesticating Our Species, and Making Us Smarter*. Princeton, NJ: Princeton University Press.

Hodge, A. Trevor. 1992. *Roman Aqueducts & Water Supply*. London: Duckworth.

Holling, C. S. 1973. "Resilience and Stability of Ecological Systems." *Annual Review of Ecology and Systematics* 4: 1–23. https://doi.org/10.1146/annurev.es.04.110173.000245.

Hull, David L. 1988. *Science as a Process: An Evolutionary Account of the Social and Conceptual Development of Science*. Chicago: University of Chicago Press.

Kelly, Thomas. 2005. "The Epistemic Significance of Disagreement." In *Oxford Studies in Epistemology*, vol. 1, edited by Tamar Szabó Gendler and John Hawthorne, 167–96. Oxford: Oxford University Press.

Kim, Jaegwon. 1988. "What Is 'Naturalized Epistemology'?" *Philosophical Perspectives* 2: 381–405. https://doi.org/10.2307/2214082.

Kitcher, Philip. 1993. *The Advancement of Science: Science without Legend, Objectivity without Illusions*. New York: Oxford University Press. ISBN 9780195046281.

Kornblith, Hilary. 2002. *Knowledge and Its Place in Nature*. Oxford: Clarendon Press.

Kuhn, Thomas S. 1996. *The Structure of Scientific Revolutions*. 3rd ed. Chicago: University of Chicago Press (originally 1962). ISBN 9780226458083.

Kvanvig, Jonathan L. 2012. "Coherentism and Justified Inconsistent Beliefs: A Solution." *Southern Journal of Philosophy* 50(1): 21–41. https://doi.org/10.1111/j.2041-6962.2011.00090.x.

Ladyman, James, and Don Ross. 2007. *Every Thing Must Go: Metaphysics Naturalized*. Oxford: Oxford University Press.

Lakatos, Imre. 1970. "Falsification and the Methodology of Scientific Research Programmes." In *Criticism and the Growth of Knowledge*, edited by Imre Lakatos and Alan Musgrave, 91–196. Cambridge: Cambridge University Press.

Laudan, Larry. 1977. *Progress and Its Problems: Towards a Theory of Scientific Growth*. Berkeley: University of California Press.

Lehrer, Keith. 1990. *Theory of Knowledge*. Boulder, CO: Westview Press.

Li, Y., et al. 2025. "Pragmatics in the Era of Large Language Models: A Survey on Datasets, Evaluation, Opportunities and Challenges." arXiv preprint arXiv:2302.02689.

Longino, Helen E. 2002. *The Fate of Knowledge*. Princeton, NJ: Princeton University Press.

Lugones, María. 2003. *Pilgrimages/Peregrinajes: Theorizing Coalition against Multiple Oppressions*. Lanham, MD: Rowman & Littlefield.

Mallapaty, Smriti. 2020. "AI models fail to capture the complexity of human language." *Nature* 588: 24. https://doi.org/10.1038/d41586-020-03359-2.

Lynch, Michael P. 2009. *Truth as One and Many*. Oxford: Clarendon Press.

Meadows, Donella H. 2008. *Thinking in Systems: A Primer*. Edited by Diana Wright. White River Junction, VT: Chelsea Green Publishing.

Mesoudi, Alex. 2011. *Cultural Evolution: How Darwinian Theory Can Explain Human Culture and Synthesize the Social Sciences*. Chicago: University of Chicago Press.

MLPerf Association. 2023. "MLPerf Training Results." https://mlcommons.org/benchmarks/training/.

Moghaddam, Soroush. 2013. "Confronting the Normativity Objection: W.V. Quine's Engineering Model and Michael A. Bishop and J.D. Trout's Strategic Reliabilism." Master's thesis, University of Victoria. http://hdl.handle.net/1828/4915.

Newman, Mark. 2010. *Networks: An Introduction*. Oxford: Oxford University Press.

Olsson, Erik J. 2005. *Against Coherence: Truth, Probability, and Justification*. Oxford: Oxford University Press.

Patterson, Orlando. 1982. *Slavery and Social Death: A Comparative Study*. Cambridge, MA: Harvard University Press.

Peirce, Charles S. 1992. "How to Make Our Ideas Clear." In *The Essential Peirce: Selected Philosophical Writings*, vol. 1 (1867–1893), edited by Nathan Houser and Christian Kloesel, 124–41. Bloomington: Indiana University Press (originally 1878).

Peter, Fabienne. 2024. "Moral Affordances and the Demands of Fittingness." *Philosophical Psychology* 37(7): 1948–70. https://doi.org/10.1080/09515089.2023.2236120.

Popper, Karl. 1959. *The Logic of Scientific Discovery*. London: Hutchinson (originally 1934).

Price, Huw. 1992. "Metaphysical Pluralism." *Journal of Philosophy* 89(8): 387–409. https://doi.org/10.2307/2940975.

Pritchard, Duncan. 2016. "Epistemic Risk." *Journal of Philosophy* 113(11): 550–571. https://doi.org/10.5840/jphil20161131135.

Psillos, Stathis. 1999. *Scientific Realism: How Science Tracks Truth*. London: Routledge.

Putnam, Hilary. 2002. *The Collapse of the Fact/Value Dichotomy and Other Essays*. Cambridge, MA: Harvard University Press.

Quine, W. V. O. 1951. "Two Dogmas of Empiricism." *Philosophical Review* 60(1): 20–43. https://doi.org/10.2307/2181906.

Quine, W. V. O. 1960. *Word and Object*. Cambridge, MA: MIT Press. ISBN 9780262670012.

Rescher, Nicholas. 1996. *Process Metaphysics: An Introduction to Process Philosophy*. Albany: State University of New York Press.

Roda, Weston C., Marie B. Varughese, Donglin Han, and Michael Y. Li. 2020. "Why Is It Difficult to Accurately Predict the COVID-19 Epidemic?" *Infectious Disease Modelling* 5: 271–281. https://doi.org/10.1016/j.idm.2020.03.001.

Rorty, Richard. 1989. *Contingency, Irony, and Solidarity*. Cambridge: Cambridge University Press.

Rosenstock, Sarita, Cailin O'Connor, and Justin Bruner. 2017. "In Epistemic Networks, Is Less Really More?" *Philosophy of Science* 84(2): 234–52. https://doi.org/10.1086/690717.

Rottschaefer, William A. 2012. "Moral Realism and Evolutionary Moral Psychology." In *The Oxford Handbook of Moral Realism*, edited by Paul Bloomfield, 123–45. Oxford: Oxford University Press.

Rottschaefer, William A. 2012. "The Moral Realism of Pragmatic Naturalism." *Analyse & Kritik* 34(1): 141–56. https://doi.org/10.1515/ak-2012-0107.

Russell, Bertrand. 1903. *The Principles of Mathematics*. Cambridge: Cambridge University Press. ISBN 9780521062749.

Ladyman, James, and Don Ross. 2007. *Every Thing Must Go: Metaphysics Naturalized*. Oxford: Oxford University Press. ISBN 9780199573097.

Rorty, Richard. 1979. *Philosophy and the Mirror of Nature*. Princeton, NJ: Princeton University Press.

Sevilla, Jaime, Lennart Heim, Anson Ho, Tamay Besiroglu, Marius Hobbhahn, and Pablo Villalobos. 2022. "Compute Trends Across Three Eras of Machine Learning." arXiv preprint arXiv:2202.05924.

March, James G. 1978. "Bounded Rationality, Ambiguity, and the Engineering of Choice." *The Bell Journal of Economics* 9, no. 2: 587–608. https://doi.org/10.2307/3003600.

Sims, Matthew. 2024. "The Principle of Dynamic Holism: Guiding Methodology for Investigating Cognition in Nonneuronal Organisms." *Philosophy of Science* 91(2): 430–48. https://doi.org/10.1017/psa.2023.104.

Snow, John. 1855. *On the Mode of Communication of Cholera*. London: John Churchill.

Krag, Erik. 2015. "Coherentism and Belief Fixation." *Logos & Episteme* 6, no. 2: 187–199. https://doi.org/10.5840/logos-episteme20156211.

Staffel, Julia. 2020. "Reasons Fundamentalism and Rational Uncertainty – Comments on Lord, The Importance of Being Rational." *Philosophy and Phenomenological Research* 100, no. 2: 463–468. https://doi.org/10.1111/phpr.12675.

Taleb, Nassim Nicholas. 2012. *Antifragile: Things That Gain from Disorder*. New York: Random House. ISBN 9781400067824.

Tauriainen, Teemu. 2017. "Quine on Truth." *Philosophical Forum* 48(4): 397–416. https://doi.org/10.1111/phil.12170.

Tauriainen, Teemu. 2017. "Quine's Naturalistic Conception of Truth." Master's thesis, University of Jyväskylä.

Turchin, Peter. 2003. *Historical Dynamics: Why States Rise and Fall*. Princeton, NJ: Princeton University Press.

Worrall, John. 1989. "Structural Realism: The Best of Both Worlds?" *Dialectica* 43(1–2): 99–124. https://doi.org/10.1111/j.1746-8361.1989.tb00933.x.

Watson, James D., and Francis H.C. Crick. 1953. "Molecular Structure of Nucleic Acids: A Structure for Deoxyribose Nucleic Acid." *Nature* 171(4356): 737–738. https://doi.org/10.1038/171737a0.

Wright, Sewall. 1932. "The Roles of Mutation, Inbreeding, Crossbreeding and Selection in Evolution." *Proceedings of the Sixth International Congress of Genetics* 1: 356–66.

Zagzebski, Linda Trinkaus. 1996. *Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical Foundations of Knowledge*. Cambridge: Cambridge University Press.

Zollman, Kevin J. S. 2013. "Network Epistemology: Communication in the History of Science." *Philosophy Compass* 8(1): 15–27. https://doi.org/10.1111/phc3.12021.