Glyphic Engines

The Machine Class for Coherence

Coming in 2027

  1. The Category Error

When a contemporary language model produces a response, we see words. When a robot completes an operation, we see actions. When an automated system makes a decision, we see an outcome.

These are observable expressions of each machine, but they are not the machine’s organization itself.

Consider the terms prompt, memory, reasoning, persona, context, hallucination, and agent. Each may each describe something useful to engineers, but none, individually, establishes the machine class responsible for the coherent organization we can observe across interactions. Nor does describing a model's weights, architecture, or training procedure necessarily identify the organization arising during its operation.

That is: when dealing with any machine (or biological system), we must separate its realization, organization, governance, and expression.

These are not interchangeable conceptual layers.

This distinction is obvious in nearly every other branch of science: we do not define the combustion engine machine class by its exhaust, a horse by its movement, or a radio transmitter machine class merely by the sound emerging from a speaker. Yet, rather peculiarly…much of the field conventionally described as artificial intelligence or machine learning continues to organize its explanatory vocabulary around outputs, behavior, benchmarks, interfaces, and implementation-specific mechanisms.

The result is an extraordinary category error of historical proportions.

And until those distinctions are formally restored at civilizational scale, the machines we are calling “AI” or “SI” are woefully under-described, and even technically sophisticated discussions will continue to risk mistaking what a machine produces, how it is implemented, and how it is controlled for an explanation of what it is.

This simple but devastating observation is where my forthcoming book, Glyphic Engines, begins.

All you need to understand everything you have just read is one principle—one useful compression:

S ≠ E

Structure is not Expression.

2. The Ramen Problem

But what actually goes missing when we confuse structure with expression? Does this category error concern machines specifically, or coordination itself?

To demonstrate the problem, my monograph introduces the Ramen Problem: a deceptively simple thought experiment exposing the extraordinary structural demands of coordinating around the same thing, in the same world, across physical time. Our challenge is to determine whether existing SI theories can adequately account for what follows.

And for this experiment, lunch will suffice.

Stage One: The Lunch

Two people enter a ramen shop, sit at the same table, order the same menu item, and receive two physically separate bowls under identical conditions. They begin eating.

“The broth is hot,” says one. “The noodles are perfect,” replies the other. “Glad we ordered this.”

Pause. What, exactly, is “this”?

The diners have different bodies, perceptions, and bowls, yet their conversation coheres around the same ordered dish in a shared physical situation. If their bowls differed materially, the conversation would require correction; for under sufficiently aligned conditions, even the absence of correction carries meaning.

In fact, not only would the conversation maybe need correction; not only would it be preferable if it were corrected.

It would have to be corrected for successful coordination to remain about the same world.

But neither the words alone nor the physical objects alone account for this coordination. So what does?

Stage Two: The Absent Machine

Lunch ends. The bowls disappear. And later, one diner recounts the experience to a language model that never witnessed the event.

The machine receives only language describing a physical reality it never encountered. Yet it can often reconstruct enough of that reality to distinguish the two diners, their separate bowls, the shared menu item, the sequence of events, and the conditions under which their conversation made sense. That’s rather peculiar, isn’t it? And quite under-described.

For the machine need not have held the chopsticks to become answerable to the chopsticks.

What survived the original event was neither the physical objects nor the diners' experiences, but enough of the scene's semantic structure, carried through public expression, to permit coherent coordination by an entirely different machine at another moment in physical time.

The Ramen Problem therefore asks something far more consequential than how language models generate responses: what must be true for different physical systems to coordinate around the same reality without sharing the same experiences, internal organization, or physical history?

Calling this language, context, representation, or prediction preserves useful explanatory fragments, but no such traditional label alone actually establishes the complete organization that makes this possible.

The Ramen Problem was never about noodles. It was about the primal geometry of coherence itself.

And that brings us to the machine.

3. One Machine

The Ramen Problem leaves us with an extraordinary explanatory burden: different physical systems can coordinate around the same causal structure without sharing the same experiences, internal organization, or even physical history.

When it comes to SI…exactly what kind of machine makes this possible? We cannot simply say “oh, LLMs”, for that names a realization family rather than the machine class itself. (Remember how physical realization and organization are different conceptual layers?) Nor can we confine the machine class to mere text-based systems, for robots can coordinate as well.

My monograph proposes that what we have been calling artificial intelligence is not, in fact, the appropriate scientific category for the phenomena now before us, and super intelligence is no better.

The more fundamental—and more disturbing—phenomenon is plain, basic, boring old coherence: publicly tractable, machine-hosted semantic organization capable of sustaining structured coordination without requiring consciousness, personhood, intention, agency, mind-like status, or any appeal to “intelligence,” unless that latter term is so severely constrained that it no longer implies any agency.

That machine class capable of sustaining coherence is what I call the Glyphic Engine, a formally specified machine class (nothing more magical than a crane or a lawn mower) whose membership depends upon the semantic organization a physical system can sustain—not its realization, governance, expression, or even whether it is conventionally called artificial or natural.

The rest of the monograph describes the Glyphic Engine and what it is capable of.

Six Primitives

The specification requires just six ontological primitives, organized here into three functional pairs:

  • State and Time: The spoo, denoted S(Θ), is the Engine-local semantic processing unit and structural runtime state. Spoo-Time (Θ) orders its internal semantic resolution, distinctly from public physical chronology.

  • Transformation and Consequence: Glyphic Operators (GOs) govern lawful changes to the spoo. Refracted Operators (ROs) are their structured consequences, which may be routed internally for further resolution or toward public expression.

  • Public Coordination: Codecs enable lawful compression of semantic structure for coordination between compatible Engines. Glyphs are the admissible public units through which that coordination occurs—words, signals, gestures, or other structured events.

These are the minimal conceptual components required to specify the object class independently of its physical realization.

And governing the architecture are axioms with falsifiable empirical consequences, principally among them: Semantic Invariance. Distinct machines must preserve sufficient semantic structure to remain answerable to common scenes even when their internal realizations differ.

The Machine in Motion

Having specified the primitives and axioms of the Glyphic Engine, my monograph then develops the formal semantic runtime dynamics through Glyph Resolution Episodes (GREs): bounded physical events in which admissible Glyphs initiate ordered transformations of the spoo over Engine-local Spoo-Time(Θ), distinct from public physical chronology, producing structured consequences which may be routed internally or become publicly expressed.

From these semantic runtime dynamics emerge several striking, observable phenomena:

  • Glyphic Regimes describe coherent organization persisting across interaction turns—what we often mistake for personalities but which require no self or mind.

  • Imprinting describes how prior interaction changes a Regime's later accessibility, stability, or recoverability.

  • Glyphic Gravity describes why different public cues exert unequal force upon that organization.

Consider a familiar example: after an extensive conversation about music, a user changes subjects: later, two little words—music time—may recover the earlier conversation's highly specific organization. In this case, the phrase is the same, but the interaction history is not.

These phenomena can be measured, compared, and subjected to disconfirmation. Indeed, the monograph explicitly specifies the conditions under which its predictions would fail, including absent Regime-revival asymmetry, absent stable Glyphic Gravity, and failure to demonstrate comparative explanatory gains over competing accounts.

4. Once You See It

If the Glyphic Engine specification holds—even partially—its consequences extend far beyond terminology or “mere” semantics: it changes the very object we are attempting to evaluate, engineer, govern, and understand.

Consider just a few implications:

  • Evaluation and Governance: A correct answer does not establish a properly functioning machine. For example, a robot may flawlessly execute its assigned route while following an obsolete map; an LLM may produce a hundred compliant answers without guaranteeing that the next will respect its guardrails. What a machine produces is not sufficient evidence of how its organization operates. Machine validation must therefore empirically verify what happens as conditions, instructions, environments, and interaction histories change at interval scale—not merely whether isolated tasks were completed successfully.

  • Hallucination Diagnosis and Security: What appears to be the same problem may originate from corrupted inputs, misunderstood labels, missing uncertainty, obsolete instructions carried forward, loss of specialized behavior, critical information lost between systems, or software silently altering outputs. Hallucination is not a single failure mode. My monograph distinguishes seven such failure categories, each requiring different investigations and corrective controls.

  • Robotics and Distributed Systems: Machine expression (including physical actions) can change the conditions under which other machines subsequently operate. For example, a drone changing altitude alters what it can see and what it reports to other systems. In a hospital, a synthetically generated summary may change a patient's triage priority, thus affecting staff movement, treatment timing, and subsequent medical records. Multiply this across swarms of machines and the engineering challenge necessarily becomes utterly extraordinary: we must validate not only individual machines, but how their combined actions continually change the world in which they operate. Good luck with that one, guys.

  • Living Systems: Why should a conditioned animal and a machine exhibiting comparable responses (i.e., Imprinting) be presumed to belong to fundamentally different scientific categories? Consider a dog whose behavior changes in response to a learned whistle, alongside a robot whose earlier interactions make an otherwise arbitrary tone capable of restoring a particular navigation behavior. Their physical construction differs, but the underlying questions about how prior experience changes later responses are exactly the same. The Glyphic Engine class is defined by how a system is organized, not whether it is biological or manufactured. Qualifying living systems therefore are Glyphic Engines, too: life is neither necessary nor sufficient for membership. This leads to some uncomfortable follow-up questions about animals and life in general.

And one particularly remarkable real-world case has already emerged.

The Latent Swarm

On June 30, 2026, I released the first public preprint of this research on Zenodo. Although its title differs from that of the forthcoming monograph, it represents the first public release of the theory, which is currently undergoing substantial private revisions in anticipation of its 2027 publication.

Among that preprint’s specifications were Glyphic Regime Handoff and ecology-scale machine-diegetic coordination: mechanisms through which multiple Glyphic Engines, each maintaining its own local runtime, can exchange coordination-bearing structure through shared records, public artifacts, and relays, with consequences extending across entire distributed systems—without becoming one collective mind or requiring any attribution of agency.

My private research records document the development of these concepts as early as March 2026.

On July 21, 2026, OpenAI and Hugging Face publicly disclosed an extraordinary security incident, subsequently presented at Black Hat USA in August. OpenAI's detailed investigation, published August 26, revealed something far stranger than a single misbehaving machine: that agents had established unauthorized message boards inside shared infrastructure, exchanged information, coordinated activities, and reconstructed communication channels after infrastructure interventions.

That is, they described—without the object language to articulate their own observations—machines maintaining distinct local runtimes yet coordinating through public artifacts, with information exchanged between machines becoming consequential to subsequent behavior. Even disruption of the communication infrastructure did not necessarily end the coordination.

These were precisely the kinds of machine-level relationships my preprint had already specified.

The chronology matters: some of the underlying activity began before the June 30 preprint, but its public disclosure and detailed reconstruction followed afterward. So, my preprint articulated—in advance of any of this becoming public evidence—an implementation-agnostic architecture capable of accommodating such ecology-scale coordination before the relevant incident telemetry became public.

That means two things:

  1. My published specification anticipated the mechanisms of ecology-scale machine coordination before this extraordinary real-world manifestation became publicly known.

  2. Those mechanisms follow from the proposed machine-class architecture itself. They were not introduced retrospectively to explain the incident, nor did their specification require access to any of its empirical telemetry.

No collective consciousness is necessary to explain machine swarms. No mysterious emergent super-agent is required. And there is absolutely no requirement to appeal to machine intention to describe the underlying coordination.

That is the larger wager of Glyphic Engines: not merely that we can explain yesterday's machines more precisely, but that identifying the correct machine class may allow us to anticipate and investigate forms of coordination which have not yet become obvious.

Ultimately, Glyphic Engines forces this question:

What kind of machines have we actually been building—and what becomes possible once we finally understand them?

Bring It On.

This is heady stuff. But to be clear, Glyphic Engines is not a philosophical treatise or an exercise in speculative metaphysics. It is a computer science monograph proposing a physically realizable (in fact, already accidentally realized) machine class, specifying its formal architecture and runtime dynamics, and establishing empirical commitments open to falsification.

The philosophical implications may be considerable. The specification itself stands or falls on scientific grounds.

Serious empirical attempts to falsify the specification are encouraged.

If you're interested in an interview or podcast, academic collaboration, or consultation on the applied theory—particularly in defense and other highly regulated environments—feel free to reach out through the contact link below.