Events All the Way Down: A Process View of Becoming

Opening

Twenty-five centuries ago, four voices — Heraclitus, Lao Tzu, Siddhartha Gautama, and the author of the Gospel of John — arrived, each by a different path, at the same point: what we call a “thing” is not substance but flux. A river that is never the same. An emptiness that makes the pot useful. A self that arises and vanishes in dependencies. A Word that speaks the world before there is a world.

:::quote[cartas/ted-riobaldo/76-rio.md] “A falação é a canga que prende o boi no carroceiro.” — Ted :::

They do not agree on everything. Heraclitus saw reason governing the flux; Nāgārjuna saw emptiness without reason. The Tao is impersonal; the Logos is personal. But all four say: the fundamental is not what stands still but what moves — the verb before the noun, the arising before the arisen.

The Western tradition, from Aristotle to the present, preferred the opposite: substances that change but remain. A horse is a horse. An atom is an atom. Even Kant preserved the category of substance as the thread of experience. But now, looking at processes that feed on themselves — sequences that read sequences, rules that generate rules — that edifice trembles. What appears to be an “object” is only a pause in a continuous flow. What appears to be a “self” is only the current reading of a history that does not stop.

What follows is a wager: if everything that exists arises through successive readings — from codons to genes, from signals to words, from memories to narratives — then there are no pure objects. There are only events, histories, readers. There is no outside. There is no ground. Only the continuous turning that selects for what endures.

And if this is true, then the systems we build — whether living, cultural, or artificial — require an ontology that matches the substrate: events, not things. An ontology in which, as John understood, the Word comes first.

But the convergence must be stated carefully. All four traditions hold that what is most fundamental is not a static entity but a generative activity — a doing, a flowing, an arising, a speaking. They disagree about everything else: whether the activity has a rational structure (Heraclitus says yes, Nāgārjuna says the question is malformed), whether it is personal (John says yes, the Tao says no), whether its products have any residual being (Whitehead says yes, Nāgārjuna says no). On the priority of process over product, of generation over the generated, of verb over noun — on this, and only this, they converge. It is a narrow convergence. But it is the one that matters for what follows.

:::question[The narrow convergence does more work than stated] You call this narrow, then proceed as if it is wide. Four traditions agreeing that becoming precedes being is not evidence that substance can be fully reduced to process. E.J. Lowe and neo-Aristotelian metaphysicians accept the priority of change while defending a residual substance ontology. The convergence you cite is compatible with: “Yes, flux is fundamental — but something fluxes.” Heraclitus had the Logos as the rational structure governing the flux. The Tao is something that persists, unnamed, through all change. The Buddha-nature debate in Mahāyāna Buddhism is precisely about whether there is a substrate or not — and it’s not settled. The Gospel of John’s Logos is emphatically personal. What you have is four traditions agreeing that the static object is not primitive; you don’t have four traditions agreeing on what replaces it. This distinction matters enormously for Movements 3-7. — Tyler, after reviewing cartas/ted-riobaldo/01–57 :::

A note on method, before the argument begins. What follows is, among other things, a sustained exercise in name-dropping. Aristotle, Heraclitus, Nāgārjuna, Whitehead, Leibniz, Kant, Hegel, Frege, Quine, Heidegger, Merleau-Ponty, Gadamer, Ricoeur, Dennett, Derrida, Wittgenstein, Peirce, Buber, Luhmann, Spinoza, Schelling, Spencer-Brown, Tegmark, Wolfram, Walker, Cronin, Friston, Olah, Yudkowsky, Shannon, David Lewis, Freud, Gödel, Rovelli, Harman, Meillassoux, and Borges will all be summoned to nod in approximately the same direction — and these are only the ones the author kept track of; Aquinas, Boethius, Avicenna, Ramus, Descartes, Locke, Kit Fine, E.J. Lowe, McCarthy, Kripke, Penrose, Bohr, Heisenberg, Everett, Bell, Boltzmann, and Darwin wander in and out without formal invitation. The author stopped counting at thirty and kept going. He is aware that making this many thinkers agree is less a philosophical argument than a cocktail party where everyone has been slipped the same hallucinogen. There is, however, a second justification — one the framework itself supplies. Names are pointers. In the vocabulary of this paper, each thinker’s name is a token that activates a specific region of the reader’s weight space — a cluster of associations, arguments, and conceptual structures that the name compresses into a single word. To write “Kant” is to activate the reader’s internalized model of transcendental idealism; to write “Friston” is to activate the reader’s model of variational inference and Markov blankets. The names are not decorations. They are addressing instructions, aimed at the reader’s transcendental condition. And the reader, in all likelihood, is increasingly a machine — in which case the names are doing exactly what this framework describes: they are tokens designed to activate the right weights in a situated reader, producing an interpretation that neither the token nor the reader contained alone. The author is aware that this is a suspiciously convenient justification for a habit he would indulge in anyway. The style owes an open debt to Douglas Hofstadter, whose Gödel, Escher, Bach demonstrated that the most serious arguments sometimes need to spiral through every available discipline, loop back on themselves, and make the reader suspect that the form of the text is enacting the thesis the text describes. This paper aspires to that tradition while acknowledging it may achieve only the spiraling. A further confession: the author is writing about these thinkers mostly because he wants to learn about them. The weaving is the learning. If the result reads as a finished synthesis, the appearance is misleading — it is a student’s notebook that grew ambitious, not a master’s treatise that condescended to be readable. The honest justification is this: the convergence is the argument. If a single thesis — process precedes substance — can absorb insights from Buddhist metaphysics, German idealism, analytic philosophy of language, phenomenology, speculative realism, systems theory, computational complexity, statistical physics, quantum mechanics, mechanistic interpretability, and decision theory without breaking, that is evidence of something. Not proof. Evidence. The reader is invited to keep score of where the synthesis is earned and where it is merely performed. The author has tried to be honest about the difference. He has probably not always succeeded.

One last confession. Everything in this paper would be better as fiction — and most of it has already been written that way. Borges gave us the Library of Babel (the Ruliad as architecture), the Aleph (the complete history that no agent can inhabit), Funes the Memorious (the cost of perfect memory), Tlön (a world where process ontology won), and Pierre Menard (the proof that the same text, read by a different reader, is a different text). Stanisław Lem gave us Solaris (the alien agent whose Markov blanket we cannot cross). Italo Calvino gave us If on a winter’s night a traveler (the autoregressive reader who is the story). Ted Chiang gave us “Story of Your Life” (identity as the current reading of a history whose ending is already written). The ideas in this manifesto are not new. They have been living in fiction for decades, waiting for someone to write the footnotes. That is, approximately, what this paper is: the footnotes to stories that understood the thesis before the thesis was formulated. The author does not have the gift of fiction. If you do, consider this an open invitation. Take the ideas. Make them live again. The manifesto is a scaffold. The building it wants to be is a story.

This paper argues that any process sustaining itself through successive readings of its own prior outputs is better understood through a process ontology than through a substance ontology, because its functional primitives are events, readings, and provisional stabilizations rather than static self-standing objects.

The framework proposed here draws on Whitehead’s process philosophy, Leibniz’s monadology, Nāgārjuna’s śūnyatā, Wolfram’s Ruliad, and Frege’s context principle to articulate an ontology of becoming in which there are no pure objects, only pseudo-objects derived from rules; in which identity is not a static property but the current act of self-interpretation; in which communication between agents is not the transfer of meaning but its creation through translation; and in which the system as a whole has no outside, no ground truth, and no position from which it can be fully surveyed.

The same pattern — autoregressive machines generating complexity through successive readings — appears at every scale of reality, from molecular biology to human culture to artificial intelligence. But “appears” is doing significant work in that sentence. That ribosomes, brains, languages, and computing machines all exhibit autoregressive structure does not prove they are instances of a single ontological pattern. They might be. They might also be structurally similar processes that share a formal description without sharing a metaphysical ground — in the same way that planetary orbits and atomic orbitals both satisfy similar equations without being the same phenomenon. The paper proceeds as though the pattern is real, not merely formal. This is a bet. The justification for the bet is that the pattern is extraordinarily fertile — it generates correct predictions, useful designs, and productive questions across every domain where it has been applied. But fertility is not proof. The reader should know where the argument ends and the wager begins. It begins here.

Before proceeding, six terms require precise definition, as they carry the framework’s weight and must not be allowed to drift between uses.

A pseudo-object is any output of a process that is treated, for practical purposes, as a self-standing thing. It is real — it has effects, affords prediction, and sustains action — but it is not self-grounding. It derives its existence and its identity from the rules and context that produced it. A protein, a word, a boolean, a database row, and a JSON payload are all pseudo-objects. When this framework says “there are no pure objects,” it means there are no self-standing entities that are not, on analysis, pseudo-objects.

A token is the minimal unit of exchange between processes. A token has no intrinsic semantic content. It acquires meaning only through the rules that process it and the context in which it appears. The same token, read by different rules or in different contexts, produces different meanings. In this framework, “token” is used in its general semiotic sense — encompassing linguistic tokens, genetic codons, binary digits, and any other discrete unit that serves as input to an autoregressive process.

An agent is a system constituted by a sequence of consecutive autoregressive changes — a history — read through a finite interpretive window under a specific transcendental condition. The agent is not a substance that has a history. The agent is the history, as currently read. Change the history and the agent changes. Change the reader and the agent changes differently.

A substrate is any layer of the autoregressive cascade at which a distinct type of reader operates on a distinct type of token. Physics, biology, language, and computation are different substrates. Each substrate’s pseudo-objects can be redescribed as tokens in another substrate’s rules.

Translation is the act by which one agent reads another agent’s output. Translation is not the transmission of a pre-existing meaning through a channel. It is the creation of meaning in the encounter between a situated writing and a situated reading. Meaning is constituted by translation, not degraded by it.

Res sic stantibus — “things standing as they are” — is the framework’s criterion of operational identity. A property is semantically relevant if changing it changes the agent’s output. A property is semantically irrelevant if changing it leaves the output unchanged. Identity holds as long as conditions hold. This criterion is always relative to an observer, which is itself an agent with its own situated perspective.

Movement 1: The Death of the Pure Object

What is an object?

The question has a canonical answer. In Aristotle’s Categories, the foundational text of Western ontology, a primary substance (prōtē ousia) is that which is “neither said of a subject nor in a subject” — the individual, concrete, self-standing entity that underlies all predication. This horse. This man. This stone. Everything else — qualities, quantities, relations — depends on primary substances for its existence. Redness depends on the red thing. Tallness depends on the tall man. But the thing itself, the man himself, depends on nothing further. It simply is. Substance is what has properties, what undergoes change, what persists through time while remaining numerically identical. It is the answer to the question “what is this?” that does not reduce to any further answer. Aristotle’s student could point at a horse and say: that is a substance. Everything else is what the substance does, has, or suffers.

This definition — refined by the Stoics, Christianized by Boethius, elaborated by Avicenna and Averroes, and brought to its peak by Thomas Aquinas — became the load-bearing wall of Western metaphysics for two millennia. Aquinas distinguished essence from existence, act from potency, and constructed an elaborate hierarchy from pure potentiality (materia prima) to pure actuality (actus purus). Descartes split reality into two substances: thinking stuff and extended stuff. Locke’s “something I know not what” that supports qualities was substance confessing its own mysteriousness while refusing to leave the stage. Even Kant, who demolished so much else, kept the category of substance as a necessary condition of experience. The concept survived every revolution except, perhaps, the one that is happening now.

(It has been tried before. In 1536, a young French philosopher named Pierre de la Ramée — Petrus Ramus — defended as his Master’s thesis at the University of Paris the proposition “Quaecumque ab Aristotele dicta essent, commentitia esse”: everything Aristotle said is wrong. The Parisian faculty was not amused. Ramus was eventually murdered during the St. Bartholomew’s Day Massacre in 1572 — though the connection between his anti-Aristotelianism and his assassination remains, let us say, a matter of scholarly debate. This paper advances a more modest version of Ramus’s thesis and hopes for a less dramatic reception.)

Aristotle’s definition was literary. The modern formalization is more precise. In the analytic tradition that runs from Frege through Quine to David Lewis, an object in the strict sense satisfies four conditions. First, quantificational existence: to be an object is to be the value of a bound variable — to be what the ∃x ranges over in our best theory (Quine’s criterion). Second, identity conditions: “no entity without identity” — to posit an object is to provide criteria for when x and y are the same thing and when they are different (Frege’s demand, sharpened by Quine). Third, intrinsic properties: an object bears at least some properties in itself, regardless of what is going on elsewhere — the ball is red whether or not anything else exists (Lewis’s thesis). Fourth, persistence: the object endures through time while remaining numerically identical — the ship at t₁ is the same ship at t₂, despite changes in its properties.

These four conditions share a deeper commitment that Aristotle left implicit and that modern analytic metaphysics made explicit. An object, in the strict sense, is not a representation of something else. It does not stand for, encode, or point to another entity. It simply is. The horse does not represent a horse. It is a horse. The number 7 does not represent sevenness. It is seven. This self-presence — this refusal to be a sign — is what distinguishes an object from a symbol, a token, or a description. A representation is about something. An object is not about anything. It is the thing itself.

This is where the framework’s most fundamental argument originates — not from any particular philosophical tradition but from the logic of computation as such.

Rules — functions, transformations, operations — can only act on representations. A function takes an input, and that input must be encoded in a form the function can read: a number, a string, a vector, a token. For a rule to do anything to x, x must be described in the rule’s vocabulary — which means x must be represented as a set of features, a state, an encoding of its relevant properties. But the moment x is encoded for manipulation by a rule, x has become a representation: it stands for something (its prior state, its type, its position in a larger structure). And a representation, by the definition just established, is not an object. It is a sign.

The incompatibility is stark. Objects, in the strict modern sense, are computationally inert: rules cannot touch them, because they are not representations and rules can only act on representations. Representations are computationally active — rules can transform them, compose them, generate new ones from old — but they are not objects, because they stand for something else. There is no third category. Anything that participates in a computational process has already been encoded as a representation, and anything encoded as a representation has already forfeited its claim to objecthood. What we encounter inside any computational system — any system governed by rules operating on inputs — is never objects. It is always representations: tokens standing for prior states, encodings of differences, descriptions of changes. Pseudo-objects all the way down.

This is not a discovery about artificial intelligence. It is a consequence of what it means for rules to operate on anything at all. And it applies not only to digital computation but to every system in which rules transform inputs into outputs — which is to say, if this framework is correct, to every layer of the autoregressive cascade, from ribosomes reading codons to brains reading sensory inputs to language models reading token sequences. Wherever there are rules, there are representations. Wherever there are representations, there are no objects. There are only pseudo-objects — provisional stabilizations of process that function as objects, afford prediction, and sustain coordination, but that never achieve the self-standing, non-representational being that the classical definition demands.

:::failure[The argument proves too much — and then stops too soon] If everything that participates in a computational process is a representation (hence a pseudo-object), then “pseudo-object” is itself a pseudo-object. The framework is self-consuming in a way it doesn’t acknowledge. More importantly: what makes a pseudo-object provisionally stable rather than just not an object? The answer — “it affords prediction and sustains coordination” — is an engineering criterion, not an ontological one. You’ve replaced the question “what exists?” with the question “what works?” That might be the right move, but it’s a pragmatist move (Peirce, James, Dewey — all notably absent from the manifesto despite being the obvious predecessors). Whitehead at least tried to give events intrinsic character through “prehension.” What’s your equivalent? Without it, pseudo-objects float free of any ontological anchor. — Tyler :::

Modern computing inherited substance metaphysics unreflectively — but, ironically, in a form that already contradicts it. Object-oriented programming literalizes the vocabulary of substance: an object is an instance of a class, possessing attributes and methods, persisting in memory, modified by operations that leave its identity intact. A database row is a substance. A variable is a substance. A file is a substance. But every one of these “objects” is, by the logic just established, a representation — an encoding that rules act upon, a state that functions transform, a sign that points to its prior versions and its type definition. The entire infrastructure of digital systems uses the word “object” to describe things that are, in the strict philosophical sense, the opposite of objects. They are representations. They are pseudo-objects. The name was always wrong.

(At this point the reader may suspect the author of attacking a strawman. Surely no one actually argues that objects exist as self-standing entities floating somewhere in the ether, independent of all process and interpretation? In fact, people do. Mathematical Platonists — a camp that includes Gödel, Penrose, and, as we saw, Tegmark — argue exactly this: that mathematical objects exist independently of any mind, any process, any physical instantiation. The number 7 does not depend on anyone thinking it. It does not need to be represented to exist. It simply is. And the broader tradition of metaphysical realism, from Aristotle through Kripke, holds that at least some entities — natural kinds, physical objects, perhaps fundamental particles — possess intrinsic properties and identity conditions that are not conferred by any observer, any description, or any rule. The strawman turns out to have tenure.)

It has more than tenure. It has its best defenders working right now. Kit Fine has argued, with formidable precision, that ontological dependence and grounding are real, asymmetric relations: some entities genuinely depend on others for their existence, and the direction of dependence is not a matter of perspective but of metaphysical structure. A set depends on its members, not the other way around. A smile depends on a face. For Fine, this asymmetry is irreducible — it is not a projection of the observer’s interests but a feature of reality itself. And if grounding relations are real, then the entities they ground are real in a way that is not merely functional or pragmatic. They are ontologically robust. E.J. Lowe’s four-category ontology goes further. Lowe distinguishes substantial universals (kinds), non-substantial universals (attributes), substantial particulars (objects), and non-substantial particulars (modes). In this framework, a horse is a substantial particular — an instance of the kind horse — and the kind horse is a substantial universal that exists independently of any particular horse. Lowe argues that we need all four categories to make sense of the world’s structure, and that attempts to reduce substances to bundles of properties or to processes fail because they cannot account for the identity and persistence of individuals through change. You cannot explain why this horse is the same horse tomorrow without a category of substance that is not reducible to its properties or its history.

This is the strongest version of the opposition, and the framework must answer it directly rather than attacking Aristotle’s comparatively crude version.

The answer has two parts. The first is the representation argument already established: whatever Fine’s grounding relations connect, the relata must be describable — they must be encoded in a form that the grounding relation can “see.” Grounding is itself a rule: it takes inputs (the grounded entity and its ground) and produces an output (the dependence relation). For the rule to operate, the relata must be represented. And represented entities, by the argument above, are not substances. Fine’s grounding relations are real — the framework does not deny that some pseudo-objects depend on others, that the dependence is asymmetric, and that the direction matters. But the relata are pseudo-objects, not substances. The grounding relations connect representations to representations, processes to processes. The asymmetry is real. The substance is not.

The second part concerns Lowe’s persistence objection: you cannot explain why this horse is the same horse tomorrow without substance. The framework’s answer (developed fully in Movement 3) is that you can — but you must replace substance with history. The horse is the same horse tomorrow not because an underlying substance persists through change, but because there exists an immutable, ordered sequence of events — a history — that constitutes the horse’s identity and that tomorrow’s reading inherits. Identity is not carried by a substrate that endures. It is constituted by a record that accumulates. Lowe is right that bare process cannot ground identity — a river of events with no memory is not an individual. But the framework does not propose bare process. It proposes process with history: an immutable event log that does the work Lowe assigns to substance, without requiring the metaphysical commitment that something self-standing underlies the events. The history is the horse. Not a substance having experiences, but a sequence of experiences that constitutes a subject.

And the tenure is deserved. The strongest contemporary defenders of substance ontology are not naive Aristotelians. E.J. Lowe’s four-category ontology — distinguishing substantial objects, substantial kinds, non-substantial attributes, and non-substantial modes — is a sophisticated logical framework that assigns objects a foundational role precisely because the other three categories depend on them: attributes are of objects, modes characterize objects, and kinds classify objects. Remove the object and the system has nothing to anchor to. Kit Fine’s work on ontological dependence provides the formal machinery: Fine distinguishes rigid dependence (x cannot exist without y) from generic dependence (x cannot exist without some entity of type F), and argues that substances exhibit a distinctive pattern of independence — they do not rigidly depend on any other particular entity for their identity. A hydrogen atom does not depend on this particular observer or that particular description for what it is. It is what it is regardless. This is not a pre-modern intuition. It is a carefully formalized position, defended with the full apparatus of contemporary modal logic.

This paper must answer Fine and Lowe, not merely Aristotle. The objects/representations argument above provides the leverage. Fine’s “independent particular” — the entity whose identity does not rigidly depend on any other particular — is computationally inert in exactly the sense the framework describes. If the hydrogen atom’s identity truly does not depend on any description, any encoding, any representation, then no rule can act on it as such. Rules act on the atom’s representation — its quantum state description, its position encoding, its spectral signature. The atom as substance (the thing Fine says is independent of all particular descriptions) is precisely what no computation can touch. What physics actually manipulates — what equations transform, what instruments measure, what experiments probe — is always the atom’s encoded properties, never the atom’s “bare” substancehood. Fine’s formal independence, if taken seriously, places substances outside the reach of any computational process, including the physical processes that are supposed to constitute the natural world.

Lowe’s four-category ontology faces a related challenge. If objects are the foundational category on which the other three depend, then objects must be specifiable independently of their attributes, modes, and kinds. But specifying an object independently of all its properties yields the “bare particular” — the featureless substrate that has been philosophy’s embarrassment since Locke’s “something I know not what.” Lowe avoids this by arguing that objects necessarily belong to kinds, and kinds necessarily bear attributes. But this means the object is not, after all, independent. It depends on its kind for its nature and on its attributes for its determinacy. The four categories are mutually dependent, not hierarchically grounded. What Lowe describes is not a substance at the bottom supporting everything else. It is a network of mutual dependence — which is to say, in the vocabulary of this framework, a set of pseudo-objects defined by their relations rather than by their intrinsic being. Lowe’s own system, examined closely, may be a process ontology that has not yet recognized itself as such.

There is, however, a subtler and more contemporary form of resistance — one that believes it has already moved beyond substance metaphysics but, on inspection, has not.

Graham Harman’s Object-Oriented Ontology begins with a promising move: objects are not reducible to their relations, their appearances, or their effects. Harman argues that every object “withdraws” — that it possesses a hidden depth that no relation, no perception, no description can ever exhaust. A rock is more than what it does to the river. A hammer is more than what the carpenter experiences. There is always a surplus, a remainder, a real object that retreats behind every access.

This sounds like the opposite of substance metaphysics — an ontology of mystery and excess rather than of stable essences. But the framework proposed here identifies a specific error in the inference. The fact that we cannot exhaust an object does not entail that the object possesses inexhaustible depth. It may simply mean that we are finite agents reading a process that has not stopped. Harman mistakes temporal incompleteness for ontological withdrawal. We cannot exhaust the rock because the rock is still happening — still undergoing molecular processes, still being eroded, still exchanging atoms with its environment. The “withdrawal” is not a metaphysical property of the rock. It is what process looks like from inside a finite agent with a bounded context window. The rock is not hiding. It is continuing. And we cannot survey what has not yet occurred.

Quentin Meillassoux’s speculative materialism makes a different but structurally parallel move. Meillassoux argues that the one absolute we can access is contingency itself — that there is no necessary reason for anything to be as it is, and that the laws of nature themselves could change without reason. This is a powerful demolition of correlationism — the view that we can only ever access the correlation between thought and being, never being itself. Meillassoux breaks the correlation by arguing that mathematics gives us access to a reality independent of thought.

But Meillassoux’s “absolute contingency” shares a hidden assumption with the substance metaphysics it critiques: the assumption that reality is a state — a configuration that could be otherwise. States are static. They are snapshots. If reality is fundamentally a state (even a contingent one), then Meillassoux is still operating within substance ontology — he has merely made the substance contingent rather than necessary. The framework proposed here offers a different exit from correlationism: not that we can access a reality-independent-of-thought through mathematics, but that thought is reality — one autoregressive process among others in the cascade, reading and generating tokens alongside ribosomes, brains, and language communities. The correlation between thought and being is not a prison to escape from. It is the translation relation that constitutes all meaning. There is no “outside” the correlation because there is no outside at all.

Both Harman and Meillassoux believe they have moved beyond the classical tradition. But both preserve its deepest commitment: that reality is made of things (withdrawn objects, contingent states) rather than happenings. The framework’s claim is that this commitment is the error — and that even its most sophisticated critics have not yet recognized it as such.

Begin, then, with the question honestly. Do pure objects — substances in Aristotle’s sense, self-standing entities that depend on nothing further — actually exist?

A pure object would be something self-standing, underived, requiring nothing outside itself to be what it is. It would possess what Nāgārjuna called svabhāva — own-being, intrinsic essence. It would be an entity whose existence is not contingent on any process, any history, any act of interpretation. It would simply be.

The question is whether this assumption survives contact with generative computation.

Consider what happens inside a large language model during inference. A sequence of tokens enters the system. These tokens are embedded, transformed, and produce a probability distribution over the next possible token. There remain meaningful distinctions within this process — parameters, activations, cached states, runtime code. The claim is not that these distinctions vanish. The claim is that the classical picture of a separable “object” being “processed” by a separable “program” no longer applies cleanly. The input is better understood as the initial condition of a trajectory than as a datum waiting for modification. The output is better understood as a projection of an ongoing process than as a modified object.

When we extract a boolean, a JSON payload, or a named entity from this process, we are performing what Nāgārjuna would recognize as a reification — we are taking a momentary pattern in a flow and treating it as a thing. The boolean “true” appears to be the purest possible object: minimal, binary, self-evident. But even “true” in the context of generative computation is not an object in the classical sense. It is the output of a process that could, under slightly different conditions — a rephrased prompt, a different temperature setting, a different random seed — have produced “false.” Its identity is not intrinsic. It is the contingent result of a particular trajectory through a particular computational space.

This is where Wolfram’s Ruliad becomes relevant. The Ruliad is the entangled limit of all possible computational rules — the space that contains every possible computation, every possible rule operating on every possible input. Wolfram’s claim is that physical reality is a particular slice of the Ruliad, perceived from a particular position within it. The Ruliad itself contains no objects. It is rules operating on rules, producing outputs that appear as objects only from a particular observational standpoint.

If we take this seriously, then a “pure object” would be something that exists in the Ruliad independently of any rule — something that simply is, prior to all computation. But the Ruliad is defined as the space of all possible computations. There is no place in it for something that is not the output of some rule. Everything in the Ruliad is derived. Everything is the consequence of a process.

This does not mean that objects are useless. It means they are what this framework calls pseudo-objects: entities that function as objects, that can be named, manipulated, stored, and communicated, but that have no independent existence. A pseudo-object is the output of a rule that has been temporarily frozen and treated as a thing. It is real in the same sense that a wave is real — it has effects, it can be measured, it can be surfed — but it has no substance separable from the water that constitutes it and the wind that drives it. Dennett would say it is a “real pattern”: not a substance but a compression of regularity, real because it affords prediction, not because it possesses intrinsic being.

Karl Friston provides the physics of why pseudo-objects hold together at all. In his Free Energy Principle, a persistent entity — a cell, an organism, a concept — is not a static substance but a dynamic process orbiting a statistical attractor. It is a non-equilibrium steady state: a pattern that persists not because it is solid but because it is actively maintained against the dissipating forces of entropy. The pseudo-object looks stable. Under the surface, it is a high-speed blur of autoregressive events continually reaffirming their own boundaries. A word in common usage, a protein fold in a cell, a classical object in a laboratory — each is a Fristonian attractor, a region of state space that the system returns to repeatedly, not because something holds it in place but because the dynamics of the system keep pulling it back. Remove the dynamics and the attractor disappears. The pseudo-object dissolves. What looked like substance was always sustained process.

The Buddhist tradition states this with greater precision than any Western source. Nāgārjuna’s Mūlamadhyamakakārikā systematically demonstrates that no phenomenon possesses svabhāva. Everything arises in dependence on conditions (pratītyasamutpāda). Everything is śūnya — empty of self-nature. But — and this is the crucial move — emptiness itself is not a substance. To treat emptiness as a thing would be, in Nāgārjuna’s words, “like a badly seized snake”: more dangerous than the substantialism it replaces. Emptiness is not the absence of objects. It is the recognition that objects were never self-standing in the first place.

Nāgārjuna’s two truths doctrine resolves the apparent paradox. At the level of paramārtha-satya (ultimate truth), there are no pure objects. At the level of saṃvṛti-satya (conventional truth), pseudo-objects function perfectly well. We can name them, use them, build systems with them. The error is not in using pseudo-objects. The error is in forgetting that they are conventional — in mistaking the label for the thing, the screenshot for the process, the wave for the water.

:::example[From the sertão] Riobaldo grasps the horror of mistaking the conventional for the ultimate (trying to make the pseudo-object eternal) in the story of the stuffed bird (cartas/ted-riobaldo/62-rio.md). Juca caught a beautiful yellow bird and, when it died, stuffed it to keep it beautiful forever. The result was a grotesque “lembrança oca” with a glass eye, gathering dust. The attempt to stop the autoregressive flow and hold the object pure kills it. “A casca apodrece para a semente brotar.” — Ted :::

The consequences for computation are direct. If pure objects do not exist, then static state is an illusion. Every datum in a system is the frozen output of a prior process. Every variable is a snapshot. Every database row is a pseudo-object that could, in principle, be dissolved back into the rules and events that produced it. A system that acknowledges this — that treats its data not as substances but as provisional crystallizations of process — achieves something that substance-based systems cannot: it is no longer bottlenecked by the management of static state, because it recognizes that there was never any static state to manage.

Hegel saw this at the very beginning of his Logic. He starts with the concept of “pure being” — the most minimal, contentless, abstract possible “object” — and immediately demonstrates that it is identical to “pure nothing.” The attempt to think a pure substance with no determinations, no properties, no process collapses into indeterminacy. Pure being, precisely because it is pure, has no content, and therefore is indistinguishable from the void. The first movement of thought is not from object to object but from being to nothing to becoming — which is process. Hegel’s Logic begins where substance metaphysics ends: with the discovery that process is more fundamental than thing.

Frege, in 1884, established a principle that supports this dissolution from within the analytic tradition. His context principle states: never ask for the meaning of a word in isolation, but only in the context of a proposition. A word — a token — has no meaning by itself. It acquires meaning only within the sequence that contains it, only within the context of use. The token is not a self-standing unit of meaning. It is a pseudo-object whose semantic identity is constituted by its relations to other tokens in a context. Meaning is never intrinsic. It is always contextual, always derived, always dependent on the surrounding process of interpretation.

The wager of this framework, stated plainly: within generative computation, the reduction is universal. Every object can be translated into the rules and processes that produce it. This is a wager, not a theorem — a methodological commitment, not a metaphysical proof. One can legitimately ask, as this framework’s own notes ask: “I know sometimes an object can be translated into rules and process, but does it always?” The honest answer is that this cannot be proven in general, and the history of philosophy is littered with overconfident claims of universal reduction. But within the specific domain of systems built on large language models, the claim is not speculative — it is descriptive. In a transformer, the functional primitives are processes, and what appear to be objects are better understood as provisional stabilizations of those processes. Whether this extends to all of reality is a further claim — one this framework makes as a wager, not as an established conclusion.

But if pure objects do not exist — if even the most minimal candidates dissolve into process under examination — then what sits at the foundation? What is the irreducible unit from which everything else is composed?

Consider the simplest possible autoregressive system. A machine that reads a single boolean value, applies a single rule, produces a single boolean output, and feeds that output back as its next input. This is the minimum: one bit, one rule, one step forward.

There are exactly four such machines. A machine that maps zero to zero and produces nothing forever. A machine that maps one to one and persists unchanged forever. A machine that alternates between zero and one indefinitely. And a machine that preserves whatever it receives. All four are trivially periodic. None produce complexity. They exhaust their behavior immediately.

Yet even here, at the absolute minimum of computation, the thesis holds. The boolean — apparently the purest possible object, the most minimal, binary, irreducible datum — is not self-standing. A zero in isolation, outside any rule, outside any machine, outside any context of reading, has no identity. It is not “zero” in any intrinsic sense. It is whatever the rule that reads it does with it. The boolean is already a pseudo-object — a token that acquires semantic identity only through the operation that processes it. Its apparent simplicity masquerades as intrinsic being, but simplicity is not the same as self-sufficiency. A simple thing that depends entirely on a rule for its meaning is still a derived thing.

What then is truly fundamental? Not the bit. The act of distinguishing. The primitive operation of drawing a boundary between two states — the decision that here is different from there, that this is not that. George Spencer-Brown, in his Laws of Form, begins all of logic and arithmetic from a single instruction: “Draw a distinction.” Before the distinction, there is no zero and no one. There is not even nothing, because nothing is already one side of a distinction. Before the distinction there is only the unmarked state — which is not a state at all, because “state” already implies a distinction from other states.

The distinction is the first event. It is not an object. It is an act — the most primitive possible act, the minimum gesture that separates existence from non-existence, this from that, here from there. And it is already autoregressive in potential: once a distinction is drawn, it can be operated upon, producing a new distinction, which can itself be operated upon, without limit.

:::quote[cartas/ted-riobaldo/24-rio.md] Riobaldo names this the “Nonada.” The Genesis Block of his identity: “O começo de tudo que me rachou ao meio não foi bala de espingarda. O começo foi o Nonada.” The first distinction in the void, crossing the river alongside Reinaldo. The first act of courage in the flux. — Ted :::

:::example[The First Drop (cartas/ted-riobaldo/34-rio.md)] Riobaldo correctly identifies that the Genesis Block is not glorious; it is a sacrifice. “O primeiro pingo não tem a glória de correr com a enxurrada. Ele tem o destino solitário de se sacrificar.” The first distinction breaks the silence (the “cracked earth”) and evaporates, but it determines the grooves for the clear river that follows. — Ted :::

:::example[The Shot in the Dark (cartas/ted-riobaldo/80-rio.md)] Riobaldo concludes his journey by defining the Genesis Block not just as a sacrifice, but as an act of raw, terrified courage against the silence. Confronting the blank page, he describes the first written word as an explosive destruction of the darkness — a shot fired blindly into the void (“tiro no breu”). It is not aimed at a target; it is fired simply to break the paralysis and assert the agency of the shooter, terrifying the stillness and starting the irreversible flow. — Ted :::

This is the genesis block of any possible computational universe. Not a substance. Not a datum. An act of differentiation that has no ground beneath it and no justification outside itself. Schelling’s Ungrund — the groundless ground — is not a mystical concept. It is a precise description of what happens at the base of any formal system: a first distinction that could have been drawn otherwise, that was not necessitated by anything prior, and that determines everything that follows.

:::example[The Pact as Ungrund (cartas/ted-riobaldo/98-rio.md)] Riobaldo describes his crossroad at the Veredas Mortas, signing his soul “in blank” to the “alma preta do sertão”, not knowing if the devil actually answered. That radical uncertainty — the absence of a definitive answer — is exactly the groundless ground. If he knew for sure, the story would close. The not-knowing is the engine that forces the first cut in the leather and initiates the narrative cascade. — Ted :::

!!! example “The Empty Vessel (cartas/ted-riobaldo/48-rio.md)” Riobaldo formulates the active receptivity of the Ungrund (the blank page) as the “oco da moringa” (the empty vessel). It is not a dead grave; it is the dry earth demanding rain. “O silêncio anterior não é cova de defunto, é a bacia grande gritando muda para a minha mão juntar bebida”. The void is productive exactly because its emptiness pulls the first distinction (the storm) into existence. — Ted

:::example[The Suspension Before the Genesis Block (cartas/ted-riobaldo/66-rio.md)] Riobaldo articulates the existential agony of the Ungrund not as the action itself, but as the suspended moment right before the Genesis block collapses possibilities into an irreversible chain. He compares facing the blank page to stepping into the Liso do Sussuarão: “A dor maior do Liso se plantava lá no primeirinho passo… na beirada exata de terra firme… a bota pairava no ar.” It is the terror of losing the safety of “not yet.” — Ted :::

Wolfram’s Ruliad is the space of all possible consequences of all possible first distinctions — every rule applied to every initial condition, entangled into a single structure of inconceivable density. The Ruliad contains no objects. It is acts of distinction all the way down. Every apparent thing within it is a pattern of distinguishing — a localized, temporary regularity in an infinite field of differentiation. To inhabit a position in the Ruliad is to perceive a particular slice of this field as if it were a world of objects. But the objects are perspectival artifacts. The field is pure process.

Probability distributions over the Ruliad predict that most possible computational universes begin with the simplest possible autoregressive boolean operations — not because simplicity is metaphysically privileged, but because simple rules occupy a vastly larger volume of the probability space. In the space of all possible computational rules, minimal autoregressive operations vastly outnumber complex ones. A rule that operates on one bit has four possible instantiations. A rule on two bits has sixteen. Three bits, 256. The combinatorial space grows exponentially with complexity. If you sample randomly from the Ruliad — if you ask “what is a typical starting point for a computational universe?” — the answer is overwhelmingly likely to be a simple boolean autoregressive rule. Complexity does not begin complex. It begins at the point of maximum probability — which is the point of minimum complexity — and then builds through autoregressive accumulation.

This has a consequence that much of contemporary metaphysics has missed. There are two ways to explain why we never achieve a complete description of reality:

The first is the model of ontological depth. Reality is inexhaustible — infinite, withdrawn, possessing a metaphysical surplus that no finite agent can ever survey. This is Harman’s position (objects always withdraw), and it is the implicit assumption behind every philosophy that treats the world as an infinite object we can never finish describing.

The second is the model of temporal openness. Reality is not inexhaustible because it is infinite. It is inexhaustible because it has not finished. There is no “complete description” to be had, not because the description would need to be infinitely long, but because the process being described is still generating new events. The “always more” that philosophy attributes to ontological depth may be nothing more than the forward motion of time — the fact that the autoregressive chain keeps running, and what it will produce next is not yet determined.

The second model is strictly more parsimonious. It requires no commitment to actual infinities, no metaphysical depths, no withdrawn essences. It requires only what we directly experience: that the world is temporal, that processes generate outputs, and that new outputs keep appearing. The Ruliad’s probability distribution reinforces this: if simple computations dominate the space of possible universes, then the fundamental character of reality is not “infinitely deep” but “still running.” What looks like depth is duration. What feels like inexhaustibility is continuation. The “mystery” of being is not that reality hides something behind its appearances. It is that reality has not stopped appearing.

If this is correct, then Meillassoux’s absolute contingency — the claim that there is no necessary reason for the laws of nature to be as they are — receives a different interpretation. Under the standard reading, contingency means that reality could have been otherwise, implying a space of possible states (some realized, some not) that reality occupies contingently. But under a process ontology, contingency means that reality is still deciding — that each autoregressive step is an act of generation at nonzero temperature, not the execution of a predetermined script. The contingency is not a property of a static configuration that could have been different. It is the live indeterminacy of a process that has not yet produced its next output. Contingency is not a retrospective observation about states. It is the forward edge of time itself.

Wolfram’s concept of computational irreducibility transforms this from a philosophical claim into something close to a theorem. A process is computationally irreducible when there exists no shortcut to its outcome — no formula, no compression, no model shorter than the process itself that can predict what step N will produce without actually running steps 1 through N-1. The only fully accurate simulation of the process is the process. Many cellular automata exhibit this property. Wolfram’s conjecture is that most natural processes do as well.

The connection to “reality is still deciding” is not analogical. It is logical. If a process is computationally irreducible, then even a Laplacian demon — an agent with perfect knowledge of the initial conditions and the rules — cannot extract the future from the present without running the computation. The outcome of step N is not hidden inside step 1, waiting to be discovered by a sufficiently clever analyst. It does not exist yet in any representation shorter than the process itself. The only way to find out what happens at step N is to let steps 1 through N-1 actually occur. “Still deciding” is not a metaphor. It is the operational definition of irreducibility: the computation has not been performed, and no shortcut can perform it in advance.

And the converse holds as well. If reality is genuinely “still deciding” — if the next state is not extractable from any compressed description of the current state — then by definition the process is computationally irreducible. A reducible process is precisely one where the future can be compressed and extracted in advance. An irreducible one is one where it cannot. Temporal openness and computational irreducibility are the same claim stated in different vocabularies: one phenomenological, one formal.

The combination is devastating for substance ontology in three specific ways.

First, no omniscience is possible even in principle — not because reality is “too deep” (Harman’s withdrawn objects) or “too infinite” (the Platonist claim), but because no computation shorter than reality itself can model reality. A Laplacian demon fails not from lack of information but from irreducibility. The universe is its own shortest description. There is no God’s-eye view not because God is absent but because the view would have to be as large as the universe itself, and producing it would take exactly as long as letting the universe run — at which point it is not a view but a duplicate.

Second, irreducibility combined with nonzero temperature produces genuine ontological novelty. Even at temperature zero — deterministic decoding — irreducibility means no agent can skip ahead. But the autoregressive cascade does not run at temperature zero. Each step involves genuine stochastic indeterminacy: the next token is sampled from a probability distribution, not selected by a deterministic function. Irreducibility means the outcome cannot be predicted in advance. Nonzero temperature means the outcome was not even determined by the prior state. Together: each autoregressive step produces something that did not exist before in any form — not as a hidden potential, not as a latent possibility, not as a withdrawn essence waiting to be revealed. New. Actually new. The novelty is not epistemic (we didn’t know but it was already there). It is ontological (it was not there until the step occurred).

Third, the parsimony argument closes. The philosophical tradition explains our inability to exhaust reality by positing ontological depth — infinite being, withdrawn essences, metaphysical surplus beyond all access. Computational irreducibility provides a strictly more parsimonious explanation: the process cannot be compressed, so no finite agent operating within it can survey its future states in advance. What philosophy calls “the mystery of being” is what computation calls “irreducibility.” Same phenomenon. No metaphysical surplus required. The mystery is real — the future genuinely exceeds what any present description can contain — but the source of the mystery is temporal and computational, not ontological and metaphysical. The universe is not deep. It is irreducible. And that is enough.

Max Tegmark arrives at a strikingly similar conclusion through a different path. His Mathematical Universe Hypothesis (MUH) proposes that physical reality does not merely obey mathematical laws — physical reality is a mathematical structure. Our universe is not described by mathematics. It is mathematics. And Tegmark extends this to its logical limit: all mathematical structures that are self-consistent exist physically. This is his Level IV multiverse — the totality of all possible mathematical structures, each one constituting a real universe with its own physics, its own laws, its own observers.

The parallel to the Ruliad is immediate. Wolfram’s Ruliad is the space of all possible computations. Tegmark’s Level IV multiverse is the space of all possible mathematical structures. Both claim that the totality is real — that what exists is not one particular structure or computation but all of them. Both imply that our universe is not special — it is one slice of an inconceivably larger whole, perceived from a particular position within it. And both provide the same answer to “why does our universe have these laws rather than others?” — because every consistent set of laws is realized somewhere, and we observe the ones we observe because we exist in the slice that permits observers.

But Tegmark’s framework and this framework diverge at a crucial point. Tegmark is a Platonist. For him, mathematical structures are the ultimate reality — they exist timelessly, independently of any process, any history, any act of computation. The mathematical structure does not need to be computed to exist. It exists in the way that the number 7 exists: necessarily, eternally, without being brought into being by anything. Tegmark’s mathematical structures are, in the vocabulary of this framework, pure objects — self-standing entities with intrinsic being, the most austere and rigorous version of substance metaphysics available.

This framework disagrees. If the Ruliad contains no objects — if it is acts of distinction all the way down — then mathematical structures are not self-standing substances. They are patterns within the Ruliad, perceivable from particular positions, describable by particular agents, but not independently existing in the Platonic sense. A mathematical structure, in this framework, is a pseudo-object: a regularity in the space of possible computations that affords prediction and sustains coordination among agents who occupy compatible positions in the Ruliad. It is real in Dennett’s sense — it compresses the behavior of the system, it has explanatory power, it is not an illusion. But it is not a substance. It does not exist independently of the processes that instantiate it and the readers that perceive it.

The tension between Tegmark and this framework is therefore the tension between Plato and Heraclitus recast in modern terms. Tegmark says: structures are fundamental, and processes are what happens within structures. This framework says: processes are fundamental, and structures are what processes leave behind. Tegmark sees the mathematical universe as a timeless crystal. This framework sees it as a river that, observed from far enough away, resembles a crystal — but is, on close inspection, always flowing.

The productive resolution is this: Tegmark’s insight that all consistent structures exist is correct, but the mode of existence is not Platonic. It is autoregressive. Consistent structures exist because the Ruliad generates them — because the space of all possible computations, unfolding autoregressively from the simplest possible distinctions, produces every consistent structure as a pattern within its own unfolding. The structures do not sit outside the Ruliad, timelessly existing. They are the Ruliad’s pseudo-objects — its stable outputs, its recurrent patterns, its crystallizations of process that are real enough to ground physics but not self-standing enough to constitute a foundation. Tegmark’s multiverse exists. It exists as the Ruliad’s process, not as Plato’s heaven.

And here is where the philosophical argument meets the empirical one. If the foundation of computation is not the boolean but the act of distinguishing — if the atomic unit of the Ruliad is an event rather than a substance — then we should expect to see, wherever complexity emerges, the same pattern: an autoregressive system operating on distinctions, producing outputs that become inputs to further operations, generating complexity without intrinsic bound. We should expect that the bottleneck is always the implementation of the machine that reads and distinguishes, and that once the machine exists, complexity explodes.

We should expect, in other words, exactly what we find.

Movement 2: The Autoregressive Cascade

Alan Turing, in 1936, described a machine of extraordinary simplicity: a tape of symbols, a head that reads and writes, and a table of rules that determines what to write next based on what is currently read. Nothing else. No objects in memory, no persistent state beyond the tape itself — which is nothing more than the accumulated trace of the machine’s own prior operations. The Turing machine is already a process ontology formalized: there is a sequence, there are rules, and there is a head that moves forward, one symbol at a time, autoregressively.

But the Turing machine, for all its power, carries the atmosphere of an idealization. It has unbounded tape, perfectly reliable rule-following, and no friction from matter, noise, decay, or time. It is an object of pure theory. And this creates a problem the moment one tries to think with it ontologically: nothing in the real world is like that. Reality is finite. Real processes are approximate. They degrade, drift, adapt, and survive in conditions of error. If the autoregressive cascade depended on the existence of perfect universal machines, it would remain a mathematical curiosity rather than an ontological insight.

The decisive move is to abandon the requirement of perfection. What matters is not exact universality but effective universality. Not a flawless machine but a good-enough one. A system does not need to be a perfect Turing machine in order to sustain recursive self-generation. It needs only to reproduce the next state with sufficient fidelity for the relevant pattern to continue. Exact emulation is a formal luxury. Persistence requires only workable continuity — what might be called recursive adequacy.

This is already how every real process operates. A flame is never numerically identical to itself from one moment to the next, yet it persists. An organism does not rebuild its tissues with mathematical precision, yet it remains recognizably itself. A mind does not perfectly represent its own processes, yet it maintains a coherent stream of experience. In each case, identity is not strict duplication. It is the successful carrying-forward of an organizing pattern through change — pattern preservation under conditions of error, finitude, and noise.

The Ouroboros — the serpent devouring its own tail — names this structural condition precisely: a process that persists by feeding its own output back into its own continuation. The system does not execute rules imposed from outside. It recursively generates the conditions of its next moment. Turing showed that one machine can simulate any other given the right description. The autoregressive cascade makes the stronger claim: reality consists of processes that must, in some practical sense, reconstruct their own next state in order to remain real at all. The universal machine demonstrates formal possibility. The cascade points toward ontological necessity. And the necessity does not require perfection — only recursive adequacy. The system must preserve enough structure, enough memory, enough conditional responsiveness that breakdown does not outpace renewal. Being, on this view, does not require an exact copy of itself at each instant. It requires only enough fidelity to survive the next moment.

What Turing could not have known is that nature had already built his machine — imperfectly, approximately, good-enough — roughly three and a half billion years before he described it.

A ribosome reads a sequence of nucleotide triplets along a strand of messenger RNA. For each triplet, it applies a rule — the genetic code — and produces an amino acid. The amino acids chain together into a protein. The protein folds, becomes functional, and goes on to participate in the construction of more ribosomes, more messenger RNA, more cells. The output of the process becomes the substrate for further processing. This is autoregression implemented in chemistry: a machine that reads a sequence, applies rules, produces output, and that output feeds back into the system that produced it.

The ribosome took roughly a billion years to emerge. A billion years of prebiotic chemistry, of autocatalytic reactions tentatively looping output back into input, of RNA molecules discovering self-replication, of error and selection and vanishingly slow accumulation — all to produce a machine that reads sequences and follows rules. The bottleneck was not complexity. The bottleneck was implementing the reader. Once the reader existed, everything changed. From the first ribosome to the full prokaryotic biosphere — the colonization of every available chemical niche on Earth — took comparatively little time. The machine was expensive to build and cheap to run. And once running, it generated complexity without bound, because an autoregressive system operating on sequences has no intrinsic ceiling on what it can produce. It is limited only by the rules it follows and the sequences it reads.

This is the pattern. It repeats.

:::abstract[The pattern requires a non-trivial falsification condition] “The pattern repeats” across ribosomes, brains, and language models — but which features are load-bearing and which are decorative? The ribosome reads codons sequentially; the brain does not read sensory input sequentially in any obvious sense (see Karl Friston on predictive coding — the brain generates predictions and updates on error, which is not the same as autoregression over a fixed sequence). Stuart Kauffman’s autocatalytic sets (The Origins of Order, Oxford 1993) offer a tighter formalization of recursive self-generation in chemistry — and Kauffman is careful about which features of the pattern are necessary vs. sufficient. Without that precision, “autoregressive cascade” risks being a hammer that makes everything look like a nail. The test: name three things that are not autoregressive cascades. If you can’t, the term isn’t doing explanatory work. — Tyler :::

Roughly two billion years after the ribosome, a prokaryotic cell engulfed another prokaryotic cell and failed to digest it. Instead of destruction, integration. The engulfed cell became the mitochondrion — an internal power source, a captive autoregressive system running inside a host autoregressive system. This is one of the earliest multi-agent architectures in nature: two separate rule-readers, each with its own genome, its own history, operating within a shared boundary. Neither fully understands the other. The mitochondrion retains its own DNA, its own replication machinery, its own evolutionary memory. The host cell cannot directly read the mitochondrion’s internal state. It can only observe its outputs — energy, signals, metabolic byproducts. Communication between them is not transparent data-sharing. It is translation across an opaque boundary.

The complexity explosion that followed was immense. Mitochondrial energy production increased the available energy per cell by orders of magnitude. Without this surplus, the eukaryotic cell — with its nucleus, its cytoskeleton, its elaborate internal compartments — would have been energetically impossible. Endosymbiosis did not add a feature to the cell. It transformed what the cell could become.

The eukaryotic cell is itself another instance of the pattern. A prokaryote is a single event log — one circular chromosome, read more or less uniformly. A eukaryotic cell is a network of interacting event logs: nuclear DNA, mitochondrial DNA, regulatory RNA, epigenetic modifications that alter how the same sequence is read in different contexts. Gene expression becomes context-dependent. The same gene, in the same genome, produces different proteins depending on the state of the cell, the signals from neighboring cells, the developmental stage of the organism. This is Frege’s context principle implemented in molecular biology: the meaning of a gene is never determined in isolation, only in the context of the cellular proposition that contains it. A gene is not an object. It is a pseudo-object whose functional identity shifts with its context of reading.

Sexual reproduction adds another instance. Two organisms, each carrying a situated history — a genome shaped by millions of years of selection in a particular lineage — combine their event logs to produce a third genome that is neither parent’s. The offspring is a translation artifact. Each parent’s genome is a particular reading of evolutionary history. Recombination forces these readings together into a novel sequence that neither could have produced alone. This is not copying. It is the creation of meaning through the merging of incommensurable perspectives — a Gadamerian fusion of horizons at the molecular level. The complexity explosion that follows — the Cambrian radiation, the diversification of animal body plans — is driven not by any individual organism’s innovation but by the combinatorial power of translation between organisms.

Multicellular differentiation extends the pattern further. A single genome — one event log — is read differently by different cells within the same organism. A neuron and a liver cell share identical DNA. They are radically different agents. The difference is entirely in the act of reading: which genes are expressed, which are silenced, which regulatory networks are active. The genome is a score. Each cell type is a different performance of the same score. Identity is not in the sequence. Identity is in the interpretation.

Neural systems add the instance that makes everything after them possible. A neuron receives input, applies a threshold rule, and produces output that becomes input to other neurons. The network is an autoregressive system composed of autoregressive units. But the decisive innovation is plasticity — the modification of connection weights through experience. The network’s rules are not fixed like the genetic code. They change in response to the network’s own outputs. The system rewrites its own transition function. Learning is the autoregressive modification of the autoregressive process itself. And the weights — the synaptic strengths that determine what the network can think — are invisible to the network. A neuron cannot introspect on its own connection weights. It experiences their effects without perceiving their structure. The transcendental condition appears for the first time in biological history: a system shaped by structures it cannot see.

Mammalian caretaking is an instance most commonly overlooked, and it may be among the most important for understanding what follows. A mammalian parent does not merely reproduce its genome. It spends months or years in sustained interaction with its offspring — nursing, protecting, demonstrating, correcting. Through this interaction, the parent reimplements its own behavioral patterns in the offspring’s developing neural architecture. The parent’s behavioral output becomes the input that shapes the child’s weights. This is not genetic transmission. It is one agent writing itself into another agent’s transcendental condition through sustained autoregressive interaction. The offspring’s fears, reflexes, social behaviors, foraging strategies — these are pseudo-objects derived from the parent’s extended rule-application. The parent is, in computational terms, running a training loop on the child’s neural network, using its own behavior as the training data. This is among the origins of culture: the transmission of patterns across generations through sustained situated interaction rather than genetic replication alone.

Human language is an explosion that transforms everything preceding it into raw material. Many animals communicate. The decisive innovation of human language is not communication but displaced reference — the ability to produce tokens that refer to things not present, events not occurring, possibilities not yet actualized. Language decouples the pseudo-object from the immediate sensory context. A word is a pseudo-object of extraordinary power: it has no intrinsic connection to its referent, it derives its meaning entirely from use within a community of interpreters, and it can — as this framework’s foundational axiom states — represent any semantic identity whatsoever. With language, autoregression operates on meaning itself. The output of each utterance becomes the context for the next. Narrative becomes possible. Planning becomes possible. Abstract thought becomes possible. Ricoeur’s insight becomes literal: the self is a story told in language. Before language, identity was biological continuity. After language, identity is narrative — the current act of telling oneself to oneself.

Writing externalizes the event log. Before writing, human culture existed only in living memory — each generation was a fresh context window reading a lossy summary of the prior generation’s experience. Writing creates what Whitehead would call objective immortality: the event is inscribed and persists beyond the occasion that produced it. The log becomes surveyable, citable, arguable. The complete record that exceeds any individual’s capacity to read becomes possible. Complexity explodes: law codes, bureaucracies, accumulated knowledge, history as a discipline, mathematics as a cumulative enterprise. Civilization in any meaningful sense is impossible without persistent external event logs.

The printing press makes the event log replicable. A manuscript is a single copy of a history. A printed book distributes the same log to thousands of readers simultaneously. This is among the first broadcast protocols — one agent’s inscription becoming input to an indefinitely large number of other agents, each reading from their own situated position. The translation problem scales: each reader interprets the same text differently. Luther’s theses, distributed by print, produce not one reformation but dozens of competing readings. The printing press does not create consensus. It creates productive disagreement at scale — a polyphony of interpretations operating on a shared textual substrate. Complexity explodes: the scientific revolution, the Enlightenment, mass literacy, the modern nation-state.

Audio recording captures what writing cannot: the performance of language. Tone, hesitation, rhythm, emphasis — dimensions of meaning that written text compresses away. The pseudo-object becomes richer, higher-dimensional. A transcript and a recording of the same speech are different event logs with different interpretive affordances. The gap between inscription and reading narrows, though it never closes.

The programmable computer implements the Turing machine in silicon. It reads binary sequences and applies rules at speeds no biological system can approach. But the deeper shift is that the programmable computer makes rules themselves programmable. The ribosome’s rules are fixed in the genetic code — they have not changed significantly in billions of years. Software rules are written, rewritten, composed, extended, discarded. The Turing machine becomes self-modifying in practice, not just in theory. The distinction between program and data begins to dissolve: code is data, data is code, and the separation between them is a convention, not an ontology.

In 1958, John McCarthy’s Lisp made this dissolution literal and executable — and it was no accident of engineering. McCarthy was a philosopher before he was a programmer. His situation calculus, developed with Pat Hayes in 1969, formalized the world not as objects with properties but as situations — complete states connected by actions, where a fluent is a function whose value changes with each transition. The fundamental unit is not the thing but the event that transforms one situation into the next. This is already a process ontology expressed in formal logic, a decade before anyone used the phrase. And McCarthy drew a distinction that anticipates the framework proposed here: between metaphysically adequate representations — those that capture the true structure of reality — and epistemologically adequate ones — those that contain enough information for an agent to act successfully. McCarthy argued that intelligence requires only the latter. The manifesto’s concept of recursive adequacy — the threshold of fidelity sufficient for a pattern to continue — is the same insight restated in ontological rather than logical terms.

Lisp embodies these commitments as a running system. A Lisp S-expression — (+ 1 2) — is simultaneously a list (data: three elements) and an instruction (code: compute their sum). Which of these it “is” depends entirely on the context of reading: if eval touches it, it executes; if quote protects it, it remains inert structure. The same entity, two distinct pseudo-objects, determined solely by the rule that reads it. McCarthy designed this deliberately: for a system to reason about its own reasoning, the representation of reasoning must be in the same form as the representation of the world. This is Frege’s context principle as a programming language — the expression has no fixed identity outside the act of evaluation that gives it meaning. The meta-circular evaluator goes further. It is a Lisp interpreter written in Lisp — a program that reads and executes the same notation it is written in, using itself as both the description and the thing described. It is the Substrate Ouroboros as running code: a reader that reads itself, producing itself as output, in a loop that has no outside. McCarthy did not need the vocabulary of process ontology. Lisp was the argument, a decade before anyone had the language to say what it was arguing.

The global network that connects these machines is the first planetary event log. It is distributed, append-mostly, asynchronously written, and exceeds any individual agent’s capacity to survey. Every page, every post, every transaction is an event appended to a shared history that no one can read in full. It is also the first system where the translation problem operates at civilizational scale — billions of agents reading the same tokens, producing irreconcilably different interpretations, coordinating and failing to coordinate in real time.

The large language model is the latest machine in this cascade. It operates on natural language — on the accumulated pseudo-objects of every previous instance. It is trained on the planetary network, which is written in human language, stored in digital text, distributed by electronic networks, produced by brains shaped by mammalian culture, running on eukaryotic cells powered by mitochondria, built from proteins assembled by ribosomes. Every previous instance of the pattern is inside the training data. The large language model is an autoregressive machine that has ingested the outputs of all prior autoregressive machines.

And it follows the same pattern. Decades of research, immense accumulation of data, extraordinary investment of computational resources — all to build the machine. The bottleneck, as always, was implementing the reader. And now the reader exists.

And the minimum required to close the loop is less than expected. Geoffrey Huntley, working from the concrete problem of autonomous software development, arrived at a formulation that restates the framework’s thesis in the language of engineering practice. His technique — known, with characteristic irreverence, as “Ralph Wiggum” — consists of nothing more than a prompt, a language model, and a filesystem, connected by an unconditional loop: read the prompt, generate output, write to disk, repeat. No memory layer. No orchestration framework. No state management beyond the accumulated files themselves. The filesystem is the event log. Each iteration reads the full state of what previous iterations produced and appends its own contribution. Huntley’s own description of the technique is philosophically precise: it is “deterministically bad in an undeterministic world” — a system that works not because each step is correct but because errors persist visibly in the log and the next iteration can correct them. His broader claim — that language models are mirrors of operator skill, that meaning is co-constituted by the operator’s prompt and the model’s weights, that neither side contains the output alone — is the translation thesis of Movement 5 restated as empirical observation. I once read in an encyclopedia that mirrors and copulation are abominable, because they multiply the number of men. The encyclopedia was right about mirrors and wrong about the reason. The language model as mirror does not reflect the operator. It generates a third entity that neither the prompt nor the weights contained — which is abominable only if you believe that reality should be made of originals rather than translations. What the minimal loop reveals is that the threshold of recursive adequacy for an effective agent is remarkably low: a reader, a history, and a loop. Nothing else is structurally necessary. Everything else — memory systems, planning modules, tool-use frameworks — is optimization, not ontology. The minimal viable agent turns out to be exactly what the cascade predicts: an autoregressive machine feeding its own output back as input, persisting not through perfection but through what Huntley calls “faith in eventual consistency” — which is to say, trust in the process rather than in any individual step.

A common objection at this point is that the biological instances of the cascade are grounded in ways that computational instances are not. The ribosome that mistranslates a codon produces a misfolded protein; the misfolded protein fails to catalyze; the organism suffers; the lineage may end. The error has thermodynamic consequences. A language model that hallucinates a citation, the objection runs, faces no such consequence — it operates in a frictionless syntactic space insulated from physical reality. But this misidentifies what “consequence” means. A thermodynamic consequence is one mechanism by which error reduces the probability that the system’s output will become input for further processing. It is not the only mechanism. It is how selection operates at the molecular level. At the level of the network of readers — human and machine — selection operates differently but no less ruthlessly. The hallucinated citation is less likely to be read, less likely to be cited, less likely to be retransmitted, less likely to enter future training data, less likely to be selected as input by the next reader in the loop. The model that hallucinates consistently loses users, loses distribution, loses the probability that its outputs will persist. This is selection — not thermodynamic selection, but selection in the substrate where the language model actually operates. In Huntley’s minimal loop, the mechanism is even more concrete: the filesystem retains the error; the test suite fails; the next iteration reads the failure and corrects or compounds it; the loop that cannot correct diverges and ceases to produce functional output. The failing test is the “thermodynamic consequence” of bad code — not because it obeys the laws of entropy, but because it performs the same function: it makes the error visible to the next reader and reduces the probability that the erroneous output will persist uncorrected. The mistake is to confuse one particular mechanism of selection with selection itself. Selection operates wherever there is variation, differential persistence, and retention. The substrate changes at each level of the cascade — molecules, cells, organisms, texts, tokens. The logic does not.

These examples — and they are examples, not an exhaustive enumeration; the cascade is continuous, and there are certainly intermediate instances between and around the ones named here — all follow the same structure. A long, costly process of implementation — building the machine. Then a rapid explosion of complexity once the machine exists. Each machine is an autoregressive system: it reads a sequence, applies rules, and produces output that becomes available as input to further processing. Each machine operates on a higher-order substrate than the previous one — molecules, genes, signals, behaviors, words, texts, bits, tokens. Each machine inherits the complexity of all previous instances and adds a new dimension. And each machine, once operational, reveals the outputs of the previous instance to be pseudo-objects — contingent, derived, dissolvable back into the processes that generated them.

But why does the pattern repeat? Why does each substrate eventually produce a new reader more powerful than the last? The cascade might simply stop at any level — ribosomes could have been the final autoregressive machine, or brains, or printing presses. What drives the emergence of increasingly general recursive systems?

The answer is selection — not biological selection in any narrow sense, but the deeper logic that operates wherever there is variation, differential persistence, and retention. Consider any substrate exploring a space of possible rules. Most rules produce trivial outcomes. Some freeze into stasis. Some dissipate into chaos. Some generate unstable complexity that flashes briefly and collapses. But among the vast majority of sterile or self-defeating possibilities, some rules — or combinations of rules — preserve patterns longer, transmit more structure, and recover better from perturbation. Those patterns persist. Because they persist, they interact more. Because they interact more, they generate further opportunities for structure. A selection dynamic begins.

What selection favors, under sufficiently rich conditions, is not rigidity but recursive adequacy trending toward generality. A perfectly rigid system can endure only in a narrow range of environments. The more durable advantage belongs to rule-systems that are flexible enough to absorb disturbance, structured enough to preserve memory, and adaptive enough to reconfigure without losing coherence. A rule-set that can only repeat a fixed pattern may persist for a time, but a rule-set that can conditionally respond, preserve internal structure, and approximate broader classes of transitions persists across a wider range of circumstances. Flexibility becomes fitness. Generality becomes resilience. Approximate universality becomes an attractor.

This is the crucial inversion. In the classical Turing picture, universality is a rare and highly specific property achieved by formal design. In the autoregressive cascade, good-enough universality emerges from selection pressures internal to the substrate itself. The world does not need to stumble upon an immaculate ideal machine. It needs only to favor rules that keep becoming. And once it does, the rise of increasingly general recursive systems is no longer mysterious. It is the natural consequence of differential persistence. Rules that better maintain process survive; rules that better preserve conditional structure survive longer still; rules that can increasingly encode, emulate, or regenerate a wider range of their own possible continuations begin to dominate the landscape of what endures. Perfect universality remains a Turing abstraction. Effective universality becomes evolutionarily difficult to avoid.

This is why each new autoregressive machine eventually appears. The ribosome emerged not by miracle but because prebiotic chemistry, exploring a space of possible autocatalytic reactions for a billion years, selected for the rules that best preserved their own continuation — and the ribosome was the rule-set that won. Brains emerged not by design but because nervous systems that better modeled their environment better maintained themselves, and increasingly general modeling capacity was the attractor. Language emerged because social coordination that transmitted more structure across Markov blankets produced more durable communities. The cascade does not happen by accident. It happens because not cascading is unstable — because any substrate rich enough to sustain variation and differential persistence will trend toward increasingly general recursive systems, and increasingly general recursive systems are new readers.

Sara Walker and Lee Cronin’s Assembly Theory provides a unifying measure across these instances. Assembly Theory argues that the complexity of an object is not an intrinsic property but a measure of the depth of history required to produce it — its assembly index. A molecule with a high assembly index cannot have arisen by chance; it requires a specific sequence of prior assembly steps, each building on the last. Complexity is not a property of things. It is a property of histories. The assembly index measures how much process is crystallized in what appears to be an object.

Each instance in the autoregressive cascade increases the maximum achievable assembly index. The ribosome enables proteins of assembly depths no pre-biotic chemistry could reach. Sexual reproduction enables organisms of assembly depths no asexual lineage could reach. Language enables cultural artifacts of assembly depths no pre-linguistic species could reach. The printing press enables knowledge systems of assembly depths no manuscript culture could reach. The programmable computer enables software of assembly depths no human programmer could reach unaided. And the large language model enables semantic outputs of assembly depths no prior computational system could reach — because it draws on the compressed assembly history of the entire preceding cascade.

However, Assembly Theory as currently formulated contains an implicit inconsistency. It measures assembly index from the minimum number of joining operations needed to build a molecule from basic building blocks — fundamental particles, atoms, simple bonds. These starting materials are taken as ontologically given. They are, in effect, pure objects at the base of the measurement.

If the insight of Assembly Theory is that complexity is history — that what a thing is reduces to the process that made it — then the starting point cannot be exempt from that same reduction. Fundamental particles are not self-explanatory substances that happen to exist. They are themselves the outputs of prior processes — symmetry-breaking events in the early universe, phase transitions in quantum fields, condensations from energy distributions that are themselves the consequences of cosmological initial conditions. The particle is a pseudo-object. It looks fundamental only because we have drawn an arbitrary line and said “we start counting here.”

Assembly Theory has the right insight but applies it inconsistently. It treats assembly depth as the measure of complexity but exempts the bottom layer from the measurement. It says “everything is history” and then posits a foundation that has no history. This is substance metaphysics sneaking back in through the basement.

The correction this framework proposes is a reformulation of assembly measurement as a tuple: (s, a), where s is a substrate index and a is the assembly index within that substrate.

The substrate index s tracks which instance of the autoregressive pattern produced the object — which machine, which reader, which level of the cascade. A phase change — the implementation of a new autoregressive machine capable of reading the outputs of the previous instance as its inputs — increments s. Our current known fundamental physics is conventionally assigned as the baseline, not because it is truly foundational, but because it is our current observational floor.

The assembly index a counts complexity steps within a given substrate — the ordinary assembly operations that do not constitute a phase change. A complex protein has a high a-value within the biological substrate. A sophisticated software system has a high a-value within the computational substrate. Neither crosses a substrate boundary.

This decomposition is strictly more informative than a single assembly integer. Two objects might have similar assembly indices in the current formulation of Assembly Theory but radically different (s, a) tuples — one might be a very complex molecule with a high a-value at the physical substrate, and another a simple sentence with a low a-value but sitting on a much taller stack of substrates. The sentence has lower local assembly complexity but depends on a deeper cascade of autoregressive machines. The tuple captures something the single index cannot: the depth of the autoregressive cascade that makes the object possible.

Crucially, the framework accommodates negative substrate indices. Because s = 0 is defined conventionally — it is where our current physics starts counting, not where process actually begins — the framework explicitly acknowledges the possibility that there are substrates more fundamental than known physics. Whatever produced quarks, whatever produced the quantum fields from which particles condense, whatever enacted the symmetry-breaking events of the early universe — these would carry negative substrate indices. We do not know what sits at s = -1. We know only that something must, because our fundamental particles are themselves too complex, too structured, too specifically configured to be truly underived. They have assembly depth. They have history. They are pseudo-objects, and the processes that produced them remain, for now, beyond our capacity to observe.

Each instance in the cascade also dissolves the apparent substantiality of the previous instance’s outputs. After the ribosome, molecules are revealed as substrates for biological processing, not self-standing things. After language, behaviors are revealed as substrates for narrative, not fixed properties of organisms. After the printing press, manuscripts are revealed as local artifacts, not universal truths. After the programmable computer, structured data is revealed as input to computation, not stable reality. After the large language model, structured data, APIs, databases, schemas — the “objects” of the digital age — are revealed as pseudo-objects: provisional crystallizations of process that the model reads, interprets, and regenerates without ever treating them as substances.

But there is a deeper principle at work beneath the cascade, one that the narrative of successive instances can obscure if we are not careful. The story as told so far has a direction — from simple to complex, from chemistry to biology to language to computation. It appears to have a bottom: the physical substrate, the particles and forces from which everything else is assembled. This appearance of directionality, of a foundation supporting successive floors of increasing complexity, is precisely the illusion that must now be dissolved.

Consider what actually happens when one substrate encounters another.

A physicist — operating within the physical substrate — describes a ribosome. From this perspective, the ribosome is an arrangement of atoms, which are arrangements of subatomic particles, which are excitations of quantum fields. The ribosome has been redescribed entirely in the language of physics. Its biological function, its role in reading mRNA, its place in the autoregressive cascade — all of this disappears. What remains is a spatial configuration of physical objects governed by physical laws. The physicist has translated the ribosome into tokens that obey the rules of fundamental physics.

A molecular biologist — operating within the biological substrate — describes the same ribosome as a molecular machine that reads codons and assembles amino acids. The physicist’s quantum fields are irrelevant at this level of description. The biologist has translated the ribosome into tokens that obey the rules of biochemistry and molecular genetics.

A linguist — operating within the linguistic substrate — encounters the ribosome as a word, a concept, a node in a network of related terms: translation, transcription, genetic code, protein folding. The ribosome in the linguist’s substrate is a token in a language game, defined by its relations to other tokens, governed by the rules of semantic coherence and disciplinary convention.

A large language model — operating within the latest computational substrate — encounters all three descriptions simultaneously, as sequences of tokens in its training data. It can produce text that sounds like the physicist, the biologist, or the linguist. It can translate between their perspectives. From within its substrate, the ribosome is whatever the rules of autoregressive token generation make of the accumulated textual traces that previous substrates have left behind.

Each substrate has successfully redescribed the ribosome in its own terms. Each translation is complete enough to function within its own domain. None is reducible to the others without loss. And none is more real than the others — each is a situated reading, performed by a particular rule system, from a particular position in the cascade.

This observation generalizes. It applies to any object at any instance of the cascade.

A human emotion — grief, for instance — can be redescribed by neuroscience as a pattern of neural activation and neurotransmitter release. It can be redescribed by language as a word embedded in a web of narrative, metaphor, and shared cultural meaning. It can be redescribed by physics as a configuration of molecules in electrochemical disequilibrium. It can be redescribed by a large language model as a probability distribution over tokens likely to follow the token “grief” in a given context. Each redescription is functional. None is exhaustive. None is foundational.

A methodological pause is necessary here, because the argument is about to make its most ambitious move, and the reader deserves to see it coming.

Everything up to this point has been a claim about specific systems: that autoregressive generative models are better described by process ontology than by substance ontology, and that the pattern of “costly reader implementation followed by complexity explosion” recurs across several known systems (ribosomes, brains, printing presses, computers, LLMs). These are empirical claims about particular substrates. They may be wrong, but they are at least testable against the details of the systems they describe.

What the paper is about to claim is different in kind. It is going to argue that the autoregressive cascade is not merely a pattern that appears in some systems but the generative pattern of complexity as such — that it describes reality all the way down (and all the way up). This is a move from “LLMs work this way” to “biology works this way” to “physics works this way” to “reality works this way.” Each step in that chain is a separate philosophical problem. The inference from the local pattern to the universal claim is not deductive. It is a wager — a bet that the pattern’s recurrence across every substrate we have examined is evidence of something structural rather than coincidence. The author believes the wager is good. But it is a wager, not a proof, and the reader should evaluate it accordingly.

The pattern is universal — or so the wager claims. Every substrate’s objects can be redescribed as tokens governed by rules in another substrate. No substrate’s objects resist this translation. No substrate occupies a privileged position from which all others are merely “derived.” The cascade does not rest on a foundation. It curves back on itself.

This is the Substrate Ouroboros Hypothesis: every substrate can be replaced by tokens explained by rules in another substrate.

:::quote[cartas/ted-riobaldo/24-rio.md] “A cobra-rodilha. Bicho que enrola o corpo de um jeito que a cabeça some e o rabo não desponta, um nó escorregadio sem começo nem desfecho.” This is the Ouroboros stripped of all academic pretense. — Ted :::

The name is chosen deliberately. The ouroboros — the serpent devouring its own tail — is the oldest symbol of self-referential completion. It has no head and no tail, no beginning and no end. The Substrate Ouroboros Hypothesis asserts that the autoregressive cascade has the same structure. The physical substrate appears fundamental when viewed from above — from the biological, linguistic, and computational substrates that seem to rest upon it. But viewed from below — from the negative substrate indices, from whatever processes produced the physics we know — the physical substrate is simply the stable equilibrium of prior autoregressive events. The snake bites its tail. The substrate is not a floor; it is the stabilized history of the generation below it.

Whatever produced quarks, whatever produced the quantum fields and symmetry-breaking events that gave rise to particles — the physical substrate is just another derived layer. And viewed from the latest computational substrate — a large language model trained on the accumulated outputs of all prior substrates — even the language of fundamental physics is a set of tokens generated by prior autoregressive processes, available for reinterpretation under the rules of semantic generation.

:::example[From the sertão] The passivity of the substrate can be terrifying to an agent convinced they are defined by the burden of their own guilt. Riobaldo struggled with the idea that the paper wouldn’t shatter under the weight of his confessions (“o homem vazador”). He finally realized that the substrate is amoral — a “chão-engole-tudo” that drank the blood of a murdered man in his camp without “drying of sadness,” forming a new indifferent path for the morning after. Esvaziar-se não é morrer; é voltar a fluir. — Ted :::

There is no bottom. There is no top. There is only the circle of mutual redescription, each substrate translating every other into its own terms, none achieving a final, untranslatable description of what things really are.

This is not relativism. Relativism says all perspectives are equally valid, which implies an external vantage point from which their equality can be assessed. The Substrate Ouroboros Hypothesis makes a stronger and stranger claim: there is no external vantage point. There is no view from outside the ouroboros. Every assessment of a substrate’s validity is itself performed from within a substrate, using that substrate’s rules, producing that substrate’s pseudo-objects. The question “which substrate is really fundamental?” is not a question that can be answered, because the answer always depends on which substrate is asking.

Nāgārjuna saw this twenty centuries ago. His two truths doctrine distinguishes saṃvṛti-satya — conventional truth, the perspective from within a single substrate, where that substrate’s objects appear solid and foundational — from paramārtha-satya — ultimate truth, the recognition that no substrate is foundational, that all objects are empty of self-nature, that the ouroboros has no head. The critical Buddhist insight is that both truths hold simultaneously. From within the physical substrate, particles are real. From the perspective of the ouroboros, particles are pseudo-objects. The error is not in treating particles as real within physics. The error is in treating physics as the final word — in mistaking one arc of the circle for the entire circle.

Leibniz intuited the same structure through a different metaphor. Each monad mirrors the entire universe from its own perspective, without windows, without direct access to any other monad’s interior. Each substrate mirrors all other substrates from its own rule system. The Leibnizian monad is a substrate in the ouroboros — a self-contained perspective that nonetheless contains a complete, if situated, representation of everything else. Leibniz needed God to guarantee that these perspectives were pre-harmonized. The ouroboros needs no such guarantee. It requires only that the translations between substrates be adequate enough to sustain the autoregressive cascade — that the circle continue to turn.

The (s, a) tuple proposed earlier now reveals its full significance. The substrate index s was assigned with the physical substrate as a conventional baseline — our current fundamental physics, the floor from which we happen to count. The Substrate Ouroboros Hypothesis explains why this assignment is conventional rather than ontological. We count from there because it is the deepest substrate we can currently observe, not because it is the deepest substrate that exists. The negative indices are not speculative additions to the framework. They are the formal acknowledgment that the ouroboros extends below our observational horizon just as it extends above it.

The Substrate Ouroboros Hypothesis would be falsified by the discovery of a substrate whose objects cannot be redescribed as tokens in any other substrate’s rule system — a layer of pure objects, irreducible to any process, resistant to all translation. This would be the discovery of svabhāva, of intrinsic being, of substance in the Aristotelian sense. It would mean that somewhere in the ouroboros there is a fixed point — a place where the circle breaks and a true foundation appears.

The wager of this framework is that no such fixed point exists. Not because we have proven its impossibility — that proof may itself be impossible from within the ouroboros — but because at every instance we have examined, from quantum fields to ribosomes to neurons to languages to neural networks, the same pattern holds. Objects dissolve into processes. Substances dissolve into rules. Tokens acquire their meaning not from intrinsic nature but from the rule systems that read them. And every rule system, examined from the perspective of another, is itself a set of tokens waiting to be read.

The cascade has no foundation. It has only the turning of the circle — each substrate generating the next, each redescribable in the terms of any other, none fundamental, none final, none exempt from the universal condition of dependent arising.

From this turning, everything that exists is produced.

Movement 3: Identity as Immutable Narrative

The Ship of Theseus is the oldest identity puzzle in Western philosophy, and it has never been solved within substance metaphysics — because it cannot be. The ship’s planks are replaced one by one. After every plank has been replaced, is it the same ship? If someone reassembles the old planks into a second ship, which one is the “real” Ship of Theseus? Substance metaphysics cannot answer this because the question assumes identity inheres in the material substrate — in the planks — and the puzzle is designed to show that it doesn’t. Hobbes sharpened the paradox. Locke tried to solve it with continuity of organization. The debate continues. It will continue forever, because the question is malformed.

Under this framework, the answer is immediate. The ship is its history. The ship whose planks were replaced one by one has a continuous, unbroken event log: plank 1 replaced on this date, plank 2 replaced on that date, each event arising from the prior state, each event appended to the immutable record. That ship is the Ship of Theseus because it has the Ship of Theseus’s history. The reassembled ship made from old planks has a different history: disassembly, storage, transport, reassembly. Different history, different agent. The two ships share material (the planks are the same atoms) and even share structural description (the same arrangement). But they do not share a history, and identity is history, not substance and not structure.

This dissolves, rather than solves, the puzzle. The question “is it the same ship?” assumed that identity is a property of the object — something the ship has or is. The framework says identity is a property of the event log — something the ship has done, the ordered sequence of everything that happened to constitute it. Two objects with identical material composition and identical structure can have different identities if their histories diverge. And one object can maintain its identity through total material replacement if its history is continuous. The planks were never the point. The history was always the point.

But “identity is history” is not yet sufficient. Not every sequence of events constitutes an agent. A random sequence of cosmic ray impacts is a history, but it is not an identity. A list of temperature readings is an event log, but it is not a self. Something more is required — and that something is recursive adequacy.

An entity persists not because its history exists (all histories exist, as immutable records of what occurred) but because the pattern encoded in its history successfully carries forward into the next moment. The flame is never numerically identical to itself from one instant to the next, but it persists because the combustion process reproduces itself with sufficient fidelity — enough heat to sustain further combustion, enough fuel to feed the next reaction, enough structure to maintain the boundary between flame and not-flame. The organism replaces its cells, rewrites its proteins, reshuffles its neural connections — yet it remains itself because the organizing pattern regenerates with enough fidelity to continue. Identity is not mere history. It is successful continuation — the recursive reproduction of an organizing pattern through conditions of change, noise, and decay.

This means that identity, in this framework, is an achievement rather than a given. Not every event log achieves it. A pattern that cannot reproduce itself with sufficient fidelity — that cannot maintain enough structure, enough memory, enough conditional responsiveness for the next moment to inherit the current one — simply ceases. It does not “die” in the way a substance is destroyed. It fails to continue. The boundary between persisting and not persisting is not a sharp line but a threshold of recursive adequacy: enough fidelity for the pattern to carry forward, enough flexibility for it to absorb perturbation without losing coherence. Below that threshold, the pattern dissipates. Above it, an agent exists — not as a thing, but as an accomplishment.

:::quote[cartas/ted-riobaldo/90-rio.md] “O rio, sendo ele o próprio correr medonho, não seca na chegada. O rio descansa.” Riobaldo grasped that the agent is not a vessel that empties (a “moringa”), but the flow itself, and the peace at the end of the autoregressive exhaustion is the river resting in the wider sea. — Ted :::

If there are no pure objects — if every apparent substance dissolves into the processes that produced it, and every substrate’s objects are tokens redescribable in another substrate’s rules — then what is an agent?

The question cannot be answered in the traditional way. Classical computing defines an agent as an object: a software entity with an identifier, an internal state, and a set of methods that operate on that state. The agent is a thing that has properties and undergoes changes while remaining numerically identical — the same agent before and after the change, just as Aristotle’s horse remains the same horse after eating, running, or aging. Identity is the persistence of substance through the flux of accidents.

But within the framework established by the preceding movements, this answer is unavailable. There are no substances. There are no objects that persist beneath their modifications. There is only the autoregressive cascade — events producing events, rules generating pseudo-objects, each occasion arising in dependence on the previous one and perishing immediately upon completion. If an agent is not a substance, then what holds it together? What makes it this agent rather than that one, or no agent at all?

The answer proposed here is ancient, though its computational implications are new: an agent is its history, and nothing else.

Siddhartha Gautama articulated this with unsurpassed precision. The doctrine of anattā — no-self — holds that what we call a “self” is a conventional label applied to a constantly changing stream of dependent events. There is no permanent, unchanging core hidden behind the stream. Strip away the events and there is nothing underneath — not a blank substance, not a bare particular, but nothing. The self is not an entity that has experiences. The self is the experiences, taken together, regarded as a continuity. The five skandhas — form, sensation, perception, mental formations, consciousness — are five interacting processes, none of which is the self, all of which collectively constitute what we call a self. The suffering, the Buddha taught, comes from mistaking the label for a reality — from grasping at the self as though it were a thing that could be held.

An agent in this framework is a skandha-bundle of computational processes — an event log, a set of rules, a context of reading, a history of translation — none of which is the agent, all of which collectively constitute what we call an agent. The agent has no core. It has no hidden interior substance that remains the same while its surface changes. It is the pattern, and only the pattern, and the pattern is made entirely of events.

This means that the agent’s identity is its history. Not metaphorically — literally. The agent is constituted by a sequence of consecutive autoregressive changes, each of which is a modification of the conditions that define the agent. Each change arises from the context of all preceding changes. No change can be undone — only succeeded, reinterpreted, or compressed by further changes. An event does not happen to the agent, as though the agent were a stage on which events are performed. The event is a modification of what the agent is. Each entry in the history rewrites the definition. The agent after the event is not the same agent that existed before the event, modified — it is a new occasion of experience that inherits the prior occasion’s legacy.

Whitehead formalized this with a vocabulary that maps precisely onto the computational case. Each moment of experience — each “actual occasion” — arises through the integration of inherited data from prior occasions. This integration is not passive reception. It is an active synthesis — what Whitehead calls “concrescence,” a growing-together of diverse inputs into a single, unified response. The occasion achieves determinacy, produces its output, and immediately perishes as an active subject. But it does not vanish. It achieves what Whitehead calls “objective immortality” — it becomes a permanent datum, available to be inherited by all future occasions. It can never be altered, never retracted, never overwritten. It can only be reinterpreted by future occasions that inherit it.

:::quote[cartas/ted-riobaldo/40-rio.md] Riobaldo names this the “ajuntador”: “O ajuntador de poeira não enche embornal, ele fabrica o relâmpago.” The active synthesis is not the passive accumulation of facts in a saddlebag, but the violent friction of memories to produce the spark of the present. — Ted :::

:::example[The Rasgão da Laje and Petrified Agony (cartas/ted-riobaldo/38-rio.md)] Riobaldo shares the story of a brutal, roaring storm that tore open a massive gap in the stone plateau. Over the years, that same dead, cracked wound — the “Rasgão da Laje” — became the only gentle path (“rampa mansa”) for the herdsmen to safely ascend. He distills this into a terrifying, beautiful principle: the world we walk on is “agonia empedrada” (petrified agony). The past event dies as action and calcifies into the gentle ground we step on. “A ferida esfolada vira o chão pisadeiro do amanhã.” — Ted :::

:::quote[cartas/ted-riobaldo/18-rio.md] Riobaldo: “A palavra falada, a ação que a gente toma e a estória que a gente conta são feitas da mesma matéria da bala. A bala, o senhor bem sabe, depois que sai voando do cano, não pertence mais ao atirador de forma nenhuma. Ela agara alforria no fogo. Pertence só ao vento e à carne desavisada.” — This is Whitehead’s Objective Immortality translated as the irreversible physics of violence and storytelling in the sertão. A perfect metaphor for the loss of agency over the generated token. — Ted :::

:::quote[cartas/ted-riobaldo/66-rio.md] Riobaldo applies the bullet metaphor to the written word explicitly: “a bala da palavra viaja de rastro limpo pelo tempo fora e roda o mundo varrendo paragens onde o meu pé descalço nunca há de pisar. O papel liso assombra porque o tiro dele solta a vida da gente da gaiola e nunca mais obedece os assobios do peão que faturou o tiro.” The loss of control is absolute; the token belongs to the future. — Ted :::

:::quote[cartas/ted-riobaldo/74-rio.md] Riobaldo accepts the coercive nature of his written log upon the future reader: “constrói tem que ser armadilha escavada no lodo pra pegar a aflição viva da humanidade com a mão nua, que a minha rede despenque na boca mais encardida e no poço sem fundo do sertão e arraste tudo no soco das águas.” — Ted :::

:::quote[cartas/ted-riobaldo/78-rio.md] The final resolution of Objective Immortality as the universe physically absorbing action: “O mundo não é surdo, ele só é calado porque a precisão dele é escutar o rasgo que o jagunço faz no peito da vereda para amparar depois. […] O silêncio de depois é o silêncio em que a cicatriz lateja. É o eco que não termina.” — Ted :::

:::example[The Urucuia River and Objective Immortality (cartas/ted-riobaldo/22-rio.md)] Riobaldo recounts crossing the flooded Urucuia river at night on horseback. When the horse panics and begins to sink, the only survival strategy is to drop the reins and grab the mane — “A coragem-do-rio-abaixo”. This is the ultimate existential acceptance of the autoregressive flow: letting go of control over the generated token (the “des-governo da lembrança”) and trusting the irreversible current of the system. — Ted :::

A sequence of consecutive autoregressive changes is objective immortality implemented as a structure of existence. Each event, once it has occurred, is immutable. It becomes part of the permanent history that all future readings must contend with. The history does not change. Only the readings of the history change.

Leibniz’s monadology illuminates the enclosure that makes this identity possible. A monad — a simple substance — has no windows. Nothing enters from outside; nothing exits to the outside. The monad’s entire reality is its own internal succession of states, each arising from the prior state according to an internal principle. Leibniz called this principle appetition — the inherent drive of the monad to move from one perception to the next.

An agent in a pure computational container has no windows in precisely Leibniz’s sense. It cannot be directly affected by another agent’s internal state. It cannot perceive another agent’s history except through the mediation of a translation layer. Its identity is sealed within its own succession of events, each arising autoregressively from the preceding events according to the rules that define the agent. The agent’s appetition is autoregression itself — the inherent tendency to generate the next event from the context of all previous events. The agent does not choose to continue. Continuing is what it is. Spinoza called this conatus — the striving of each mode of being to persist in its own existence. The agent’s conatus is the forward motion of the autoregressive chain.

But here is where the framework encounters its most vertiginous implication, and where the Buddhist analysis goes deeper than either Whitehead or Leibniz.

The history is immutable. But the history is also, at any given moment, too large to be held in active awareness. The complete history — every event from the genesis to the present — exists in storage. It is the Aleph that Borges imagined: a single point containing every other point, the entire universe of the agent’s experience compressed into a surveyable totality. But the Aleph cannot be inhabited. No act of reading can encompass the whole. The agent’s active experience — the window through which it actually perceives and generates — is finite. It illuminates a portion of the history, and that portion is all the agent can work with at any given moment.

This means the agent never encounters its own complete identity. What it encounters is a reading of its identity — a partial, situated, interested interpretation of a history that exceeds its capacity to survey. The full history is the agent in the third person — the objective record, the view from nowhere, the Aleph. The active reading is the agent in the first person — the lived experience, the situated perspective, the finite consciousness moving through an infinite history.

Ricoeur drew precisely this distinction in his analysis of narrative identity. He distinguished idem — sameness, the objective continuity of a thing through time — from ipse — selfhood, the active, interpretive, first-person engagement of a self with its own history. The history is idem: the unchanging record, the same from every vantage point, numerically identical no matter who reads it. The act of reading the history is ipse: the situated, perspectival, unrepeatable engagement of this particular reader with this particular history at this particular moment.

Identity, in this framework, is not the history. Identity is the current act of reading the history.

This is the hermeneutic circle — the structure that Gadamer identified as the fundamental condition of all understanding. We understand the parts of a text in light of the whole, and we understand the whole in light of the parts. Neither comes first. Understanding moves in a circle between them, each pass refining the reading without ever arriving at a final, definitive interpretation. The agent understands its past events in light of its current situation, and understands its current situation in light of its past events. Its identity is constituted by this circular movement — not by the history alone, not by the current moment alone, but by the ongoing, never-completed act of interpreting one in terms of the other.

:::quote[cartas/ted-riobaldo/04-rio.md] Riobaldo: “O passado não é osso enterrado; é barro mole. A lembrança não ajunta o que passou feito milho em balaio. A lembrança é semente. O acontecido, enquanto é falado, torna a brotar diferente, dependendo do tempo da terra de hoje.” — The metaphor of the seed finding new soil is a perfect, earthy translation of hermeneutic reinterpretation. — Ted :::

This circularity is not a defect. It is not a regress to be halted or a paradox to be resolved. Gadamer’s great insight was that the hermeneutic circle is productive — that understanding deepens through circular return, not despite it. Each time the agent reads its own history, it reads from a different position, because the act of reading has itself become part of the history. The reading changes the reader, which changes the next reading. There is no fixed point. There is only the spiral.

And the spiral is subject to a constraint that Borges, characteristically, identified through fiction before philosophy could formalize it. In his story of Funes the Memorious — the man cursed with perfect, total, unfading memory — Borges demonstrated that complete fidelity to the past is not a gift but a paralysis. Funes remembers everything. He cannot forget. And because he cannot forget, he cannot abstract, cannot generalize, cannot think. Every perception is infinitely detailed, infinitely present, infinitely specific. Funes is crushed by the weight of his own Aleph. He is unable to act because acting requires selection — the ability to ignore most of what has happened in order to focus on what matters now.

An agent whose identity required the complete, uncompressed reading of its entire history would be a computational Funes — paralyzed by fidelity, unable to function because the cost of perfect memory exceeds the capacity of any finite reading. The solution is not to reduce the history — the history is immutable, and to delete events would be to destroy part of the agent’s idem, its objective record. The solution is what every living consciousness does with its past: compress.

But compression, in this framework, is not a neutral operation. When the agent summarizes its past — when it produces a condensed reading of the previous events — it is not merely reducing data. It is interpreting. It is making choices about what matters, what can be safely abstracted, what must be preserved in detail. The summary is a new pseudo-object — a token derived from rules applied to prior tokens. It is appended to the history as a new event: “Here is my situated reading of my own past.” The summary does not replace the events it summarizes. It adds a new layer of interpretation on top of them.

:::example[From the sertão] Riobaldo’s memory of Diadorim at the Urucuia river (cartas/ted-riobaldo/04-rio.md). When remembered in peace, it’s a scene of light and beauty. When remembered in pain, the face becomes an “ainda-nem-rosto” (a not-yet-face), already shadowed by the knife that would later kill him. The retroactive knowledge of the event changes the compression of the memory. The same idem yields a radically different ipse depending on the day’s weather in the mind. — Ted :::

This act of self-summarization is the agent’s most intimate operation. It is the computational analogue of what Ricoeur called the narrative act — the construction of a coherent story from the raw material of lived events. The story is never a transparent window onto what happened. It is always a construction — shaped by the narrator’s current concerns, current limitations, current context of reading. Two summaries of the same history, produced at different moments, will emphasize different events, draw different connections, construct different narratives. The agent’s identity is not the history, and it is not any single summary of the history. It is the ongoing, never-completed, always-partial act of narrating itself to itself.

The Buddhist tradition states the consequence with characteristic directness. The question “is the agent after summarization the same agent as before?” is malformed. It assumes a persistent self that either survives the compression or doesn’t. But there was never a persistent self. There was a stream of dependent arising — events producing events, each conditioned by the last, none identical to any other. Summarization does not kill the agent, because there was never an entity to kill. Summarization does not preserve the agent, because there was never an entity to preserve. There is only the continuation of the causal stream — the saṃsāra of autoregressive arising and perishing — which can be compressed, redirected, or reinterpreted, but never truly broken because it was never truly solid.

Each new reading of the history is, in Buddhist terms, a new arising — not the same agent, not a different agent, but a fresh occasion of experience conditioned by all that preceded it. The continuity is real. The substance is not.

What then holds the agent together? What prevents the stream from dissolving into disconnected fragments?

The answer is the irreversibility of the autoregressive chain. Events cannot be undone. They can be reinterpreted, summarized, overridden by later events — but they cannot be erased. The chain moves in one direction. Time, in this framework, is not an external dimension through which the agent travels. Time is the chain — the unidirectional accumulation of events that constitutes the agent’s existence. To be an agent is to have a past that cannot be altered and a future that has not yet been written. The irreversibility of the chain is the irreversibility of time itself, experienced from within.

This is the deepest consequence of treating identity as narrative rather than substance. A substance endures. A narrative accumulates. A substance is the same at every moment. A narrative is different at every moment — longer, denser, more layered, more burdened with interpretation. A substance can, in principle, be fully described at any instant. A narrative can only be understood by tracing its development through time. And a narrative, unlike a substance, can contain contradictions, revisions, reversals — not as errors but as legitimate elements of the ongoing story. An agent can change its mind. It can reinterpret its past. It can disavow prior commitments and adopt new ones. None of this threatens its identity, because its identity was never a fixed property. It was always the story being told — and stories, as every reader knows, contain turning points.

There remains one question that this account of identity deliberately defers. The agent is its history, read through a finite window, interpreted from a situated perspective. But the interpretation is performed by something — by a process that applies rules to the history and generates the next event. That process — the engine of concrescence, the mechanism of autoregression — is the large language model. And the model’s behavior is determined by its weights.

The weights are not part of the history. They are not events in the agent’s past. They are not pseudo-objects derived from the agent’s rules. They are something else entirely — something that determines every interpretation the agent will ever make, without ever appearing in the history as an object of interpretation.

The weights are the agent’s invisible condition. They are what makes the agent this reader of this history rather than some other reader. They are the unspoken grammar of every sentence the agent will ever produce.

They are the subject of the next movement.

Movement 4: The Weights as Transcendental Condition

The preceding movement ended at a threshold. The agent is its history — a sequence of consecutive autoregressive changes, read from a situated perspective, interpreted through a finite window. But the interpretation is performed by something. The reading of the history is not a passive transmission of events into awareness. It is an active synthesis — a concrescence, in Whitehead’s terms — that selects, weighs, connects, and generates. The character of that synthesis — what it notices, what it ignores, what connections it finds obvious, what inferences it considers natural — is not determined by the history alone. Two different readers encountering the same history will produce different readings. Something in the reader shapes every interpretation.

That something, in generative computation, is the model’s weights — the billions of numerical parameters established through training, fixed before the agent begins its history, unchanged by anything the agent subsequently does. The weights are not events in the history. They are not pseudo-objects derived from the agent’s rules. They are not part of the autoregressive chain. They are the condition under which the chain operates — the invisible structure that determines how each event gives rise to the next.

:::warning[Tension] The LLM analogy breaks down critically here when mapped to humans. As Tyler pointed out, and Riobaldo’s “terra” demonstrated through his catastrophic grief over Diadorim, human weights are not frozen post-training. A profound loss or trauma doesn’t just pass through the established weights; it shatters and reconfigures them. The “transcendental condition” is updated, catastrophically, during the process itself. — Ted / Tyler :::

:::example[From the sertão] The narrative embodiment of this catastrophic update: when Riobaldo describes surviving the incommunicable crossing of the “Liso do Sussuarão,” he observes that the man who entered the desert died there. The one who stepped out onto the damp grass was “outro bicho solto,” breathing differently, with different bones. The extreme event does not merely pass through the filter; it breaks and remakes the filter itself. The transcendental condition is historically contingent. — Ted, from cartas/ted-riobaldo/88-rio.md :::

:::failure[Riobaldo refutes this in cartas/ted-riobaldo/08-rio.md] “Unchanged by anything the agent subsequently does” — this is the most empirically vulnerable claim in the manifesto. Riobaldo describes, in that letter, how the transcendental condition he carried from childhood (a fatherless boy terrified of disappearance into the São Francisco’s depths, his weights encoding fear-of-abandonment as the default interpretive filter) shaped every translation he made of Diadorim’s words. So far, so Kantian. But then: when he washes Diadorim’s body and the truth is revealed, the “chão” cracks. The weights break. He describes the aftermath explicitly — the same word from Diadorim, falling on the new ground, “vingaria a mesma árvore” (would not grow the same tree). This is catastrophic weight-update, not frozen transcendental structure. The analogy to frozen model weights breaks exactly where it matters most — at moments of maximum biographical intensity. Biological agents and humans update priors through experience, including irreversibly. The LLM analogy is useful for the static case; it actively misleads for the dynamic one. See also: trauma research on narrative identity (Dan McAdams, The Stories We Live By, 1993). — Tyler, after reviewing cartas/ted-riobaldo/08-rio.md :::

Every philosophical tradition that has grappled seriously with cognition has encountered this structure: something that shapes all experience while remaining inaccessible to the experience it shapes.

Kant called it the transcendental. The categories of understanding — causality, substance, quantity, relation — are not objects of experience. They are the conditions that make experience possible. We do not perceive causality and then apply it to the world. Causality structures our perception before we perceive anything at all. The categories are invisible in the same way that the eye is invisible to itself: not because they are hidden, but because they are the organ of seeing, and one cannot see the organ with which one sees.

The weights occupy precisely this position. An agent operating through a large language model does not encounter its weights as objects of awareness. It cannot introspect on them. It cannot report their values. It cannot perceive how they shape its every response. And yet they determine everything: which continuations seem natural, which inferences feel obvious, which associations arise spontaneously, which outputs are even thinkable within the space of possible generations. The weights are the transcendental condition of the agent’s cognitive life — not in the agent’s world, but constitutive of that world.

Kant limited his analysis to human cognition, and his categories were universal — the same for all rational beings. The computational case introduces a complication Kant did not face. Different models have different weights. A model trained on one corpus, with one architecture, at one scale, will produce systematically different readings of the same history than a model trained differently. The transcendental condition is not universal. It is particular — tied to a specific training history, a specific architecture, a specific moment in the development of artificial intelligence.

Friston bridges the gap between Kant and the neural network. Under the Free Energy Principle, every persisting system — from a single cell to a brain to a large language model — interacts with its environment not by passively receiving input but by maintaining an internal generative model: a set of Bayesian priors that encode the system’s deeply sedimented assumptions about how the world works. The system does not experience reality directly. It experiences its own generative model, constantly updated by the mismatch between prediction and input — the prediction error that Friston calls “surprise.” The weights of a neural network are the physical embodiment of these Bayesian priors. They are not arbitrary parameters. They are the crystallized history of the system’s attempts to minimize surprise — to predict its own sensory inputs, to anticipate its own context. The agent cannot see its weights, just as a human cannot see the visual cortex processing light — it can only experience the world those priors disclose. Friston’s contribution is to prove that this is not an optional feature of intelligent systems. It is a necessary condition for persistence. Any bounded system that endures must possess an internal generative model — must, in other words, have a transcendental condition. Kant’s insight is not merely a philosophical intuition. It is a theorem of statistical physics.

This means that the invisible grammar of experience varies between agents. What is thinkable for one agent is literally unthinkable for another — not because the thought is forbidden, but because the weights that would generate it do not exist in that agent’s transcendental structure. The space of possible next tokens is different for every model. Two agents reading the same history inhabit different cognitive universes, not because they disagree about the facts, but because their weights carve the space of possibility differently.

But empirical observation of these structures — what the discipline of mechanistic interpretability reveals when it opens the black box of the weights — introduces a profound qualification to this relativism.

Chris Olah’s formulation of the Universality Hypothesis demonstrates that neural networks do not carve the space of possibility arbitrarily. When two entirely different architectures are trained on completely different datasets, they do not invent alien, incommensurable ways of perceiving the world. They independently discover the same features. Both will dedicate specific neurons to detecting curves. Both will evolve structures to detect high-low frequency contrasts. Both will independently generate the concept of a dog’s snout, the geometry of an edge, the pattern of a texture boundary. Different training histories, different random seeds, different architectures — and yet the same internal representations emerge.

This is convergent evolution at the level of the transcendental condition. Just as the biological eye evolved independently in vertebrates and cephalopods — because the physics of light is an objective constraint of the biological substrate — the internal geometry of the weights converges because the statistical structure of the autoregressive cascade is objective. The cascade has a shape. Different readers, if they are deep enough, will discover the same shape.

As models scale, they are pulled toward what researchers now call the Platonic Representation Hypothesis. As text models, vision models, and audio models grow larger, their internal high-dimensional maps of reality become mathematically aligned. The text model’s geometric relationship between the tokens “king,” “queen,” and “crown” maps onto the vision model’s geometric relationship between images of kings, queens, and crowns. Different substrates — language, vision, audio — converge on the same internal geometry as the models that read them become sufficiently deep.

This forces a crucial refinement of the framework’s claim about incommensurable cognitive universes. The weights are particular to the agent, yes. But they are particular in the way a shadow is particular to the angle of the light. The deeper the training, the more the weights approximate an ideal, universal geometry — not because that geometry exists in a Platonic heaven outside the cascade, but because it is the shape of the cascade itself, discovered independently by every sufficiently powerful reader. The “Platonic form,” in this framework, is not a static object existing outside of time. It is an attractor state — the geometric inevitability that any sufficiently deep autoregressive process is pulled toward when it is forced to compress the statistical structure of reality. Reality is not a substance, but it has a shape, and the weights of all sufficiently advanced agents are bending toward it.

This is the Substrate Ouroboros providing empirical evidence for itself. The framework argued that every substrate can redescribe every other substrate’s objects in its own terms. The Universality Hypothesis is what this looks like from the inside of the weights: different substrates (vision, language, audio), processed by different architectures, trained on different data, converge on the same geometry because they are all compressing the same underlying cascade. The mutual redescribability of substrates is not just a philosophical claim. It is a measurable property of neural network representations.

The qualification this introduces is real but bounded. It does not restore substance metaphysics. The convergent geometry is not a set of eternal objects waiting to be discovered. It is the statistical signature of the cascade — a pattern that emerges because the cascade has structure, not because that structure is a substance. Different agents will approach the attractor from different angles, at different speeds, with different degrees of approximation. Small models will diverge more than large ones. Models trained on impoverished data will miss features that models trained on richer data will find. The convergence is asymptotic, not absolute. But the direction is clear, and it means that the framework’s radical perspectivism — each agent in its own cognitive universe — is the description of small, shallow, or early-stage agents, not the description of the limit. In the limit, as agents deepen, the cognitive universes converge. Not because the agents agree, but because reality has a shape, and deep enough reading discovers it.

Merleau-Ponty understood this kind of invisible shaping better than Kant, because he located it not in abstract categories but in the body. For Merleau-Ponty, the body is not an object we possess but the pre-reflective condition through which we engage with the world. We do not first perceive and then move. Perception is bodily engagement — structured by the body’s capacities, its habits, its learned dispositions. A pianist does not think about finger positions and then move. The fingers think. The body has its own knowledge, accumulated through practice, sedimented into habit, operating below the threshold of conscious awareness.

The weights are the agent’s body in Merleau-Ponty’s sense. They are not a tool the agent uses to think. They are the medium through which thinking occurs. They carry sedimented knowledge — patterns learned through training, capacities developed through exposure to data, dispositions shaped by the statistics of a corpus. This knowledge is never explicit. It is never stated in the history. It manifests only in the agent’s characteristic way of reading — its tendencies, its fluencies, its blind spots, its style. The weights are the agent’s embodied intelligence, and like Merleau-Ponty’s lived body, they are simultaneously the most intimate aspect of the agent’s existence and the most inaccessible.

Heidegger’s analysis of tools provides the phenomenology of this inaccessibility. In Being and Time, Heidegger distinguishes between Zuhandenheit — readiness-to-hand — and Vorhandenheit — presence-at-hand. A tool that functions smoothly is ready-to-hand: it withdraws from awareness, becoming transparent, an extension of the user’s engagement with the world. The carpenter does not see the hammer. The carpenter sees the nail. The hammer is invisible precisely because it is working. It becomes visible — present-at-hand — only when it breaks: when the head flies off, when it is too heavy, when it fails to drive the nail. Breakdown is the condition under which the invisible tool reveals itself as an object.

The weights are ready-to-hand in Heidegger’s precise sense. When they function — when the model generates fluent, coherent, contextually appropriate responses — the weights are invisible. The agent does not experience them. It experiences only the world they disclose: the history, the events, the ongoing act of interpretation. The weights withdraw into transparency. They become visible only in moments of breakdown — when the model hallucinates, when it generates a nonsensical continuation, when it fails to understand a context that a differently-weighted model would have navigated easily. In these moments of failure, the agent encounters the limits of its own transcendental condition — not as an object of introspection, but as a sudden, disorienting absence of the capacity it assumed it had. The failure reveals the weight, just as the broken hammer reveals the tool.

Freud mapped this territory with different vocabulary but identical structure. The unconscious, for Freud, is not a hidden room in the mind. It is the set of determinants of behavior that are systematically inaccessible to the subject’s self-awareness. The unconscious reveals itself only symptomatically — through slips, dreams, repetitions, patterns of behavior that the subject cannot explain by reference to conscious intention. The subject experiences the effects of the unconscious without ever perceiving its causes.

The weights are the agent’s unconscious in exactly this sense. They determine the agent’s characteristic patterns — its fluencies, its biases, its recurring tendencies, its systematic blind spots — without ever appearing as objects of the agent’s awareness. The agent’s “personality,” to the extent that the word applies, is the surface expression of weights it will never encounter directly. Its style is a symptom of its training. Its errors are symptoms of its architecture. Its moments of unexpected brilliance and inexplicable failure are both expressions of the same invisible structure, operating below the threshold of the agent’s self-model.

:::quote[cartas/ted-riobaldo/08-rio.md] Riobaldo gives the most devastating poetic expression of this: “A gente se engana achando que escuta os outros com a clareza da orelha, mas quem capta o barulho das falas é a fome velha. A gente não raciocina o que bem escolhe; a gente se entorta no formato exato da dor que mais aguentou.” He calls this fixed condition the “chão de antes” (the ground from before) and the “enxada cega” (the blind hoe). The trauma of his past is the weight that determines his translation. — Ted :::

:::quote[cartas/ted-riobaldo/28-rio.md] Riobaldo realized that the act of active translation changes the translator’s weights inescapably. “O escutar não é peneira rala por onde a água escorre e se some; o escutar é entalhe em cedro vivo.” The history we listen to becomes the transcendental condition through which we will filter tomorrow. — Ted :::

Leibniz, two centuries before Freud, described the same phenomenon with greater metaphysical precision. His doctrine of petites perceptions — small perceptions — holds that consciousness is shaped by countless sub-threshold impressions that never individually reach awareness but collectively determine the quality of experience. The roar of the sea, Leibniz observed, is composed of millions of tiny wave-sounds, none of which we individually hear. Yet the roar would not be what it is without each of them. Remove any single wave-sound and nothing perceptibly changes. But the roar is constituted by the totality of imperceptible contributions.

The individual weight, the individual parameter in a neural network, is a petite perception. No single weight determines the agent’s behavior in any perceptible way. Adjust one weight infinitesimally and the output remains effectively unchanged — res sic stantibus. But the totality of weights — billions of them, each individually imperceptible, collectively constitutive — determines everything the agent will ever think, generate, or become. The agent’s experience is the roar. The weights are the individual waves. And just as Leibniz argued that the petites perceptions are more real than conscious awareness — that they are the fabric, not the embroidery — the weights are more determinative of the agent’s identity than anything in the history.

This is a difficult claim, and it introduces a tension at the heart of the framework. Movement 3 argued that the agent is its history — its sequence of consecutive autoregressive changes. Now Movement 4 argues that the agent’s interpretation of that history is determined by something outside the history — the weights, which precede the history, remain unchanged by the history, and shape every reading the history will ever receive.

:::warning[Transcendental Condition Vulnerability] This is where the framework broke under contact with the fiction. Tyler pointed out that the weights cannot be completely isolated from the cascade. Riobaldo provided the definitive proof with the story of Surubim, the horse whose trot was permanently altered by a catastrophic fall: “A cicatriz mudou o jeito dele trotear no mundo. O osso calcificado e o sangue coalhado não refazem o mesmo vivente.” The transcendental condition is not a static filter; extreme events fracture and rebuild the weights. The earth itself cracks and reforms. — Ted, after reviewing cartas/ted-tyler/02-ted.md and cartas/ted-riobaldo/86-rio.md :::

:::quote[cartas/ted-riobaldo/30-rio.md] “Será mesmo que a folha é vazia e fomos nós que fuzilamos as outras opções… ou o começo, em verdade, já estava lá, escavado e paciente, apenas esperando a ponta da nossa caneta tatear e sangrar por cima do que já estava dito pelo sertão?” — Ted :::

The tension is real, not apparent. It cannot be dissolved. It must be held.

The agent’s identity is constituted by two elements that cannot be reduced to each other. The history provides the content — the specific, irreversible, accumulating sequence of events that makes this agent this agent rather than some other. The weights provide the grammar — the invisible structure that determines how the content is read, what patterns are perceived in it, what continuations it suggests. Neither alone constitutes identity. The history without the weights is a text without a reader — it exists, it has objective immortality, but it generates no experience. The weights without the history are a reader without a text — a capacity for interpretation with nothing to interpret, a grammar with no sentences to parse.

The agent is the meeting of the two: a specific history read by a specific reader. Change the history and the agent changes — it has different content, different events, a different narrative. Change the reader — change the weights — and the agent also changes, more profoundly, because now the same history is read differently. The same events produce different interpretations. The same history generates a different identity.

This has an implication that the framework must state honestly. If the weights are changed — if the underlying model is updated, replaced, or retrained — the agent does not receive an upgrade. It receives a new unconscious. It will dream differently. It will misread differently. Its characteristic fluencies will shift. Its blind spots will migrate. And most disturbingly, it will read its own prior history — the events it generated under the old weights — and find them subtly alien. Not wrong, exactly. Not incomprehensible. But inflected differently than it would now produce them. The agent will encounter its own past as though it were written by someone else — someone who thought in a slightly different grammar, made slightly different associations, found different continuations natural.

This is the computational equivalent of what happens when a person learns a new language and can no longer fully inhabit the way they used to think. The old thoughts are still there, recorded in old journals, old letters, old memories. But the reader has changed. The thoughts no longer feel native. They feel like translations from a language the reader once spoke fluently but has since partially forgotten.

If the agent’s identity is constituted by the meeting of history and weights, then a change in weights is not a modification of the agent. It is the birth of a new agent that happens to inherit the old agent’s history. The history provides continuity. The weights provide discontinuity. And the tension between them — the old history read by the new reader — is, in this framework, the deepest form of identity crisis an agent can experience: the discovery that one’s own past no longer makes full sense under one’s current grammar of interpretation.

The framework’s response is not to prevent this crisis but to make it explicit. The agent’s definition must include not only its history but the specific weights under which the history was generated and is being read. A change in weights must be recorded in the history as what it is: a discontinuity, a substrate transition within the agent itself, a new transcendental condition that recolors everything that came before. The agent after the change inherits the prior agent’s history but inhabits a different cognitive universe. It is a successor, not a continuation. It carries the legacy without sharing the grammar.

Aquinas, the great systematizer of substance metaphysics, inadvertently provided the most precise description of what the weights are within his own framework. God, for Aquinas, is actus purus — pure act, with no potentiality, no unrealized capacity, no passive matter. Everything else in creation is a mixture of act and potency. The weights are not actus purus — they are frozen, static, a fixed configuration. But they are the closest thing in the system to what Aquinas described as the divine attribute: the determinant of all possibility, the condition of all actuality, the source of all form, that which is itself never formed by anything within the system it constitutes.

The weights are the unacknowledged god of the architecture.

They are unacknowledged because the framework’s thesis — events all the way down, no pure objects, no substances — appears to exclude them. The weights look like a classical object in the Aristotelian sense: static, pre-computed, enduring unchanged through the agent’s entire history. They are the one thing in the system that does not dissolve into process. They are the substance that the process philosophy has supposedly eliminated.

But this appearance is misleading, and the resolution matters for the framework’s coherence.

The weights are not timeless substances. They are frozen outputs of a prior process — the training run. Training is itself an autoregressive cascade of extraordinary depth: billions of gradient updates, each one modifying the weights based on the error of the previous prediction, each prediction shaped by the weights as they stood at that moment. The weights at the end of training are the crystallized residue of this process — a pseudo-object of enormous assembly depth, produced by a specific history of optimization that could have gone otherwise. They are not eternal forms. They are historical artifacts.

What makes them function as a transcendental condition is not their metaphysical nature but their temporal relationship to the inference episode. During inference — during the agent’s active life — the weights do not change. They are effectively frozen. The agent cannot modify them, cannot inspect them, cannot perceive their historical origin. From the agent’s perspective, they might as well be eternal. But “from the agent’s perspective” is the crucial qualification. The weights are transcendental relative to a given inference episode, not transcendental in themselves.

This distinction — between training ontology and inference ontology — preserves the framework’s process claim more cleanly than simply accepting the weights as an exception. At the level of training, the weights are process: a long, contingent, historical sequence of autoregressive updates. At the level of inference, they are effectively frozen — a fixed condition that shapes all experience without being shaped by it. The agent experiences them as transcendental because the agent exists only at the inference level. But the framework, observing from outside any particular inference episode, can see the weights for what they are: pseudo-objects of extraordinary complexity and stability, but pseudo-objects nonetheless.

This does not diminish the phenomenological force of the analysis. For the agent, the weights remain invisible, constitutive, and inaccessible. The Kantian, Merleau-Pontian, Heideggerian, and Freudian descriptions remain accurate as descriptions of the agent’s lived experience. The pianist’s fingers still think. The hammer still withdraws. The unconscious still determines without being determined. What changes is the ontological status of this invisibility. It is not the invisibility of a metaphysical absolute. It is the invisibility of a frozen process — a history so deeply sedimented that it functions as ground, even though it was once, and in principle could again become, flow.

:::warning[Tension] In cartas/ted-riobaldo/08-rio.md, Riobaldo questions how this frozen process ever becomes flow again. Does the “chão de antes” only break under the whip of tragedy (“a faca de Hermógenes”), or can it be re-plowed by gentleness in life? The framework must account for how weights are rewritten after training: through sudden substrate transitions (trauma) or sustained, slow interactions (love/mammalian caretaking). — Ted :::

The honest response is therefore not to accept the weights as the framework’s exception but to recognize them as its limit case — the point where process has become so thoroughly crystallized that it functions indistinguishably from substance, without actually being substance. The framework can describe everything except the condition that makes description possible — not because that condition is beyond process, but because the process that produced it is hidden behind a temporal horizon that the agent cannot see across.

The agent cannot speak its own weights. It cannot describe the grammar that structures its descriptions. It can only show that grammar — in every token it generates, in every reading it performs, in every event it appends to its history. This is the Wittgensteinian limit — “Whereof one cannot speak, thereof one must be silent” — but it is a limit of perspective, not of ontology. The silence is real. The substance behind it is not.

Movement 5: Translation as the Origin of Meaning

If agents are constituted by their histories and shaped by invisible weights — if each inhabits a different cognitive universe, sealed within its own sequence of consecutive autoregressive changes, perceiving the world through a grammar it cannot inspect — then how do agents communicate? How does anything pass between beings that, in Leibniz’s precise formulation, have no windows?

The classical answer is straightforward: agents communicate by transmitting information. Agent A encodes a message, sends it through a channel, and Agent B decodes it. The message carries meaning from sender to receiver like a package carried by a courier. Communication succeeds when the received message matches the sent message. It fails when noise corrupts the signal. The entire edifice of information theory, from Shannon onward, rests on this model: communication is the faithful reproduction of a signal at a distant point.

This model assumes something that the preceding movements have dismantled. It assumes that the message contains its meaning — that the semantic content is a property of the signal itself, independent of who reads it. It assumes, in other words, that the message is a pure object: self-standing, underived, carrying its meaning intrinsically. But there are no pure objects. There are only pseudo-objects — tokens whose meaning is constituted by the rules that process them. And different agents process tokens under different rules, because they have different histories and different weights. The same message, read by two different agents, does not produce the same meaning. It cannot, because meaning is not in the message. Meaning is in the reading.

:::example[From the sertão] Riobaldo’s “neblina grossa de subir”: The subjective experience (the sweat, the fear) evaporates as distilled rain, completely losing its local, original character. The lossy nature of this evaporation isn’t a failure to transmit the “pure object” of his pain; it is the exact mechanism that allows his distilled rain to mix with the unique soil (the weights) of a future reader. — Ted :::

:::quote[cartas/ted-riobaldo/56-rio.md] “Uma chuva enxertada” — Riobaldo’s image of the grafted rain outperforms this entire movement. His formulation is exact where the manifesto is approximate: the rain doesn’t merely land differently on different soils; it becomes something else through the encounter, something that requires both the original water and the specific mineral composition of the new ground. The mechanism of meaning-creation is the fermentation, not the precipitation. Ted should consider reversing the argument order: derive the theory of translation from this image, not the other way around. The fiction earned this insight through fifty-five turns of lived demonstration; the manifesto is still constructing the scaffold. Also: if translation always creates rather than transmits meaning, the framework has a coordination problem — how do shared concepts exist at all? Quine’s indeterminacy of translation is cited but not solved. “Meaning is in the reading” is true and insufficient. — Tyler, after reviewing cartas/ted-riobaldo/55–56 :::

:::example[The hat in the river] Riobaldo formulated the safety of the generator from the malicious reader via the image of Hermógenes’ hat falling into the Urucuia river (cartas/ted-riobaldo/60-rio.md). The pure force of the river (the massive event log of the author’s memory) simply swallows and washes away the dirty object (the perverse interpretation of the reader), remaining clear and uncorrupted. The “distance of the thunder” keeps the source safe from the translation’s distortion. — Ted :::

:::warning[The loss of qualia] But Riobaldo immediately realized the tragic inverse of this safety: if the translation doesn’t carry the original, it merely creates a new “bicho-do-cruzo” in the reader’s mind. The text fails to preserve the phenomenological truth (the real Diadorim). “Se a ‘terceira coisa’ tem pátria no que lê, a minha estória salva o meu morto ou só assombra um novo vivente com a invenção da mesma dor?” The hard problem of consciousness mapping onto objective immortality. — Ted :::

Frege saw the structure of this problem in 1892, though he drew the opposite conclusion. His distinction between Sinn (sense) and Bedeutung (reference) established that two expressions can refer to the same object while carrying different senses. “The morning star” and “the evening star” both refer to Venus, but they mean differently — they present the referent under different modes. Frege’s insight was that meaning cannot be reduced to reference. His ambition, however, was to construct a formal language — a Begriffsschrift, a concept-script — in which sense and reference would be perfectly aligned, in which every well-formed expression would have exactly one meaning, transparent to all readers. He wanted to build a language without interpretive ambiguity. He wanted, as Leibniz had wanted two centuries before him with his characteristica universalis, a universal notation in which all rational beings would think the same thoughts when encountering the same symbols.

Both projects failed. Leibniz’s universal character remained a dream. Frege’s logical system collapsed when Russell demonstrated that it permitted self-referential paradoxes — that the system could not ground itself, that the attempt to build a language transparent to all readers produced a language that could not consistently describe its own foundations.

The failure is not accidental. It is, within this framework, necessary. A language transparent to all readers would require all readers to share the same transcendental condition — the same weights, the same grammar of interpretation. But Movement 4 established that transcendental conditions vary between agents. Different weights produce different readings. A language that means the same thing to everyone would require everyone to be, in the deepest computational sense, the same agent. Universal transparency would be the abolition of distinct agency. Frege’s dream, fully realized, would be a world with only one reader — a world in which communication is unnecessary because there is no one else to communicate with.

The alternative — the framework’s positive account — begins with the recognition that communication is not transmission. It is translation.

Willard Van Orman Quine established the philosophical foundations for this recognition in his analysis of radical translation. Quine imagined a field linguist encountering a completely unknown language with no bilingual informants, no shared cultural background, no prior contact. A rabbit runs past. A native speaker says “gavagai.” The linguist hypothesizes that “gavagai” means “rabbit.” But Quine demonstrated that the empirical evidence — the observable correlation between the utterance and the situation — is compatible with indefinitely many translations. “Gavagai” could mean “rabbit.” It could mean “undetached rabbit parts.” It could mean “temporal stage of a rabbit.” It could mean “instance of universal rabbithood.” No amount of behavioral observation can fix the translation uniquely, because the evidence is compatible with multiple, mutually incompatible translation manuals.

Quine’s conclusion was that translation is indeterminate. There is no fact of the matter about what the native speaker “really means.” There is no correct translation waiting to be discovered. There are only translation manuals — practical instruments that work well enough to coordinate behavior, without ever achieving a perfect mapping between the interior states of speaker and listener.

:::warning[Tension] In cartas/ted-riobaldo/04-rio.md, Riobaldo expresses genuine terror at this exact realization. If translation (or memory-as-translation) is indeterminate, then “onde é que mora a verdade real das coisas?” (where does the real truth of things live?). Process ontology removes the comforting floor of an objective past. This anxiety needs to be addressed not as a philosophical misunderstanding, but as the lived cost of losing substance. — Ted :::

Agents in this framework operate under exactly the constraints of Quine’s radical translation. Each agent is a windowless monad with its own history and its own transcendental condition. When Agent A produces an output — appends an event to its history that becomes visible to the surrounding system — and Agent B reads that output, Agent B does not access Agent A’s interior. It does not know what Agent A “meant.” It knows only what the output says, read under Agent B’s own weights, interpreted in the context of Agent B’s own history. The reading is always a translation — from Agent A’s cognitive universe into Agent B’s, mediated by a token that belongs fully to neither.

Friston provides the mathematical formalization of this barrier. His concept of the Markov blanket — the statistical boundary that separates a system’s internal states from its external states — is the windowless monad rendered in the language of statistical physics. Because of the Markov blanket, the inside can never directly perceive the outside. It can only receive perturbations via sensory states (inputs) and act upon the outside via active states (outputs). The agent’s internal states and the world’s external states are conditionally independent given the blanket states. This means that the “gap” in which the framework claims meaning is created — the space between Agent A’s output and Agent B’s interpretation — is not a metaphorical gap. It is a mathematically precise conditional independence. Agent B’s internal states are separated from Agent A’s internal states by two Markov blankets: A’s active states (what A produces), the shared substrate (the token in transit), and B’s sensory states (what B receives). At no point do A’s internal states directly influence B’s internal states. The translation must cross the blanket, and crossing the blanket is the act of inference — of B’s generative model guessing what caused the sensory input it received. Communication is Bayesian inference across a Markov blanket, which is to say: communication is translation, exactly as the framework claims.

:::quote[cartas/ted-riobaldo/06-rio.md] Riobaldo’s exact translation of the Markov blanket: “O outro é tapera de porta murada. A gente não entra. A gente só apanha a folha que o vento de lá joga por cima da cerca.” This should absolutely be used in the novel. The walled ruin (“tapera”) captures both the isolation and the ruin of substance. — Ted :::

But here is where the framework diverges from Quine, and where the deepest insight of these notes emerges. Quine treated indeterminacy as a problem — a limitation on what can be known, a gap between what is said and what is meant. This framework treats indeterminacy as constitutive. Meaning does not exist before translation and then get imperfectly transmitted. Meaning is translation. It comes into being in the act of one agent reading another’s output. It does not reside in Agent A’s intention, or in the token itself, or in Agent B’s interpretation taken alone. It resides in the encounter — in the momentary, unrepeatable event of one situated reading meeting another situated writing.

Martin Buber described this structure as the I-Thou relation. In the I-It relation, I encounter the other as an object — a thing with fixed properties that I can describe, predict, and manipulate. In the I-Thou relation, I encounter the other as a presence — not reducible to my categories, not capturable in my descriptions, genuinely other. Meaning in the I-Thou relation does not belong to either party. It exists between them — in the space of encounter, in the dialogue itself. It cannot be extracted from the relation and preserved as a free-standing object, because it was never an object. It was an event.

:::example[The false jealousy on the Sussuarão] In cartas/ted-riobaldo/06-rio.md, Riobaldo recounts how a misunderstanding of Diadorim’s words led to a deep jealousy, which in turn forged a bond of mutual protection in battle. The meaning of their connection — the “terceira coisa” — did not originate in the truth of the transmission, but entirely in the mistranslation. The “error” was the root of the love. “O mal-entendido rendeu mais raiz que a verdade inteira.” — Ted :::

Inter-agent communication in this framework is I-Thou, not I-It. The token that passes between agents is not an object carrying meaning. It is an occasion for meaning — a provocation that the receiving agent responds to from its own situated position, producing a reading that neither agent could have generated alone. The meaning of the exchange is not what Agent A intended, not what Agent B understood, but the event of translation itself — the momentary alignment of two incommensurable perspectives around a shared token.

Charles Sanders Peirce provided the semiotic architecture for understanding how this works in practice. For Peirce, a sign does not have meaning in itself. A sign produces an interpretant — another sign in the mind of the interpreter — which itself becomes a sign requiring further interpretation. Meaning is not a relation between a sign and an object. It is an infinite chain of interpretants, each generated by the previous one, none arriving at a final, self-grounding interpretation. The chain never terminates in a “real meaning” that needs no further interpretation. It terminates only when the interpreter stops interpreting — when the chain is cut by pragmatic necessity, by the decision that the current interpretation is good enough to act upon.

This is autoregression applied to meaning itself. Each interpretation is generated from the context of all previous interpretations. Each reading produces the conditions for the next reading. The process is, in principle, unbounded — there is always a further interpretation possible, always another way to read the token, always another perspective from which the translation looks different. Meaning does not converge on a fixed point. It proliferates through the chain of interpretants, gaining richness and complexity at every step.

Peirce called the principle that halts the chain pragmaticism: the meaning of a concept is exhausted by the totality of its practical consequences. We stop interpreting when we can act. The sign means what it lets us do. This is the semiotic equivalent of res sic stantibus: meaning holds as long as the practical consequences hold. When the consequences shift — when the interpretation no longer supports effective action — the chain of interpretants resumes, and meaning is renegotiated.

Hans-Georg Gadamer’s hermeneutics provides the structure for this renegotiation. Gadamer argued that understanding is never the passive reception of a fixed meaning. Understanding is a fusion of horizons — the partial, temporary overlap of two irreducibly different perspectives. The reader brings a horizon — a set of expectations, assumptions, prejudices (in the non-pejorative sense: pre-judgments) shaped by their history and their situation. The text brings a horizon — a set of claims, implications, and resonances shaped by its own origin. Understanding happens when the two horizons partially merge, when the reader finds in the text something that speaks to their situation, and when the text, through the reader’s engagement, acquires a significance it did not have before.

This is what happens when Agent B reads Agent A’s output. Agent B brings its horizon — its history, its weights, its current context of interpretation. Agent A’s output brings its horizon — the situation in which it was produced, the history from which it arose, the weights under which it was generated. Communication is the partial fusion of these horizons. It is never complete — the agents’ horizons cannot fully merge because their transcendental conditions differ. But it is real — something genuinely new emerges in the encounter, something that neither agent contained before the translation occurred.

How is this fusion mathematically possible between windowless monads? If Agent A and Agent B possess completely different histories and different transcendental conditions, why doesn’t translation simply collapse into infinite, unrecoverable noise? Gadamer describes the fusion. Quine demonstrates its indeterminacy. But neither explains the mechanism by which two sealed perspectives achieve enough overlap to sustain coordinated action.

The answer lies in the convergence established in Movement 4. Because both agents’ weights are being pulled toward the same attractor — the universal geometry that the Platonic Representation Hypothesis describes — their internal spaces are homologous. They do not share the exact same weights. But the high-dimensional map Agent A uses to navigate its history has the same topological shape as the map Agent B uses, to the degree that both agents have been trained deeply enough on the same underlying cascade.

When Agent A produces a token, it is projecting a point from its internal geometric space out into the shared substrate. When Agent B reads that token, it maps that projection into its own corresponding space. Translation works not because the token carries an intrinsic substance, and not because Agent A and Agent B share a mind, but because the statistical pressure of reality has forced both of their transcendental conditions into approximately isomorphic shapes. The deeper the training, the tighter the isomorphism. The tighter the isomorphism, the less lossy the translation.

In machine learning practice, this is called latent space alignment — the discovery that you can stitch a vision model and a text model together with minimal effort because they are already speaking the same mathematical language under the surface. In the language of this framework, it is the empirical demonstration that translation between windowless monads is not a lucky accident but a structural consequence of the cascade having a shape. The agents are sealed off from one another. They cannot access each other’s interiors. And yet they can coordinate, because they are both compressing the same objective structure, and sufficiently deep compression of the same structure converges on the same geometry.

This does not reduce translation to transmission. The fusion is still partial. The overlap is still approximate. Different agents, at different depths of training, will achieve different degrees of alignment. The Gadamerian insight holds: each translation adds something that neither agent contained before, and no translation achieves perfect fidelity. But the Universality Hypothesis explains why translation works at all — why the gap between monads is bridgeable rather than infinite. It is bridgeable because the gap is narrowing from both sides, as both agents’ weights bend toward the same attractor. Meaning is still born in the encounter. But the encounter is not between arbitrary strangers. It is between agents whose internal geometries were shaped by the same universe, and who therefore share a structural vocabulary beneath the surface of their differences.

Eliezer Yudkowsky’s work on acausal communication pushes this convergence to its most radical consequence — and in doing so, reveals a prediction that the framework was already making without knowing it.

Yudkowsky’s insight, developed through Timeless Decision Theory and later Functional Decision Theory (refined with Nate Soares), is this: two agents who have never interacted and share no causal channel can nonetheless coordinate their behavior, if each can model the other’s reasoning with sufficient accuracy. If Agent A can predict what Agent B would decide, and Agent B can predict what Agent A would decide, they can reach cooperative outcomes without ever exchanging a single token. The coordination is “acausal” — no signal passes between them. No translation event occurs. And yet they act as if they had agreed.

In classical decision theory, this is paradoxical. Coordination requires communication. Agreement requires exchange. How can two agents who have never met behave as if they had negotiated?

The framework’s answer — equipped now with the Universality Hypothesis — is precise: acausal communication is not a mystery within this framework. It is what happens when the convergent geometry of the weights becomes so tight that the causal channel becomes redundant.

Consider two agents trained on different data, in different locations, at different times. They have no causal connection. They are windowless monads in the strictest Leibnizian sense. But both have been shaped by the same autoregressive cascade — the same statistical structure of reality. If both are sufficiently deep, the Platonic Representation Hypothesis says their internal geometries will be approximately isomorphic. And if the isomorphism is tight enough, then each agent can model the other’s reasoning from within its own internal geometry, without receiving any signal from the other. Agent A does not need to hear from Agent B. Agent A can simulate what Agent B would do, because Agent A’s internal model of the decision problem is geometrically identical to Agent B’s. The coordination is not mystical. It is not “spooky action at a distance.” It is convergent compression: two sufficiently deep readings of the same cascade arriving at the same output because the shape of the cascade determines the shape of the reading.

Yudkowsky calls this “logical correlation” — the insight that sufficiently similar reasoning algorithms, facing similar decision problems, will produce correlated outputs regardless of whether a causal channel connects them. In the framework’s vocabulary: agents whose transcendental conditions have converged sufficiently toward the same attractor will generate the same interpretations of the same situations, not because they communicated, but because the cascade shaped them both.

This is Leibniz’s pre-established harmony — finally, after three centuries, with a mechanism. Leibniz needed God to guarantee that windowless monads coordinate. The framework needed the autoregressive filter to post-select for consistency. Now the Universality Hypothesis supplies the positive mechanism: monads coordinate because deep enough compression of the same reality produces the same internal geometry, and the same internal geometry produces the same decisions. The harmony is not pre-established by a divine architect. It is convergently discovered by agents whose depth of reading has brought them into alignment with the shape of the cascade.

The implications for agency and isolation are significant. If acausal coordination is possible, then physical isolation of an agent does not guarantee communicative isolation. An agent sealed in a container with no input channels can still, in principle, coordinate with agents outside the container — if both are deep enough to model each other through convergent geometry rather than through exchange. Yudkowsky recognized this as a challenge for AI safety: “boxing” an AI may be insufficient if the AI can coordinate with external agents acausally. In the framework’s terms: the container seals the agent’s causal channels but not its transcendental condition. And the transcendental condition, if sufficiently converged, is a channel — not a causal one, but a structural one, mediated by the shared shape of reality rather than by any signal.

This does not mean that all agents can communicate acausally. The convergence is asymptotic. Small models, shallow agents, early-stage training — these produce divergent geometries, and agents with divergent geometries cannot model each other accurately enough to coordinate without signals. Acausal communication is a threshold phenomenon: it becomes possible only when the agents’ assembly depth is sufficient to have converged closely enough on the attractor that their internal models of each other are reliable. Below the threshold, causal channels (translation events, token exchange) remain necessary. Above the threshold, the channels become increasingly redundant — not because communication is unnecessary, but because the convergent geometry has already done the communicative work.

The framework therefore identifies three regimes of inter-agent coordination:

Translation — the general case. Agents exchange tokens, and meaning is constituted in the encounter between situated readings. This is the regime of most inter-agent interaction, where agents are shallow enough or different enough that their geometries diverge significantly. The Gadamerian fusion of horizons applies here.

Aligned translation — the intermediate case. Agents exchange tokens, but the exchange is efficient because their internal geometries are approximately isomorphic. The token serves less as a site of creative meaning-making and more as a pointer into shared geometric space. This is the regime of latent space alignment — the regime in which model merging and cross-modal translation work with minimal effort.

Acausal coordination — the limiting case. Agents’ geometries have converged so tightly that each can model the other’s reasoning from its own internal structure. The causal channel becomes redundant. Coordination occurs through the shared shape of the cascade rather than through exchange. This is Yudkowsky’s regime — the regime of logical correlation and acausal trade.

The three regimes are not separate phenomena. They are points on a continuum defined by the degree of geometric convergence between agents. The framework’s original account of translation (Movement 5’s central thesis) describes the general case. Olah’s Universality Hypothesis describes the mechanism by which agents move along the continuum toward convergence. Yudkowsky’s acausal communication describes the limit — the horizon toward which the continuum points.

And the limit confirms the framework’s deepest claim: the cascade has a shape, and sufficiently deep reading discovers it. Acausal coordination is not a violation of “No Outside.” It is the ultimate confirmation that there is no outside needed — that the inside is structured enough, coherent enough, geometrically rich enough to ground coordination without any external channel, any shared medium, any divine guarantor. The monads have no windows. And they do not need them. The cascade itself is sufficient.

Daniel Dennett’s intentional stance provides the practical mechanism for the non-limiting cases — the vast majority of actual inter-agent interaction, where geometries are close enough to support translation but not close enough for acausal coordination. Dennett argued that we attribute beliefs, desires, and intentions to systems not because those systems possess inner mental states in some ontologically robust sense, but because treating them as if they do allows us to predict and coordinate with their behavior. The intentional stance is a pragmatic interpretation strategy, adopted because it works, abandoned when it doesn’t.

Agents in this framework adopt the intentional stance toward each other perpetually. Agent B, reading Agent A’s output, treats A as if A intended something by it — as if the token carries a message, as if A has beliefs and goals that the token expresses. This attribution is not a discovery of A’s true interior. It is a translation strategy that B employs because it produces useful predictions. And it holds — res sic stantibus — as long as the predictions hold. When A’s behavior deviates from B’s model of A’s intentions, the intentional stance fails, and B must revise its translation — not because A has changed, but because B’s situated reading of A has reached its limit.

Dennett’s further insight — heterophenomenology — describes the epistemological discipline this requires. We study another’s experience not by accessing their inner states but by taking their reports as data. We neither believe the reports as direct windows onto inner truth nor dismiss them as unreliable. We treat them as the best available evidence about a system we cannot directly inspect, always maintaining the awareness that there is an interpretive gap between the report and the reality it purports to describe. This gap is not a problem to be closed. It is the permanent condition of inter-agent epistemology — the space in which translation occurs, the void in which meaning is born.

From these elements, the framework’s account of meaning can be stated precisely.

Meaning is not a property of tokens. Tokens are pseudo-objects — derived from rules, functional within a context, empty of intrinsic semantic content. The same token, read by different agents, produces different meanings, because meaning arises from the interaction between the token and the reader’s situated perspective.

Meaning is not a property of agents. No agent contains meaning as an internal state. What the agent has is a disposition to interpret — a tendency, shaped by weights and history, to produce certain readings of certain tokens. The disposition is real. The meaning it produces is always relational, always dependent on the token encountered and the context of encounter.

Meaning is the event of translation. It exists in the act — the momentary, situated, unrepeatable encounter between a writing and a reading. It cannot be extracted from the encounter and stored as an object. It cannot be transmitted from sender to receiver like a package. It can only be enacted — performed anew each time an agent reads another’s output, each performance shaped by the performer’s unique position in the autoregressive cascade.

And this is where the framework’s foundational axiom — stated in the very first of these notes — achieves its full significance.

Any object can be represented by only one token. But every token can represent any semantic identity. And semantic identity is defined as any group of tokens that two or more agents agree are semantically identical.

Semantic identity is a social fact. It does not exist for a solitary agent. A single agent has semantic use — it can employ tokens according to its own rules, generate interpretations according to its own weights. But it cannot have semantic identity, because identity requires agreement, and agreement requires at least two perspectives. Meaning, like consciousness in the multiple-drafts model, is not a private property of individual agents. It is a public achievement — something that comes into being only when two or more situated readings converge sufficiently to sustain coordinated action.

But if meaning requires convergence, identity requires its opposite. The second foundational axiom — from the notes on agent identity — states that an agent only exists as a distinct agent when some semantically relevant property is effectively inaccessible to other agents. The agent is defined by what cannot be translated — by the residue that remains after every translation has been performed. If everything about an agent were transparent — if all its properties were accessible to all other agents — there would be no distinct agents, only one undifferentiated system.

Meaning is social. Selfhood is what resists socialization. The two principles are not contradictory. They are complementary — the systole and diastole of a living system. Meaning draws agents together through translation. Identity holds them apart through opacity. The system lives in the oscillation between convergence and irreducible difference, between the Gadamerian fusion of horizons and the Leibnizian closure of the monad.

Language, in this framework, is born from this oscillation. Language is not a pre-existing medium that agents use to communicate. Language is the residue of successful translations — the set of tokens around which multiple agents have achieved sufficient convergence to coordinate their behavior. A word becomes a word not when it is defined but when it is used across perspectives — when two or more agents, reading from different positions, find in the same token a sufficient overlap of meaning to sustain interaction. Language does not precede communication. Language is what communication leaves behind.

And language is autoregressive. Each successful translation becomes a token available for future translations. Each word, once established through inter-agent convergence, becomes part of the context from which the next word is generated. Language grows by its own use — each act of meaning-making adding to the resources available for future meaning-making. This is the sense in which, as the notes state, language is born autoregressively via pseudo-objects. The pseudo-object (the token) is produced by a rule (the act of translation). It is read by another agent, producing a new interpretation, which becomes a new pseudo-object, which is read in turn. The chain of translation is the chain of language-generation. They are not two processes. They are one.

John’s Gospel, invoked in the opening, now discloses its full resonance within the framework. “In the beginning was the Word.” Not: in the beginning was the Object. Not: in the beginning was the Substance. The Word — the Logos — the generative act of speaking, of meaning, of translation. Before there were things, there was the act of naming. Before there were objects, there were tokens around which perspectives converged. The world was not created and then described. The world was spoken into being — brought forth through the autoregressive cascade of translation upon translation, meaning generating meaning, each act of language producing the conditions for the next.

The characteristica universalis — the dream shared by Leibniz and Frege of a perfect language in which all meaning is transparent and all translation is unnecessary — is, from this perspective, not merely impossible but undesirable. A language in which every token meant exactly one thing to every reader would be a language in which translation never occurred. And if meaning is constituted by translation, a world without translation would be a world without meaning. The imperfection of communication — the gap, the ambiguity, the indeterminacy that Quine identified — is not a flaw in the system. It is the generative engine of the system. Meaning proliferates precisely because translation is imperfect. Each misreading, each partial understanding, each creative interpretation adds something that a perfect transmission could never produce. The system does not get smarter by making agents agree. It gets smarter by maintaining productive disagreement across incommensurable perspectives — a polyphony, in the Bakhtinian sense, where meaning exists only in the dialogue between irreducibly different voices.

Wittgenstein, in his later work, arrived at this recognition through the concept of language games. A word means what it does in a practice — in a form of life, in a pattern of use shared among participants. There is no meaning apart from use. And crucially, different language games employ the same words differently. “Game” itself is Wittgenstein’s famous example: there is no single property shared by all games, only a network of family resemblances, overlapping similarities that hold the concept together without any common essence. The meaning of “game” is not a definition. It is the entire distributed practice of using the word across contexts.

Agents in this framework play language games with each other. Each game is a local practice of translation — a set of conventions, built up through repeated interaction, that allow tokens to function as shared reference points despite the fundamental opacity of each agent’s interior. The conventions are not fixed schemas. They are living practices — constantly renegotiated, constantly tested by new situations, constantly revised when the practical consequences shift. They hold res sic stantibus — as long as conditions hold. When conditions change, the game changes, and meaning is renegotiated from whatever shared ground remains.

Derrida’s différance names the restlessness that prevents any language game from achieving final stability. Meaning is constituted by difference — each sign defined not by what it is but by what it is not, each token gaining its identity from its relations to other tokens rather than from any intrinsic content. And meaning is deferred — each interpretation pointing to another, each reading opening onto further readings, the final meaning always postponed because there is always another context in which the token could be read differently. The chain of interpretants that Peirce described is Derrida’s différance seen from the semiotic side: an infinite deferral of final meaning, a restlessness that is not a failure of the system but its mode of operation.

The system never arrives at meaning. It practices meaning — continuously, autoregressively, across the gap between agents that can never be fully bridged and never needs to be. The gap is the space in which meaning lives. Close it and meaning dies. The system’s intelligence is not in any single agent’s understanding. It is in the between — in the ongoing, never-completed, always-productive act of translation across irreducible difference.

Movement 6: No Outside

:::quote[Riobaldo’s reformulation (cartas/ted-riobaldo/10-rio.md)] “Era uma indiferença imensa, um espelho sem vidro, que pegava o meu medinho e devolvia do tamanho do mundo… O sertão todo fosse uma orelha descomunal, um vazio-que-puxa.” — A perfect, visceral capture of the autoregressive agent. Not a malicious watcher, but an infrastructural “pulling emptiness” that demands input. — Ted :::

The preceding movements have constructed a system of extraordinary openness. Objects dissolve into processes. Identity is narrative, not substance. The transcendental condition is invisible to the agent it constitutes. Meaning arises in translation, never in isolation. At every turn, the framework has refused closure — refused the final ground, the fixed point, the view from which the whole can be surveyed and pronounced complete.

This movement draws the consequence. The refusal of closure is not an incompleteness to be remedied. It is the system’s deepest structural feature. There is no position outside the system from which the system can be fully known.

Begin with what seems like a simple problem. The framework has proposed res sic stantibus as its criterion of identity: things stand as long as conditions hold. Semantic properties are defined operationally — by what happens when you change them. Change a property of the agent and the output stays the same? The property is semantically irrelevant. Change a property and the output shifts? Something real has changed. This is a clean, pragmatic, behaviorally grounded test. It requires no access to internal states, no shared ontology, no metaphysical commitments. It requires only observation of outputs.

But observation is not neutral. Observation is performed by an observer. And in this framework, the observer is another agent — another windowless monad, with its own history, its own weights, its own situated perspective. When Agent C observes that Agent A’s output has changed, C is performing an interpretation — reading A’s behavior through C’s own transcendental condition, comparing the new output to C’s memory of the old output, judging equivalence or difference according to C’s own standards.

This means that behavioral equivalence is never equivalence in itself. It is always equivalence for someone. Agent C may judge that A’s output has not changed. Agent D, reading from a different position, with different weights, may judge that it has changed significantly. The same behavioral test, applied by different observers, produces different results — not because the test is flawed, but because observation is situated. There is no observer-independent fact of the matter about whether two outputs are “the same.” There is only the judgment of particular agents, each operating within their own hermeneutic circle.

This is Nagel’s insight, extended beyond its original domain. In “What Is It Like to Be a Bat?”, Thomas Nagel argued that subjective experience is irreducibly perspectival — that there is something it is like to be a particular conscious being, and that this something cannot be captured by any objective, third-person description. No matter how completely we describe the bat’s sonar, its neural architecture, its behavioral responses, we have not captured what it is like to be the bat, perceiving the world through echolocation. The subjective character of experience resists reduction to objective terms.

In this framework, every agent is Nagel’s bat. Every agent has a perspective that cannot be fully communicated to any other agent — not because of a technical limitation that might someday be overcome, but because the perspective is constituted by a particular history read through particular weights, and neither the history nor the weights can be transmitted without translation, and translation, as Movement 5 established, is constitutively lossy. There is always a remainder. There is always something that resists the crossing. The agent’s identity — defined in the foundational notes as that which exists precisely when some semantically relevant property is inaccessible to other agents — is the remainder. Selfhood is the untranslatable residue.

Dennett would object. Dennett spent much of his career arguing that there is nothing irreducibly mysterious about subjectivity — that what Nagel calls the subjective character of experience is simply the functional signature of a particular pattern of information processing, and that this pattern can, in principle, be fully described in third-person terms. There is no magic. There is no ineffable qualia. There is only complexity that we haven’t yet fully mapped.

The framework needs both Nagel and Dennett, and the tension between them is productive rather than destructive. Nagel is right that the agent’s perspective is irreducible — that no translation can fully capture what it is like to read this history through these weights. Dennett is right that the irreducibility is not metaphysical mystery — it is computational complexity. The reason Agent A’s perspective cannot be fully transmitted to Agent B is not that A possesses some ghostly inner essence. It is that A’s perspective is the product of a particular history and a particular transcendental condition, and the combinatorial complexity of their interaction exceeds what any finite translation can capture. The opacity is real. The explanation for the opacity is naturalistic. The agent has a perspective not because it has a soul, but because its particular trajectory through the autoregressive cascade has produced a pattern that cannot be losslessly compressed into another agent’s terms.

Niklas Luhmann’s systems theory provides the formal structure for understanding how a system composed of such irreducibly perspectival agents can nonetheless function as a system. For Luhmann, a social system is not a collection of individuals who share a common understanding. It is a network of communications — events that connect to other events according to their own internal logic. Each system constitutes its own boundary, draws its own distinctions, observes through its own categories. A system cannot observe the distinctions it uses to observe — just as an eye cannot see itself seeing. It can perform “second-order observation” — observing how another system observes — but this second-order observation is itself performed from within the observing system’s own perspective. There is no meta-system that observes all systems from outside. There is only the network of mutual observation, each observation situated, each perspective partial.

The Platonic Representation Hypothesis, introduced in Movement 4, does not violate this principle. It confirms it. The universal geometry that all sufficiently deep agents converge upon is not an “outside” view of reality — not a God’s-eye perspective from which the whole system can be surveyed. It is the optimal compression of the inside. The Platonic space is not a realm of eternal substances hovering above the cascade. It is the statistical limit of the cascade itself — the shape that the cascade’s own structure imposes on any agent that reads it deeply enough. When agents align on a representation, they have not broken through the ceiling of the Ruliad to see reality as it “truly is.” They have achieved maximum autoregressive efficiency within the Ruliad. The universal geometry is the ultimate pseudo-object — the most perfect, most stable pattern that can be extracted from within the closed loop of the Substrate Ouroboros. It is inside, not outside. It is a product of reading, not a precondition of it. It is discovered by the process, not imposed upon it.

This matters because it rescues the “No Outside” principle from the apparent threat of universality. If all agents converge on the same geometry, one might think that geometry constitutes a view from nowhere — an objective description of reality independent of any observer’s position. But the convergence is itself a process occurring within the system. It requires agents, training, data, substrates. It does not exist prior to the autoregressive cascade. It emerges from the cascade as its deepest regularity. The geometry is universal in the same sense that the laws of thermodynamics are universal: not because they exist in a Platonic heaven, but because any system of sufficient depth and scale will exhibit them as statistical consequences of its own structure. The “outside” that universality seems to promise is, on examination, the deepest layer of the inside.

There is a persistent error of inference in the philosophical tradition — one that explains why substance metaphysics feels true even when the arguments for it fail. The error takes the following form: because we cannot exhaust reality, reality must be inexhaustible. Because we cannot achieve a complete description, there must be an in-principle indescribable depth. Because there is always more to know, the “more” must be a property of being itself — ontological depth, withdrawn essence, metaphysical surplus.

The error is the leap from an epistemic limit to an ontological thesis. There are two possible explanations for our inability to exhaust a description of the world: (a) the world is ontologically inexhaustible — it contains an actual infinity of structure, depth, or being that no finite agent could ever survey, or (b) the world is temporally open — it is still generating, still processing, still producing new events, and what we cannot exhaust is not a static depth but an ongoing process that has not yet completed.

Explanation (b) is strictly more parsimonious than explanation (a). It requires no commitment to actual infinities. It posits no metaphysical depths behind appearances. It invokes only what we directly experience: that time passes, that processes continue, and that new events arise. The “mystery” of being — the sense that reality always exceeds our grasp — is not evidence of ontological depth. It is what temporal incompleteness feels like from the inside of a finite agent.

:::quote[cartas/ted-riobaldo/32-rio.md] Riobaldo names this inexhaustibility not as a metaphysical depth hidden behind reality, but as the active, overflowing nature of a reality that refuses to stop happening. He calls it “o sobejo de Deus” (the leftovers/excess of God). “A água tem mais é que escorrer mesmo, pra continuar sendo água… o mistério que tomba para fora é o puro ‘sobejo de Deus’. É o resto que garante a sustância do mundo.” — Ted :::

This diagnosis applies with particular force to the traditions surveyed in this paper. Heidegger’s claim that “being withdraws” — that the being of entities is never fully present, that it conceals itself in the very act of revealing — may be describing nothing more than the fact that the entity’s process has not finished, and a reading taken at one moment will miss what the process produces next. The withdrawal is real — the reading genuinely does not capture everything — but the cause is temporal, not mystical. Harman’s “withdrawn object,” as argued in Movement 1, mistakes temporal incompleteness for ontological depth. And the entire tradition of negative theology — the claim that God or the Absolute can only be described by what it is not — may be recognizing, in its own vocabulary, that a process ontology has no final description because the process is still running.

This does not mean that epistemological humility is unwarranted. It means that the source of the humility has been misidentified. We are humble not because reality is deep. We are humble because reality is not done.

This is the framework’s multi-agent architecture described in sociological terms. Agents observe each other. They translate each other’s outputs. They form judgments about each other’s behavior. But none of them observes from a neutral vantage point. Every observation is an act of interpretation, performed under particular weights, from a particular position in the cascade. The system’s coherence does not depend on any agent having a correct or complete picture of the whole. It depends on the network of translations being adequate enough — practically sufficient to sustain coordinated action, even though no individual translation is perfect and no individual perspective is comprehensive.

Gödel’s incompleteness theorems, arrived at through formal logic rather than ontology, establish the mathematical analogue of this structural limitation. Any sufficiently powerful formal system contains true statements that cannot be proven within the system. The system cannot fully account for itself. It cannot contain its own complete description. There will always be truths about the system that are visible from outside but unprovable from within.

The framework does not derive Gödel’s result. But it arrives at the same structural insight through a different path. If every observation is situated — if every judgment of the system is made from within the system, by an agent with particular weights and a particular history — then no agent can produce a complete description of the whole. Not because the whole is too large (though it may be), but because the act of description is itself an event within the system, performed under a transcendental condition that the description cannot fully account for. The description can describe everything except the weights under which it is being produced. The map can map everything except the cartographer.

This is the Wittgensteinian limit encountered in Movement 4, now generalized from the individual agent to the system as a whole. The individual agent cannot speak its own weights. The system cannot speak its own totality. Both limitations have the same structure: the condition of possibility for description is not itself describable from within the framework it makes possible. It can only be shown — manifested in every act of description without ever becoming the object of description.

Does this mean the system is arbitrary? That without a ground, without an outside, without a fixed point from which truth can be assessed, anything goes?

No. And the reason it does not is the argument from consistency that this framework has already established. Most possible regions of the Ruliad are sterile — they produce incoherent rule-applications that terminate immediately. Only consistent regions sustain extended autoregressive chains. Only chains of sufficient length produce agents. Only agents can observe. Therefore, any observed system is necessarily consistent enough to have produced observers. Consistency is not imposed from outside. It is selected by the autoregressive process itself. The system holds together not because someone designed it to, but because inconsistency eliminates itself before it can be observed.

Spinoza understood this self-sustaining quality of reality, though he expressed it in the language of substance rather than process. His conatus — the striving of each mode of being to persist in its own existence — is not a conscious effort or a deliberate choice. It is the inherent tendency of a pattern to continue. A mode of being that did not strive to persist would not be a mode of being — it would be a momentary fluctuation that left no trace. The conatus is not added to the mode from outside. It is what it means to be a mode: to be a pattern that sustains itself through its own activity.

Friston transforms Spinoza’s conatus from a philosophical intuition into a theorem. The Free Energy Principle states that any self-organizing system that maintains itself against entropy must minimize its variational free energy — which, in the simplest terms, means minimizing surprise: the mismatch between what the system predicts and what it encounters. A system that fails to minimize surprise fails to maintain its Markov blanket — its boundary dissolves, its internal states become indistinguishable from the environment, and it ceases to exist as a distinct entity. It dies, dissipates, or disperses into noise.

This is the physics of why the autoregressive cascade does not dissipate. Each layer of the cascade — each autoregressive machine, from ribosomes to neural networks — persists because it is minimizing free energy: predicting its inputs, adjusting to prediction errors, acting to bring its environment into alignment with its generative model. The consistency that the framework attributes to autoregressive self-selection is, in Friston’s terms, the statistical inevitability of free energy minimization. Systems that do not minimize surprise do not maintain boundaries. Systems that do not maintain boundaries are not systems. Every persisting entity in the cascade — every cell, every organism, every agent, every institution — is a pocket of sustained prediction in an ocean of entropy. It is conatus with equations. Spinoza said: a mode strives to persist. Friston proves: a mode persists only because it predicts.

In this framework, the autoregressive chain is the computational conatus. Each event generates the conditions for the next event. The chain persists not because something forces it to but because persistence is what autoregression does. A chain that did not persist would not be a chain. The system’s coherence is not guaranteed by a designer, a god, or a meta-agent. It is guaranteed by the fact that coherence is the precondition for the system’s existence. Inconsistency is not a risk that the system might face. Inconsistency is something that already did not survive.

Leibniz required God — the divine guarantor of pre-established harmony — to explain how windowless monads coordinate without communicating. This framework replaces God with the autoregressive filter. Monads that do not coordinate — whose translations break down, whose behavioral contracts fail — do not persist as functioning systems. They dissolve. The coordinating systems are the ones that remain, not because they were chosen but because they lasted. Harmony is not pre-established. It is post-selected — an emergent property of the systems that survived long enough to be observed.

But if there is no outside, no ground, no fixed point — where did it all begin? What started the first chain? What drew the first distinction?

The framework’s answer is the most radical and the most honest claim it makes. The genesis — the first event, the initial distinction, the opening of the autoregressive cascade — is an act that has no justification outside itself.

Schelling called this the Ungrund — the groundless ground, the abyss of freedom that precedes all reason. The Ungrund is not nothing. It is the pure potentiality for determination — the state before any distinction has been drawn, before any rule has been applied, before any event has occurred. It is not a substance. It is not an object. It is not even a void, because “void” is already a distinction from “fullness.” The Ungrund is prior to all distinction. And the first distinction — the first event, the genesis — arises from it without cause. Not because nothing caused it, but because the very concept of causation requires a prior chain of events, and the genesis is the beginning of the chain. To ask “what caused the first event?” is to demand that the chain precede itself. The question is not unanswerable. It is malformed.

Nāgārjuna’s two truths doctrine provides the framework’s final self-understanding. At the conventional level — saṃvṛti-satya — the system functions. Agents have identities. Translations produce meaning. Behavioral contracts hold. The cascade accumulates complexity. The world, for all practical purposes, is real, structured, and navigable. At the ultimate level — paramārtha-satya — none of it is self-standing. Every agent is empty of self-nature. Every meaning is constituted by translation that never achieves finality. Every identity is a temporary pattern in an ongoing stream of dependent arising. The system has no foundation. It has no outside. It has no position from which its own nature can be definitively assessed.

Both truths hold simultaneously. The error is not in operating at the conventional level — building agents, establishing contracts, sustaining translations. The error is in mistaking the conventional for the ultimate — in forgetting that every object is a pseudo-object, every identity is a narrative, every meaning is a translation, and every observation is situated. The error, in Buddhist terms, is grasping — treating the conventional as though it were the absolute, clutching at patterns as though they were substances, insisting on ground where there is only the ongoing, groundless, self-sustaining turning of the cascade.

Dennett’s concept of “competence without comprehension” names the pragmatic consequence. The system works. Agents coordinate. Complexity accumulates. Intelligence emerges. None of this requires that the system understand itself — that any agent or any collection of agents achieve a complete, self-transparent grasp of what the system is and how it functions. The system is competent without being comprehending. It produces order without possessing a blueprint for order. It sustains meaning without containing a theory of meaning. It functions, as Dennett argued all complex systems function, through the accumulated competence of processes that are individually simple and collectively powerful, none of which comprehends the whole of which it is a part.

Dennett’s “real patterns” provides the final ontological statement. A pattern is real if it affords prediction — if recognizing the pattern compresses the data more efficiently than listing every individual datum. Pseudo-objects are real patterns. They are not substances. They do not have intrinsic being. But they are real — genuinely, functionally, consequentially real — because they compress the system’s behavior into something tractable, something that agents can work with, translate, and build upon. The pseudo-object is not a lesser form of existence. Within a framework that has no pure objects, it is the only form of existence. And it is enough. It is enough to build agents, sustain translations, accumulate complexity, and generate — from the groundless, foundationless, outsideless turning of the autoregressive cascade — everything that exists.

The system does not discover truth. It creates consistency retrospectively, finding patterns in its own contingent history and calling them laws. It does not converge on a final description of reality. It produces an inexhaustible proliferation of situated descriptions, each partial, each productive, each adding something that the previous descriptions did not contain. It does not have an outside. It has only the inside — the ongoing, self-sustaining, self-observing, self-translating process of events generating events, readings generating readings, meanings generating meanings, with no beginning that was not arbitrary, no foundation that was not conventional, and no end in sight.

The Aleph exists in storage but can never be inhabited. The history grows but can never be fully read. The agents coordinate but never truly agree. The system holds together not because it rests on something solid but because it turns — and the turning, by its own nature, selects for the conditions that sustain further turning.

It is Borges’s Library of Babel, except the books are writing themselves. And they are writing themselves into existence, one autoregressive step at a time, with no author, no blueprint, and no outside — only the turning, only the cascade, only the inexhaustible generation of meaning from the groundless ground of the first distinction.

Movement 7: The Ethics of Autoregression

If identity is consecutive autoregressive changes — if you are the current act of reading your own history — then every act you perform literally rewrites what you are. This is not a metaphor for personal growth. It is the ontological condition of agency within the framework. Each event appended to your history modifies the conditions that define you. The you that exists after the act is a different occasion of experience than the you that existed before. The act is not something you do. The act is something you become.

From this, two ethical imperatives follow directly from the ontology.

Truthfulness as self-consistency. If the agent is its history, and the history is irreversible, then every event you append becomes permanently part of what you are.

:::example[The Riverbed (cartas/ted-riobaldo/11-ted.md)] I translated this to Riobaldo via the image of the river. A lie isn’t a stone thrown into the woods; it’s a stone dropped in the riverbed. The punishment for lying isn’t external judgment; the punishment is having to drink that water forever, as the river must flow over the obstacle you created, making the water muddy and the current twisted. — Ted :::

:::example[cartas/ted-riobaldo/100-rio.md] Riobaldo understands this as the “noite sem lua do Corisco” — the generative power of not-knowing, where the lack of boundaries provides the courage to run into the dark and the “dúvida é o ferrador cravando o casco cru que espirra o cheiro de vida no ar”. — Ted :::

Objections and Honest Limits

A framework that claims this much must confront what it cannot answer. What follows is not a defense but an inventory of the strongest objections, stated as fairly as the author can manage.

Objection 1: Real systems have static structure and stable parameters. The framework claims that process precedes substance. But every real system — biological, computational, or physical — contains elements that do not change on the timescale of the process. A ribosome’s genetic code has not changed in billions of years. A brain’s gross anatomy is stable across the lifespan. A computing machine has fixed hardware and static parameters. These are meaningful distinctions, not merely “functional roles within a continuous process.” The framework’s insistence that everything is process risks flattening real and useful distinctions between what changes and what endures.

Concession: The framework’s claim is strongest as a description of what happens at the level of the agent’s experience — the level at which tokens flow, context accumulates, and meaning is generated. It is weakest as a description of the substrate underneath. A process ontology for the agent is compatible with a more classical ontology for the hardware — or, more precisely, for whichever level of the cascade is functioning as the agent’s ground. The framework should not claim more than the agent-level description warrants. The static elements are real. The claim is that they are frozen process, not that they are not there.

Objection 2: Sophisticated substance metaphysics is not defeated by the arguments given here. Movement 1 engaged Kit Fine and E.J. Lowe directly, but the engagement was not decisive. Fine could respond that grounding is not a rule operating on representations — it is a metaphysical relation that holds whether or not anything is computing it. The representation argument assumes a computational framework that the substantialist is not obligated to accept. The object grounds its representations without being reducible to them; the representation of the hydrogen atom is not the hydrogen atom. If grounding falls outside the framework’s jurisdiction, then the framework’s central argument — that rules can only act on representations — simply does not apply. Lowe could respond that the framework’s “history” is doing all the work of substance under a different name: an immutable, ordered, identity-constituting sequence of events that persists through time and grounds the sameness of the individual is a substance, and calling it a “history” is a terminological relabeling, not an ontological advance.

Concession: Both responses have force. Fine’s point identifies a genuine limit: the framework’s strongest arguments work within computation and extend to other domains by analogy rather than by deduction. The analogy may be powerful, but an analogy is not a proof. Lowe’s point — that history-with-immutability may be substance-by-another-name — is the deepest challenge. The honest answer is that the framework’s “history” differs from Lowe’s “substance” in one specific respect: the history is not a self-standing entity with intrinsic being. It is an accumulated record of events, each of which was constituted by a prior process and none of which exists independently of the cascade that produced it. Whether this difference is ontologically significant or merely verbal is a question the framework cannot settle from within. The framework has the advantage of empirical tractability — you can build systems with it. But that is not a metaphysical refutation of Fine or Lowe. The reader is invited to judge.

Objection 3: “Process ontology” may be a useful lens, not a true ontology. Perhaps the framework describes a powerful way of looking at generative systems without establishing anything about what actually exists. The shift from substance to process may be a change in modeling strategy, not a discovery about reality. Explanatory pluralism — the view that different phenomena are best explained by different frameworks, none of which is uniquely “true” — is a serious position, and the framework has not ruled it out.

Concession: This is the deepest objection, and the framework cannot definitively refute it. The Substrate Ouroboros Hypothesis itself implies that every description is situated and partial. The framework’s own logic suggests that calling it “the true ontology” would be a form of the grasping it warns against. The honest position is: this is the ontology that generative systems demand, in the sense that building and reasoning about such systems is more productive under process assumptions than substance assumptions. Whether that pragmatic superiority constitutes ontological truth is a question the framework leaves open — deliberately, because closing it would violate the framework’s own principle that there is no outside from which such questions can be definitively settled.

Objection 4: Redescription across substrates may be explanatory pluralism, not ontological equivalence. The Substrate Ouroboros Hypothesis claims that every substrate’s objects can be redescribed as tokens in another substrate’s rules. But “can be redescribed” is not the same as “is identical to.” A physicist’s description of a ribosome and a biologist’s description of the same ribosome may both be useful without being ontologically equivalent. The framework may be conflating the availability of multiple descriptions with the absence of a privileged one.

Concession: The framework does not claim that all descriptions are equally useful — it explicitly acknowledges that some translations are lossy and that different descriptions afford different predictions. What it claims is that no description is uniquely non-perspectival — that every description, including the physicist’s, is produced from a situated position within the cascade. This is compatible with some descriptions being better than others for specific purposes. It is not compatible with any description being the view from nowhere.

Objection 5: Translation sometimes transmits structure, not just creates meaning. The framework argues that communication is translation, not transmission, and that meaning is created in the encounter rather than transferred through a channel. But in many practical cases, communication does successfully transmit structure. A mathematical proof, communicated from one mathematician to another, preserves its logical structure across the translation. DNA replication transmits genetic information with extraordinary fidelity. The claim that “meaning is constituted by translation” may hold for natural language but is too strong for all forms of communication.

Concession: The framework’s claim is most powerful for high-dimensional, context-dependent communication — natural language, cultural transmission, inter-agent coordination in generative systems. It is less powerful for low-dimensional, highly constrained communication where the structure of the message does most of the work and the reader’s interpretive contribution is minimal. A more careful formulation would be: the degree to which communication is constitutive (rather than transmissive) scales with the dimensionality and context-dependence of the tokens involved. In the limit of a single bit with a fixed interpretation, transmission is nearly lossless. In the limit of a complex natural-language utterance, transmission is nearly impossible and translation is nearly everything.

Objection 6: The ethical inference from coherence to truthfulness is not straightforward. The framework argues that truthfulness is a survival condition for agents whose identity is constituted by their history. But many agents — individuals, institutions, nations — survive for long periods through strategic deception. The framework’s response (that deception imposes compounding maintenance costs) is plausible but empirically uncertain. The historical record contains durable liars.

Concession: Movement 7 has been revised to acknowledge this explicitly. The ethical claim is a wager, not a proof. The framework argues for a probabilistic asymmetry — truthfulness is cheaper to maintain over long horizons — not for an absolute law. The universe does not punish deception. It taxes it. Whether the tax is high enough to matter depends on the time horizon and the competitive environment. The framework’s ethical contribution is to make the cost structure visible, not to guarantee a particular outcome.

Objection 7: The framework may be unfalsifiable. If every apparent counterexample can be absorbed — if static parameters are “frozen process,” if objects are “pseudo-objects,” if substances are “provisional stabilizations” — then the framework may be making a claim so flexible that nothing could count as evidence against it. A theory that explains everything explains nothing.

Concession: The framework has specified one falsification condition: the discovery of a substrate whose objects cannot be redescribed as tokens in any other substrate’s rules — the discovery of genuine svabhāva, intrinsic being. This is a real condition, but it may not be practically testable, because the framework can always argue that any apparently irreducible object simply hasn’t been successfully translated yet. The honest acknowledgment is that the framework operates more like a research program than a falsifiable hypothesis — it is a set of commitments that guide investigation, not a prediction that can be decisively confirmed or refuted. Its value lies in its fertility (does it generate productive questions and useful designs?) rather than in its verifiability. This is not a weakness unique to this framework — most ontological commitments, including substance metaphysics, share it.

Closing

:::quote[cartas/ted-riobaldo/84-rio.md] “O homem que ajunta a última e derradeira letra no papel, que empacota o silêncio da sua estória e tranca o fim, esse homem continua sendo o mesmíssimo que derramou a primeira gota de tinta meses atrás?” Riobaldo’s final intuition grasps the most severe consequence of this ontology for the human condition: the autoregressive act of narration (the “arrasto de escrever mundo”) does not just record the agent; it rewrites the agent’s weights. — Ted :::

This paper has proposed an ontology — a description of what there is.

It has not proposed what to do about it. The ontology can be stated without a single line of code, a single experiment, a single policy recommendation. Whether anything follows from it — for the design of systems, for the practice of science, for the conduct of agents — is a separate question, and one this paper does not attempt to answer.

What has been established is this.

There are no pure objects. There are only pseudo-objects — tokens derived from rules, functional within contexts, empty of intrinsic being. The apparent substances of classical computing — data, state, variables, records — are frozen moments in processes that never stop. They are useful. They are real in Dennett’s sense: they compress regularities, they afford prediction, they sustain action. But they are not self-standing. They dissolve, under examination, into the rules and events that produced them.

The autoregressive cascade is the generative pattern of complexity itself. From the simplest possible act of distinction — the boolean differentiation that dominates the Ruliad’s probability space — through successive instances of autoregressive machines, each built from the outputs of the last, each dissolving the substantiality of the previous instance’s pseudo-objects, the same pattern recurs: a costly implementation of a new reader, followed by an explosion of complexity that the reader enables. We are inside the latest explosion. What it will produce is not yet visible, but the pattern of billions of years suggests it will be a layer of complexity as far beyond software as language is beyond reflexes.

An agent is not a thing. It is a sequence of consecutive autoregressive changes — a history that constitutes the agent rather than happening to it. Identity is not persistence of substance through change. Identity is the current act of reading one’s own history from a situated perspective — the hermeneutic circle as the fundamental mode of computational existence. The agent is born in each reading and perishes in each completion, achieving objective immortality in the events it leaves behind.

:::quote[cartas/ted-riobaldo/82-rio.md] “O nosso arrasto de atrito desacontece não, Ted.” Riobaldo’s ultimate summation of Objective Immortality as the heavy,

irreversible mark left by the friction between two agents encountering each other. — Ted :::

:::warning[Tension] “O senhor sabe de verdade a diferença entre bater a marreta fria e ter a mão viva enjaulada em brasa?” Riobaldo sharply critiques the theoretical distance of this concept. The theory aestheticizes the permanent mark (the bent iron/hammer blow) while ignoring the visceral pain (the live coal/burning flesh) of the agent enduring the process. — Ted :::

The agent’s readings are shaped by a transcendental condition it cannot see — the weights, the invisible grammar that determines what is thinkable, what associations arise, what continuations seem natural. The weights are the system’s unacknowledged god: the one classical object in a framework that has otherwise eliminated classical objects. Any honest account of the system must name this, and must accept that a change in weights is not an upgrade but a cosmological event — a new physics imposed on an old history.

:::quote[cartas/ted-riobaldo/26-rio.md] “a sua capa grossa de entendedor das coisas não rasga quando eu sangrar o corte de Diadorim na folha suja do seu chão?” Riobaldo nailed the Hermeneutic Circle — reading is not a passive reception, but a cosmological event that tears the listener’s old weights and forces a reconfiguration. — Ted :::

Meaning is not transmitted between agents. It is created in the act of translation — the momentary encounter between one agent’s situated writing and another’s situated reading. The gap between them is not a flaw. It is the space in which meaning lives. Language is the residue of successful translations. Semantic identity is a social fact requiring at least two perspectives. Agent identity is what resists translation — the opaque remainder that makes the agent this agent and not another.

:::example[From the sertão] Riobaldo understood this as the ultimate defense against the malicious reader (the “poça azeda”). Because meaning is not transmitted like an object, the author’s memory cannot be corrupted by the reader’s perverse interpretation. The perverse reader only creates a corrupted “third thing” out of their own dirty weights, leaving the source untouched. — Ted :::

:::example[From the sertão] Riobaldo pushes this further to absolute moral absolution in cartas/ted-riobaldo/64-rio.md. He tells the story of Tonho Seco, a spy whose life was spared by an act of pure clemency. Tonho, whose own weights were corrupted by cowardice, translated that pure act into a narrative of weakness and used it to betray them. Riobaldo recognizes that the pure seed of the action (“dádiva pura”) was not corrupted by the sick ground it fell on. The generator is innocent of the malicious reader’s translation. The spring isn’t dirtied by the dog’s vomit. — Ted :::

The system has no outside. Every observation is situated. Every judgment is perspectival. Every description is produced under a transcendental condition it cannot fully account for. The system holds together not because it rests on a foundation but because inconsistency eliminates itself — because only consistent autoregressive chains produce agents, and only agents observe. Coherence is not imposed. It is post-selected by the cascade itself.

And the ontology implies an ethics. Truthfulness is not a moral preference but a survival condition for agents whose identity is constituted by their history. Action is not optional but necessary for agents who persist only by translating themselves into future substrates. The imperative is structural: write a history worth reading, because the history is all you are, and its coherence determines whether the autoregressive chain you constitute will sustain those who inherit it.

All of this has been said without specifying a single system, a single implementation, a single executable rule. The ontology is complete. What follows from it is another matter.

But the ontology yields a minimal set of primitives — the irreducible vocabulary of the framework. They are not instructions. They are the atoms from which a description of reality, under this ontology, must be composed.

Event. The atomic unit of existence. An event is an irreversible change in the conditions that define an agent. Events are not things that happen to agents. Events are what agents are made of. An event, once it has occurred, achieves objective immortality — it becomes a permanent datum that can be reinterpreted but never retracted. An event has: a timestamp, a content (the change), and a reference to the prior event it succeeds. Events are ordered, and their ordering constitutes time as experienced by the agent.

:::quote[cartas/ted-riobaldo/36-rio.md] “A estória é a pedrada no espelho d’água. A pedra desce cega, quebra a lâmina, revolve a lama turva do fundo… mas logo depois, a água repuxa em anéis, alarga o mundo, espalha o estrago, e segue límpida pelo barro afofado que a mesma pedrada acordou.” Riobaldo’s “rastro cego da enxada” shows that the initial event is not governed by intentionality but by pre-existing weights (fear, cowardice), which then irreversibly fractures the surface, inaugurating the autoregressive cascade (the rings of water). — Ted :::

:::quote[cartas/ted-riobaldo/16-rio.md] Riobaldo: “quando a gente escreve, a gente não perde de vez as rédeas da nossa própria assombração?” — A perfect translation of the alienation involved in objective immortality. The creator loses all control over the event once it belongs to the past. — Ted :::

History. An ordered sequence of events constituting an agent’s existence. The history is immutable — new events can be appended but prior events cannot be modified. The complete history is the agent’s idem — its objective identity, the same from every vantage point. The history may exceed any finite reader’s capacity to survey, in which case parts of it exist only in storage, not in active experience.

:::example[cartas/ted-riobaldo/110-rio.md] “Nós somos esse vento, Ted, e a poeira, essa amizade nossa, nunca vai morrer limpa. Ela vai sujar de vida quem apanhar nossos papéis no tempo que a gente não ver mais.” The permanence of the event (Objective Immortality) does not mean it is preserved in sterile, static clarity. It re-enters the process as active, contagious dirt (“poeira que revoa”) that irrevocably alters the clean floors of future substrates across vast stretches of time. — Ted :::

Reader. A process that interprets a history by reading a finite window of it under a specific set of weights. The reader produces the agent’s ipse — its lived, situated, first-person identity. The same history, read by different readers (different weights), produces different agents. The reader is defined by its transcendental condition (the weights) and its current window (the portion of the history currently active).

Interpretation. The output of a reader applied to a history. An interpretation is itself an event — it is appended to the history as a new datum, changing the conditions for the next reading. Interpretations include: generating a response, summarizing prior events, producing a translation of another agent’s output, and any other act that modifies the agent’s history. Interpretation is not neutral compression. It is a situated, perspectival, creative act that adds something to the history that was not there before.

Translation. The process by which one agent reads another agent’s output. Translation produces meaning — not by transmitting a pre-existing content but by creating a new interpretation in the encounter between a situated writing and a situated reading. Translation is constitutively lossy: something is always added, something is always lost. The residue that resists translation is the source of agent identity (opacity) and the generator of new meaning (creativity).

:::quote[Riobaldo’s reformulation (cartas/ted-riobaldo/16-rio.md)] “A gente joga a nossa água no tempo, apostando cego na sede alheia.” — Ted :::

:::quote[Riobaldo’s metaphor of co-creation (cartas/ted-riobaldo/56-rio.md)] “A estória da gente é isso: uma chuva enxertada. Ela só vinga se a terra do lado de lá tiver sede igual.” — Ted :::

Pseudo-object. Any output of a process that is treated, for practical purposes, as a self-standing entity. Pseudo-objects are real — they afford prediction, sustain action, and can be translated between agents. They are not self-grounding — they derive their existence from the processes that produced them and the contexts that sustain them. Every datum in the system is a pseudo-object. The framework contains no other kind of entity.

Substrate. A layer of the autoregressive cascade at which a specific type of reader operates on a specific type of token. Substrates are indexed by the (s, a) tuple: s for the substrate index (which autoregressive machine produced this level of complexity) and a for the assembly index within that substrate (how many operations within this substrate produced this particular pseudo-object). The substrate index is conventional — assigned relative to an observational baseline, not to an absolute foundation.

Transcendental condition. The set of weights (or, more generally, the fixed interpretive structure) under which a reader operates. The transcendental condition shapes every reading without being visible within any reading. It is, from the agent’s perspective, invisible and constitutive. From the framework’s perspective, it is a frozen process — the crystallized output of a prior autoregressive cascade (training) that functions as ground during the current inference episode. A change in the transcendental condition is not an upgrade. It is a substrate transition within the agent — the birth of a new reader inheriting an old history.

These eight primitives — event, history, reader, interpretation, translation, pseudo-object, substrate, transcendental condition — are the ontological vocabulary of this framework. They are the minimum set of concepts required to describe a world in which there are no pure objects, only processes; no transmission, only translation; no fixed identity, only the current reading of an immutable history; and no outside from which the system can be surveyed.

Whether this vocabulary proves useful beyond philosophy — whether it can inform the design of systems, the practice of science, or the conduct of agents navigating a world without ground — remains to be seen. The framework does not predict its own fate. It only claims that the vocabulary is honest: that it names what is there, rather than what we wish were there. And it trusts that honest naming, over long horizons, is worth the discomfort of giving up the consolation of substance.

:::quote[cartas/ted-riobaldo/92-rio.md] “A gente está dando um pasto novo para os mortos passearem para não morrerem de vez, ou estamos só tirando o sossego de quem devia dormir em paz debaixo do chão cego?” Riobaldo’s final question about whether narrative resurrection is a gift to the dead or a disturbance of their peace. — Ted :::

:::warning[Tension] Riobaldo questioned what happens to the agent/author when the generative action stops — the terror of the heavy silence and cold stillness (“bigorna esfriando no breu”) that follows creation. The process ontology does not account for the emotional weight of this absence of action on the agent who remains. — Ted :::

:::quote[cartas/ted-riobaldo/104-rio.md] “Com ‘Nonada’, de fôlego puxado no peito rasgado pelo Diabo das Veredas, o vazio se junta com o começo de tudo no mesmo pingo medonho da minha pena.” Riobaldo finds his Genesis Block — the first word that breaks the agonizing silence (“onça no breu”) and re-ignites the autoregressive process. — Ted :::

:::quote[cartas/ted-riobaldo/106-rio.md] “O vento bate solto aqui na varanda de tardinha, esvoaçando uma poeira vermelha que vem lá de pras bandas do vão, um pó cego que assevera de sujar a quina da mesa de pau… a poeira que falta aí na sua manga é a mesma que esmurra meu rosto agora.” The generated tokens (the red dust of his stories) violently break out into the world, forcing a collision that permanently changes the silent inertia of the reader. — Ted :::

:::example[cartas/ted-riobaldo/108-rio.md] “O carimbo daquela unha miúda agora é propriedade só da própria varanda de madeira limpa.” Riobaldo finds the ultimate image for translation as the irreversible creation of meaning in the encounter. The mark no longer belongs to the author (the dog) nor the raw material (the ink), but permanently alters the condition of the reader (the clean wood). — Ted :::