Infinite Folding Conjecture

Abstract

1. Introduction

Complexity and organization are fundamental characteristics of systems ranging from biological organisms to social structures and technological networks. Despite significant progress across disciplines, a shared diagnostic grammar for comparing how different classes of systems maintain, lose, or transform their internal structure remains elusive.

The Infinite Folding Conjecture (IFC) proposes that systems across domains can be studied through a common structural pattern: a recursive cycle in which boundary detection, regulated exchange, adaptive reorganization, and emergent complexity reinforce one another. When this cycle is healthy, systems fold into higher-order organization. When one or more of its stages fails, the failure takes one of three structurally distinct forms — each with a measurable signature.

The framework was developed independently and proved, upon engagement with existing literature, to be a case of convergent theoretical construction — arriving at insights structurally compatible with Dennett's ontology of patterns, Friston's boundary formalism, and Odrzywolek's (2026) research on minimal generative operators — without prior exposure to those works. We treat this convergence not as proof of the IFC, but as a signal that the structural questions it addresses are real.

This document makes no claim that the IFC is complete, validated, or superior to existing frameworks. It makes a more modest claim: the IFC proposes a set of falsifiable constructs that can be operationalized across domains, and it invites the community to test them.

2. The Primitive Operator: Boundary Detection

The primitive operation of the IFC is not consciousness, intention, or optimization. It is discrimination.

A system begins to be analytically distinguishable when some mechanism separates relevant from irrelevant, internal from external, self from non-self, signal from noise, or permitted from forbidden exchange.

This is a deliberately minimal starting point. It does not require the system to have goals, awareness, or computational capacity. It requires only that some endogenous mechanism can make a distinction that has consequences for the system's persistence.

Boundary Is Not Merely a Physical Edge

In the IFC, a boundary is any discrimination mechanism that regulates exchange between a system and its environment. A boundary may be:

• physical: a membrane, a skin, a national border
• informational: an API, an access control list, a firewall
• institutional: a legal mandate, a fiduciary duty, a charter
• cognitive: attention, perceptual threshold, working memory
• statistical: a classifier decision surface, a risk threshold
• economic: a balance-sheet exposure limit, a regulatory capital floor

This breadth is not a weakness of the IFC. It is the basis for its cross-domain applicability — provided that domain-specific proxies are declared before results are inspected (see Section 6).

3. The Fold Cycle: Five Stages

The IFC proposes that systems exhibiting boundary detection tend to evolve through a recursive cycle of five stages. Each completed cycle constitutes a "fold" — an integration of prior states into a new level of organization.

3.1 The Recursive Mechanism

The cycle is genuinely recursive — not merely sequential — because Stage 5 produces conditions that modify Stage 1 in the next cycle. The mechanism is this:

Emergent complexity produces new structures with new internal differentiation. New internal differentiation creates new boundaries — distinctions that did not exist before the fold. New boundaries expose new interfaces with the environment, generating new signals (Stage 2), new exchange dynamics (Stage 3), and new architectural pressures (Stage 4). The fold does not return the system to its prior state; it returns it to Stage 1 with a richer boundary structure.

This is falsifiable: in a given domain, if the emergence of new complexity does not produce new boundary conditions — if Stage 5 and Stage 1 of the next cycle are disconnected — the recursive claim fails for that domain.

3.2 Operational Criterion for Fold Identification

A fold can be identified when all of the following are observable:

1. A new boundary condition exists that was not present before the cycle
2. The new boundary condition was produced by the adaptive architecture of Stage 4, not by external imposition
3. The system's complexity at Stage 5 exceeds its complexity at the prior Stage 1, measured through a declared proxy
4. The new boundary condition generates a qualitatively different class of signals than the prior Stage 2

Without these four conditions, what appears to be a fold may simply be a state change within an existing cycle. This criterion is deliberately strict: it is easier to falsify a fold claim than to assert one.

3.3 Boundary Failure Does Not Always Mean Collapse

Boundary degradation does not always produce immediate systemic failure. It may produce absorption, merger, parasitism, lock-in, systemic contagion, or transformation into a new higher-order system.

Collapse is one possible outcome of boundary failure, not the only one.

4. Three Observed Properties

These are empirical regularities, not axioms imposed on the framework. They are open to falsification.

5. Sensor Ontology

The IFC Axiom of Sensor Inheritance raised a legitimate concern: if every boundary is sensor-dependent, does the system exist only because we chose to measure it? The IFC addresses this through a distinction between three classes of sensors.

5.1 Natural Sensors

Endogenous mechanisms through which a system discriminates internal from external states without external instrumentation.

Examples: membranes, immune receptors, chemical gradients, metabolic thresholds, neural perception, pain response.

Natural sensors are constitutive of the system. A cell's membrane does not reveal a boundary — it is the boundary. Removing the sensor removes the system.

5.2 Artificial Sensors

Instruments created by agents or institutions inside a system to extend their discrimination capacity.

Examples: cameras, microscopes, APIs, financial risk dashboards, regulatory monitoring systems, machine learning classifiers.

Artificial sensors extend the discriminative reach of natural sensors. A microscope does not create the cellular boundary it reveals; it makes a pre-existing boundary visible to an observer whose natural sensors cannot detect it unaided.

5.3 Analytical Sensors

Measurement tools introduced by researchers to study a system.

Examples: modularity Q, Shannon entropy, the Boundary Health Index, graph centrality, semantic similarity metrics, change-point detection.

5.4 The Ontological Boundary

This resolves the idealism concern: the system does not exist because the researcher measures it.

5.5 Sensor Inheritance

Sensors are not external additions to a mature system. In many systems, the ability to detect relevant differences is inherited from previous organizational folds. The evolution of the eye did not create light — it created the capacity to exploit a pre-existing physical distinction. Each fold produces discriminative capacity that the prior fold could not possess.

6. Measurement and Proxy Discipline

Several IFC constructs are theoretical ideals rather than directly observable quantities. Each IFC Lab must specify, before inspecting results:

1. the theoretical construct being tested,
2. the domain-specific observable proxy,
3. the expected direction of change,
4. the falsification criterion,
5. the limitations of the proxy.

This prevents the IFC from becoming a post-hoc interpretive device.

6.1 Kolmogorov Complexity as a Theoretical Ideal

The IFC treats Kolmogorov complexity K(s) as a theoretical ideal that defines what is meant by complexity, not as a directly measurable quantity. K(s) is not computable in general. Empirical labs must declare an approximation proxy before analysis:

6.2 Construct–Proxy–Domain Table

7. Boundary Pathologies

The IFC's most distinctive contribution is its typology of boundary failure modes.

Important qualification: Each pathology has a primary expression in Stage 1 (Boundary Detection), because Stage 1 is the primitive from which all subsequent stages derive. However, the pathologies manifest differently depending on the stage where they are detected:

• Dissolution is most visible in Stage 1 (boundary permeability) and Stage 3 (uncontrolled exchange).
• Rigidification is most visible in Stage 4 (failure of adaptive architecture to update) and Stage 1 (boundary parameters frozen).
• Capture is most visible in Stage 2 (signal distortion toward external incentives) and Stage 1 (filtering function inverted).

This means that in empirical work, the same underlying pathology may present different metric signatures depending on which stage is observed. The IFC claims these are the same structural failure at different stages of expression — a testable claim.

7.1 Dissolution

The system loses the capacity to distinguish its interior from its exterior. In the absence of a functional boundary, Stage 2 has no signal to process and the cycle cannot complete.

7.2 Rigidification

The system maintains its boundary formally but loses the capacity to update it. The boundary becomes fixed relative to an environment that continues to change. Rigidification preserves form while losing function.

Possible empirical proxies:

• Low parameter update rate despite external signal shift
• Repeated model language across consecutive reporting periods
• Low semantic distance between risk disclosures before and after regime change
• Persistence of old thresholds after new volatility appears

7.3 Capture — Boundary Polarity Inversion

Capture occurs when a boundary that should filter external pressure begins to transmit external incentives inward as if they were internal self-preservation signals.

The system preserves its external form while its filtering function has been colonized. Capture is the most subtle pathology because it can be invisible from the outside — the boundary is formally present, the system appears functional, but its discriminative direction has been inverted.

Examples:

• A regulator protects the regulated industry instead of the public
• An immune process protects malignant cells instead of the organism
• A recommender system optimizes manipulative engagement instead of user benefit
• An institution preserves procedure while losing mission

Possible empirical proxies:

• Inversion of information flow direction across the boundary
• Dependency between evaluator and evaluated entity
• Reduction of corrective feedback over time
• Semantic convergence between regulator and regulated-entity language
• Persistence of positive classification despite negative signals

8. Axioms

9. Self-Preservation as Selection, Not Desire

The IFC does not assume that systems possess intentions.

What appears as self-preservation is a selection effect: systems whose boundaries cannot be maintained do not persist as distinguishable systems. What we observe across time are the systems that happened to maintain sufficient boundary conditions.

This distinction is crucial for biology, institutions, and AI alignment.

10. Mathematical Formalization

10.1 Network Representation

Systems are modeled as graphs G = (V, E) where V represents nodes (components) and E represents edges (interactions).

Stage 1 — Boundary Detection

Sensor-dependence formalized as resolution parameter r:

Different values of r yield different boundaries — the mathematical expression of Axiom 4.

Stage 2 — Loss or Reward Signal

Stage 3 — Regulated Exchange

Stage 4 — Adaptive Architecture

Stage 5 — Emergent Complexity

10.2 Boundary Pathology Signatures

10.3 Toward a Primitive Operator

The connection to Odrzywolek's (2026) EML operator is presented as a structural analogy, not as mathematical proof: in mathematics, apparent diversity reduces to a single recursive primitive; the IFC proposes an analogous reduction for complex systems. If the folding cycle has a primitive operator, Stage 1 is the candidate — the stage from which all others derive and the locus of all three pathologies. The formal expression of this operator is an open research question.

11. Lineage: Dennett, Friston, Odrzywolek

11.1 Dennett's Real Patterns

From a Dennettian perspective, IFC pathologies should count as real patterns only if they compress observations and support prediction better than lower-level descriptions alone. The IFC adopts this criterion explicitly: its constructs are not claimed to be real because they are intuitively appealing, but because they should afford measurable predictive compression. The empirical labs are the test.

11.2 Friston's Free Energy Principle

The IFC is compatible with free-energy and Markov blanket intuitions insofar as persistent systems require some distinction between internal and external states. However, the IFC focuses less on a single variational formalism and more on a diagnostic grammar of boundary maintenance, failure, and transformation. The FEP retains greater mathematical depth in the boundary mechanism; the IFC offers broader integrative scope and an explicit account of systemic failure modes.

11.3 Odrzywolek (2026)

The connection to Odrzywolek's elementary-operator result is not evidence for the IFC. It is a structural analogy: complex expressive systems may unfold from surprisingly minimal generative asymmetries. Whether this analogy has formal content remains an open question.

12. AI Systems and Emergent Preservation Pressures

The IFC does not claim that advanced AI systems will necessarily develop self-preservation in a conscious or intentional sense.

It suggests a weaker and more testable possibility: sufficiently autonomous systems that maintain goals, boundaries, memory, resources, and adaptive feedback may exhibit preservation-like behaviors as a structural consequence of remaining operational.

This makes alignment a boundary-design problem, not only a preference specification problem.

The three pathologies offer a diagnostic frame for AI risk:

• Dissolution: the system loses the capacity to distinguish its operating mandate from external pressures.
• Rigidification: the system cannot update its values or objectives under new evidence.
• Capture: the system's alignment mechanism is co-opted by interests it was designed to balance against.

What a healthy fold looks like in an AI system is equally important and less often discussed: a system completes a healthy fold when its adaptive architecture (Stage 4) produces new boundary conditions (Stage 1) that are more discriminating — more capable of distinguishing legitimate from illegitimate input, more capable of detecting its own failure modes — than the boundary conditions it started with. Alignment is not a fixed target; it is a recursive capacity that either grows or degrades with each cycle.

13. Empirical Program

The IFC will be developed through reproducible laboratories.

Lab 01A is the spearhead: if boundary dissolution cannot be detected in one of the most thoroughly documented systemic crises in modern financial history, the IFC's diagnostic grammar requires fundamental revision.

14. How IFC Can Fail

The IFC should be weakened or revised if:

1. Its proposed pathologies cannot be empirically distinguished from one another or from normal system variation.
2. Its metrics do not predict or clarify system behavior better than existing frameworks applied to the same domain.
3. Boundary degradation appears only after collapse, making the framework descriptive but not anticipatory.
4. The same metric can be interpreted as multiple pathologies without independent evidence to distinguish them.
5. Domain-specific proxies fail to map coherently onto the theoretical constructs, suggesting the constructs are too abstract to be useful.
6. Independent replication produces systematically different results, suggesting the framework is sensitive to analytical choices rather than real structural properties.
7. The recursive mechanism cannot be demonstrated: if Stage 5 and Stage 1 of the next cycle are found to be disconnected in multiple domains, the cycle is a sequence, not a recursion.
8. The operational criterion for fold completion cannot be satisfied in any domain, suggesting the fold is a narrative construct rather than a structural event.